At the most general level there is Traversal<S,E>
which implements Iterator<E>
, where the S
stands for start and
the E
stands for end. A traversal is composed of four primary components:
-
Step<S,E>
: an individual function applied toS
to yieldE
. Steps are chained within a traversal. -
TraversalStrategy
: interceptor methods to alter the execution of the traversal (e.g. query re-writing). -
TraversalSideEffects
: key/value pairs that can be used to store global information about the traversal. -
Traverser<T>
: the object propagating through theTraversal
currently representing an object of typeT
.
The classic notion of a graph traversal is provided by GraphTraversal<S,E>
which extends Traversal<S,E>
.
GraphTraversal
provides an interpretation of the graph data in terms of vertices, edges, etc. and thus, a graph
traversal DSL.
A GraphTraversal<S,E>
is spawned from a GraphTraversalSource
. It can also be spawned anonymously (i.e. empty)
via __
. A graph traversal is composed of an ordered list of steps. All the steps provided by GraphTraversal
inherit from the more general forms diagrammed above. A list of all the steps (and their descriptions) are provided
in the TinkerPop GraphTraversal JavaDoc.
Important
|
The basics for starting a traversal are described in The Graph Process section as well as in the Getting Started tutorial. |
Note
|
To reduce the verbosity of the expression, it is good to
import static org.apache.tinkerpop.gremlin.process.traversal.dsl.graph.__.* . This way, instead of doing __.inE()
for an anonymous traversal, it is possible to simply write inE() . Be aware of language-specific reserved keywords
when using anonymous traversals. For example, in and as are reserved keywords in Groovy, therefore you must use
the verbose syntax __.in() and __.as() to avoid collisions.
|
Important
|
The underlying Step implementations provided by TinkerPop should encompass most of the functionality
required by a DSL author. It is important that DSL authors leverage the provided steps as then the common optimization
and decoration strategies can reason on the underlying traversal sequence. If new steps are introduced, then common
traversal strategies may not function properly.
|
Important
|
4.0.0-beta.1 Release - Transactions are currently disabled in this beta release. |
A database transaction
represents a unit of work to execute against the database. A traversals unit of work is affected by usage convention
(i.e. the method of connecting) and the graph provider’s transaction model. Without diving
deeply into different conventions and models the most general and recommended approach to working with transactions is
demonstrated as follows:
GraphTraversalSource g = traversal().with(graph);
// or
GraphTraversalSource g = traversal().with(conn);
Transaction tx = g.tx();
// spawn a GraphTraversalSource from the Transaction. Traversals spawned
// from gtx will be essentially be bound to tx
GraphTraversalSource gtx = tx.begin();
try {
gtx.addV('person').iterate();
gtx.addV('software').iterate();
tx.commit();
} catch (Exception ex) {
tx.rollback();
}
The above example is straightforward and represents a good starting point for discussing the nuances of transactions in relation to the usage convention and graph provider caveats alluded to earlier.
Focusing on remote contexts first, note that it is still possible to issue traversals from g
, but those will have a
transaction scope outside of gtx
and will simply commit()
on the server if successfully executed or rollback()
on the server otherwise (i.e. one traversal is one transaction). Each isolated transaction will require its own
Transaction
object. Multiple begin()
calls on the same Transaction
object will produce GraphTraversalSource
instances that are bound to the same transaction, therefore:
GraphTraversalSource g = traversal().with(conn);
Transaction tx1 = g.tx();
Transaction tx2 = g.tx();
// both gtx1a and gtx1b will be bound to the same transaction
GraphTraversalSource gtx1a = tx1.begin();
GraphTraversalSource gtx1b = tx1.begin();
// g and gtx2 will not have knowledge of what happens in tx1
GraphTraversalSource gtx2 = tx2.begin();
In remote cases, GraphTraversalSource
instances spawned from begin()
are safe to use in multiple threads though
on the server side they will be processed serially as they arrive. The default behavior of close()
on a
Transaction
for remote cases is to commit()
, so the following re-write of the earlier example is also valid:
// note here that we dispense with creating a Transaction object and
// simply spawn the gtx in a more inline fashion
GraphTraversalSource gtx = g.tx().begin();
try {
gtx.addV('person').iterate();
gtx.addV('software').iterate();
gtx.close();
} catch (Exception ex) {
tx.rollback();
}
Important
|
Transactions with non-JVM languages are always "remote". For specific transaction syntax in a particular language, please see the "Transactions" sub-section of your language of interest in the Gremlin Drivers and Variants section. |
In embedded cases, that initial recommended model for defining transactions holds, but users have more options here
on deeper inspection. For embedded use cases (and perhaps even in configuration of a graph instance in Gremlin Server),
the type of Transaction
object that is returned from g.tx()
is an important indicator as to the features of that
graph’s transaction model. In most cases, inspection of that object will indicate an instance that derives from the
AbstractThreadLocalTransaction
class, which means that the transaction is bound to the current thread and therefore
all traversals that execute within that thread are tied to that transaction.
A ThreadLocal
transaction differs then from the remote case described before because technically any traversal
spawned from g
or from a Transaction
will fall under the same transaction scope. As a result, it is wise, when
trying to write context agnostic Gremlin, to follow the more rigid conventions of the initial example.
The sub-sections that follow offer a bit more insight into each of the usage contexts.
When on the JVM using an embedded graph, there is considerable flexibility for working with
transactions. With the Graph API, transactions are controlled by an implementation of the Transaction
interface and
that object can be obtained from the Graph
interface using the tx()
method. It is important to note that the
Transaction
object does not represent a "transaction" itself. It merely exposes the methods for working with
transactions (e.g. committing, rolling back, etc).
Most Graph
implementations that supportsTransactions
will implement an "automatic" ThreadLocal
transaction,
which means that when a read or write occurs after the Graph
is instantiated, a transaction is automatically
started within that thread. There is no need to manually call a method to "create" or "start" a transaction. Simply
modify the graph as required and call graph.tx().commit()
to apply changes or graph.tx().rollback()
to undo them.
When the next read or write action occurs against the graph, a new transaction will be started within that current
thread of execution.
When using transactions in this fashion, especially in web application (e.g. HTTP server), it is important to ensure that transactions do not leak from one request to the next. In other words, unless a client is somehow bound via session to process every request on the same server thread, every request must be committed or rolled back at the end of the request. By ensuring that the request encapsulates a transaction, it ensures that a future request processed on a server thread is starting in a fresh transactional state and will not have access to the remains of one from an earlier request. A good strategy is to rollback a transaction at the start of a request, so that if it so happens that a transactional leak does occur between requests somehow, a fresh transaction is assured by the fresh request.
Tip
|
The tx() method is on the Graph interface, but it is also available on the TraversalSource spawned from a
Graph . Calls to TraversalSource.tx() are proxied through to the underlying Graph as a convenience.
|
Tip
|
Some graphs may throw an exception that implements TemporaryException . In this case, this marker interface is
designed to inform the client that it may choose to retry the operation at a later time for possible success.
|
Warning
|
TinkerPop provides for basic transaction control, however, like many aspects of TinkerPop, it is up to the graph system provider to choose the specific aspects of how their implementation will work and how it fits into the TinkerPop stack. Be sure to understand the transaction semantics of the specific graph implementation that is being utilized as it may present differing functionality than described here. |
Determining when a transaction starts is dependent upon the behavior assigned to the Transaction
. It is up to the
Graph
implementation to determine the default behavior and unless the implementation doesn’t allow it, the behavior
itself can be altered via these Transaction
methods:
public Transaction onReadWrite(Consumer<Transaction> consumer);
public Transaction onClose(Consumer<Transaction> consumer);
Providing a Consumer
function to onReadWrite
allows definition of how a transaction starts when a read or a write
occurs. Transaction.READ_WRITE_BEHAVIOR
contains pre-defined Consumer
functions to supply to the onReadWrite
method. It has two options:
-
AUTO
- automatic transactions where the transaction is started implicitly to the read or write operation -
MANUAL
- manual transactions where it is up to the user to explicitly open a transaction, throwing an exception if the transaction is not open
Providing a Consumer
function to onClose
allows configuration of how a transaction is handled when
Transaction.close()
is called. Transaction.CLOSE_BEHAVIOR
has several pre-defined options that can be supplied to
this method:
-
COMMIT
- automatically commit an open transaction -
ROLLBACK
- automatically rollback an open transaction -
MANUAL
- throw an exception if a transaction is open, forcing the user to explicitly close the transaction
Important
|
As transactions are ThreadLocal in nature, so are the transaction configurations for onReadWrite and
onClose .
|
Once there is an understanding for how transactions are configured, most of the rest of the Transaction
interface
is self-explanatory. Note that Neo4j-Gremlin is used for the examples to follow as TinkerGraph does
not support transactions.
Important
|
The following example is meant to demonstrate specific use of ThreadLocal transactions and is at odds
with the more generalized transaction convention that is recommended for both embedded and remote contexts. Please be
sure to understand the preferred approach described at in the Traversal Transactions Section before
using this method.
|
gremlin> graph = Neo4jGraph.open('/tmp/neo4j')
==>neo4jgraph[EmbeddedGraphDatabase [/tmp/neo4j]]
gremlin> g = traversal().with(graph)
==>graphtraversalsource[neo4jgraph[community single [/tmp/neo4j]], standard]
gremlin> graph.features()
==>FEATURES
> GraphFeatures
>-- Transactions: true (1)
>-- Computer: false
>-- Persistence: true
...
gremlin> g.tx().onReadWrite(Transaction.READ_WRITE_BEHAVIOR.AUTO) (2)
==>org.apache.tinkerpop.gremlin.neo4j.structure.Neo4jGraph$Neo4jTransaction@1c067c0d
gremlin> g.addV("person").("name","stephen") (3)
==>v[0]
gremlin> g.tx().commit() (4)
==>null
gremlin> g.tx().onReadWrite(Transaction.READ_WRITE_BEHAVIOR.MANUAL) (5)
==>org.apache.tinkerpop.gremlin.neo4j.structure.Neo4jGraph$Neo4jTransaction@1c067c0d
gremlin> g.tx().isOpen()
==>false
gremlin> g.addV("person").("name","marko") (6)
Open a transaction before attempting to read/write the transaction
gremlin> g.tx().open() (7)
==>null
gremlin> g.addV("person").("name","marko") (8)
==>v[1]
gremlin> g.tx().commit()
==>null
-
Check
features
to ensure that the graph supports transactions. -
By default,
Neo4jGraph
is configured with "automatic" transactions, so it is set here for demonstration purposes only. -
When the vertex is added, the transaction is automatically started. From this point, more mutations can be staged or other read operations executed in the context of that open transaction.
-
Calling
commit
finalizes the transaction. -
Change transaction behavior to require manual control.
-
Adding a vertex now results in failure because the transaction was not explicitly opened.
-
Explicitly open a transaction.
-
Adding a vertex now succeeds as the transaction was manually opened.
Note
|
It may be important to consult the documentation of the Graph implementation you are using when it comes to the
specifics of how transactions will behave. TinkerPop allows some latitude in this area and implementations may not have
the exact same behaviors and ACID guarantees.
|
The available capability for transactions with Gremlin Server is dependent upon the method of interaction that is used. The preferred method for interacting with Gremlin Server is via websockets and bytecode based requests. The start of the Transactions Section describes this approach in detail with examples.
Gremlin Server also has the option to accept Gremlin-based scripts. The scripting approach provides access to the Graph API and thus also the transactional model described in the embedded section. Therefore a single script can have the ability to execute multiple transactions per request with complete control provided to the developer to commit or rollback transactions as needed.
Important
|
4.0.0-beta.1 Release - Transactions are currently disabled in this beta release. |
There are two methods for sending scripts to Gremlin Server: sessionless and session-based. With sessionless requests there will always be an attempt to close the transaction at the end of the request with a commit if there are no errors or a rollback if there is a failure. It is therefore unnecessary to close transactions manually within scripts themselves. By default, session-based requests do not have this quality. The transaction will be held open on the server until the user closes it manually. There is an option to have automatic transaction management for sessions. More information on this topic can be found in the Considering Transactions Section.
Important
|
4.0.0-beta.1 Release - Sessions have been removed. All scripts are now sessionless. |
At this time, transactional patterns for Remote Gremlin Providers are largely in line with Gremlin Server. As most of
RGPs do not expose a Graph
instance, access to lower level transactional functions available to embedded graphs
even in a sessionless fashion are not typically permitted. For example, without a Graph
instance it is not possible
to configure transaction close or read-write
behaviors. The nature of what a "transaction" means will be dependent on the RGP as is the case with any
TinkerPop-enabled graph system, so it is important to consult that systems documentation for more details.
Many of the methods on the GraphTraversalSource
are meant to configure the source for usage. These configuration
affect the manner in which a traversals are spawned from it. Configuration methods can be identified by their names
with make use of "with" as a prefix:
The with()
configuration adds arbitrary data to a TraversalSource
which can then be used by graph providers as
configuration options for a traversal execution. This configuration is similar to with()-modulator which
has similar functionality when applied to an individual step.
g.with('providerDefinedVariable', 0.33).V()
The 0.33
value for the "providerDefinedVariable" will be bound to each traversal spawned that way. Consult the
graph system being used to determine if any such configuration options are available.
The withBulk()
configuration allows for control of bulking operations. This value is true
by default allowing for
normal bulking operations, but when set to false
, introduces a subtle change in that behavior as
shown in examples in sack()-step.
The withComputer()
configuration adds a Computer
that will be used to process the traversal and is necessary for
OLAP based processing and steps that require that processing. See examples related to
SparkGraphComputer
or see examples in the computer required steps, like pageRank() or
[shortestpath-shortestPath()].
The withSack()
configuration adds a "sack" that can be accessed by traversals spawned from this source. This
functionality is shown in more detail in the examples for (sack())-step.
The withSideEffect()
configuration adds an arbitrary Object
to traversals spawned from this source which can be
accessed as a side-effect given the supplied key.
g.withSideEffect('x',['dog','cat','fish']). V().has('person','name','marko').select('x').unfold()
The withStrategies()
configuration allows inclusion of additional TraversalStrategy
instances to be applied to
any traversals spawned from the configured source. Please see the Traversal Strategy Section
for more details on how this configuration works.
The withoutStrategies()
configuration removes a particular TraversalStrategy
from those to be applied to traversals
spawned from the configured source. Please see the Traversal Strategy Section for more details
on how this configuration works.
Not all steps are capable of starting a GraphTraversal
. Only those steps on the GraphTraversalSource
can do that.
Many of the methods on GraphTraversalSource
are actually for its configuration and start
steps should not be confused with those.
Spawn steps, which actually yield a traversal, typically match the names of existing steps:
-
addE()
- Adds anEdge
to start the traversal (example). -
addV()
- Adds aVertex
to start the traversal (example). -
call()
- Makes a provider-specific service call to start the traversal (example). -
E()
- Reads edges from the graph to start the traversal (example). -
inject()
- Inserts arbitrary objects to start the traversal (example). -
mergeE()
- Adds anEdge
in a "create if not exist" fashion to start the traversal (example) -
mergeV()
- Adds aVertex
in a "create if not exist" fashion to start the traversal (example) -
union()
- Merges the results of an arbitrary number of child traversals to start the traversal (example). -
V()
- Reads vertices from the graph to start the traversal (example).
Gremlin steps are chained together to produce the actual traversal and are triggered by way of start steps
on the GraphTraversalSource
.
Important
|
More details about the Gremlin language can be found in the Provider Documentation within the Gremlin Semantics Section. |
There are five general steps, each having a traversal and a lambda representation, by which all other specific steps described later extend.
Step | Description |
---|---|
|
map the traverser to some object of type |
|
map the traverser to an iterator of |
|
map the traverser to either true or false, where false will not pass the traverser to the next step. |
|
perform some operation on the traverser and pass it to the next step. |
|
split the traverser to all the traversals indexed by the |
Warning
|
Lambda steps are presented for educational purposes as they represent the foundational constructs of the Gremlin language. In practice, lambda steps should be avoided in favor of their traversals representation and traversal verification strategies exist to disallow their use unless explicitly "turned off." For more information on the problems with lambdas, please read A Note on Lambdas. |
The Traverser<S>
object provides access to:
-
The current traversed
S
object —Traverser.get()
. -
The current path traversed by the traverser —
Traverser.path()
.-
A helper shorthand to get a particular path-history object —
Traverser.path(String) == Traverser.path().get(String)
.
-
-
The number of times the traverser has gone through the current loop —
Traverser.loops()
. -
The number of objects represented by this traverser —
Traverser.bulk()
. -
The local data structure associated with this traverser —
Traverser.sack()
. -
The side-effects associated with the traversal —
Traverser.sideEffects()
.-
A helper shorthand to get a particular side-effect —
Traverser.sideEffect(String) == Traverser.sideEffects().get(String)
.
-
g.V(1).out().values('name') (1) g.V(1).out().map {it.get().value('name')} (2) g.V(1).out().map(values('name')) (3)
-
An outgoing traversal from vertex 1 to the name values of the adjacent vertices.
-
The same operation, but using a lambda to access the name property values.
-
Again the same operation, but using the traversal representation of
map()
.
g.V().filter {it.get().label() == 'person'} (1) g.V().filter(label().is('person')) (2) g.V().hasLabel('person') (3)
-
A filter that only allows the vertex to pass if it has the "person" label
-
The same operation, but using the traversal representation of
filter()
. -
The more specific
has()
-step is implemented as afilter()
with respective predicate.
g.V().hasLabel('person').sideEffect(System.out.&println) (1) g.V().sideEffect(outE().count().aggregate(local,"o")). sideEffect(inE().count().aggregate(local,"i")).cap("o","i") (2)
-
Whatever enters
sideEffect()
is passed to the next step, but some intervening process can occur. -
Compute the out- and in-degree for each vertex. Both
sideEffect()
are fed with the same vertex.
g.V().branch {it.get().value('name')}. option('marko', values('age')). option(none, values('name')) (1) g.V().branch(values('name')). option('marko', values('age')). option(none, values('name')) (2) g.V().choose(has('name','marko'), values('age'), values('name')) (3)
-
If the vertex is "marko", get his age, else get the name of the vertex.
-
The same operation, but using the traversal representing of
branch()
. -
The more specific boolean-based
choose()
-step is implemented as abranch()
.
Typically, when a step is concatenated to a traversal a traversal is returned. In this way, a traversal is built up in a fluent, monadic fashion. However, some steps do not return a traversal, but instead, execute the traversal and return a result. These steps are known as terminal steps (terminal) and they are explained via the examples below.
g.V().out('created').hasNext() (1) g.V().out('created').next() (2) g.V().out('created').next(2) (3) g.V().out('nothing').tryNext() (4) g.V().out('created').toList() (5) g.V().out('created').toSet() (6) g.V().out('created').toBulkSet() (7) results = ['blah',3] g.V().out('created').fill(results) (8) g.addV('person').iterate() (9)
-
hasNext()
determines whether there are available results (not supported ingremlin-javascript
). -
next()
will return the next result. -
next(n)
will return the nextn
results in a list (not supported ingremlin-javascript
or Gremlin.NET). -
tryNext()
will return anOptional
and thus, is a composite ofhasNext()
/next()
(only supported for JVM languages). -
toList()
will return all results in a list. -
toSet()
will return all results in a set and thus, duplicates removed (not supported ingremlin-javascript
). -
toBulkSet()
will return all results in a weighted set and thus, duplicates preserved via weighting (only supported for JVM languages). -
fill(collection)
will put all results in the provided collection and return the collection when complete (only supported for JVM languages). -
iterate()
does not exactly fit the definition of a terminal step in that it doesn’t return a result, but still returns a traversal - it does however behave as a terminal step in that it iterates the traversal and generates side effects without returning the actual result.
There is also the promise()
terminator step, which can only be used with remote traversals to
Gremlin Server or RGPs. It starts a promise to execute a function
on the current Traversal
that will be completed in the future.
Finally, explain()
-step is also a terminal step and is described in its own section.
Reasoning is the process of making explicit what is implicit
in the data. What is explicit in a graph are the objects of the graph — i.e. vertices and edges. What is implicit
in the graph is the traversal. In other words, traversals expose meaning where the meaning is determined by the
traversal definition. For example, take the concept of a "co-developer." Two people are co-developers if they have
worked on the same project together. This concept can be represented as a traversal and thus, the concept of
"co-developers" can be derived. Moreover, what was once implicit can be made explicit via the addE()
-step
(map/sideEffect).
g.V(1).as('a').out('created').in('created').where(neq('a')). addE('co-developer').from('a').property('year',2009) (1) g.V(3,4,5).aggregate('x').has('name','josh').as('a'). select('x').unfold().hasLabel('software').addE('createdBy').to('a') (2) g.V().as('a').out('created').addE('createdBy').to('a').property('acl','public') (3) g.V(1).as('a').out('knows'). addE('livesNear').from('a').property('year',2009). inV().inE('livesNear').values('year') (4) g.V().match( __.as('a').out('knows').as('b'), __.as('a').out('created').as('c'), __.as('b').out('created').as('c')). addE('friendlyCollaborator').from('a').to('b'). property(id,23).property('project',select('c').values('name')) (5) g.E(23).valueMap() vMarko = g.V().has('name','marko').next() vPeter = g.V().has('name','peter').next() g.V(vMarko).addE('knows').to(vPeter) (6) g.addE('knows').from(vMarko).to(vPeter) (7)
-
Add a co-developer edge with a year-property between marko and his collaborators.
-
Add incoming createdBy edges from the josh-vertex to the lop- and ripple-vertices.
-
Add an inverse createdBy edge for all created edges.
-
The newly created edge is a traversable object.
-
Two arbitrary bindings in a traversal can be joined
from()
→to()
, whereid
can be provided for graphs that supports user provided ids. -
Add an edge between marko and peter given the directed (detached) vertex references.
-
Add an edge between marko and peter given the directed (detached) vertex references.
Additional References
The addV()
-step is used to add vertices to the graph (map/sideEffect). For every incoming object, a vertex is
created. Moreover, GraphTraversalSource
maintains an addV()
method.
g.addV('person').property('name','stephen') g.V().values('name') g.V().outE('knows').addV().property('name','nothing') g.V().has('name','nothing') g.V().has('name','nothing').bothE()
Additional References
The aggregate()
-step (sideEffect) is used to aggregate all the objects at a particular point of traversal into a
Collection
. The step is uses Scope
to help determine the aggregating behavior. For global
scope this means that
the step will use eager evaluation in that no objects continue on
until all previous objects have been fully aggregated. The eager evaluation model is crucial in situations
where everything at a particular point is required for future computation. By default, when the overload of
aggregate()
is called without a Scope
, the default is global
. An example is provided below.
g.V(1).out('created') (1) g.V(1).out('created').aggregate('x') (2) g.V(1).out('created').aggregate(global, 'x') (3) g.V(1).out('created').aggregate('x').in('created') (4) g.V(1).out('created').aggregate('x').in('created').out('created') (5) g.V(1).out('created').aggregate('x').in('created').out('created'). where(without('x')).values('name') (6)
-
What has marko created?
-
Aggregate all his creations.
-
Identical to the previous line.
-
Who are marko’s collaborators?
-
What have marko’s collaborators created?
-
What have marko’s collaborators created that he hasn’t created?
In recommendation systems, the above pattern is used:
"What has userA liked? Who else has liked those things? What have they liked that userA hasn't already liked?"
Finally, aggregate()
-step can be modulated via by()
-projection.
g.V().out('knows').aggregate('x').cap('x') g.V().out('knows').aggregate('x').by('name').cap('x') g.V().out('knows').aggregate('x').by('age').cap('x') (1)
-
The "age" property is not productive for all vertices and therefore those values are not included in the aggregation.
For local
scope the aggregation will occur in a lazy fashion.
Note
|
Prior to 3.4.3, local aggregation (i.e. lazy) evaluation was handled by store() -step.
|
g.V().aggregate(global, 'x').limit(1).cap('x') g.V().aggregate(local, 'x').limit(1).cap('x') g.withoutStrategies(EarlyLimitStrategy).V().aggregate(local,'x').limit(1).cap('x')
It is important to note that EarlyLimitStrategy
introduced in 3.3.5 alters the behavior of aggregate(local)
.
Without that strategy (which is installed by default), there are two results in the aggregate()
side-effect even
though the interval selection is for 1 object. Realize that when the second object is on its way to the range()
filter (i.e. [0..1]
), it passes through aggregate()
and thus, stored before filtered.
g.E().aggregate(local,'x').by('weight').cap('x')
Additional References
It is possible to filter list traversers using all()
-step (filter). Every item in the list will be tested against
the supplied predicate and if all of the items pass then the traverser is passed along the stream, otherwise it is
filtered. Empty lists are passed along but null or non-iterable traversers are filtered out.
Python
|
The term |
g.V().values('age').fold().all(gt(25)) (1)
-
Return the list of ages only if everyone’s age is greater than 25.
Additional References
The and()
-step ensures that all provided traversals yield a result (filter). Please see or()
for or-semantics.
Python
|
The term |
g.V().and( outE('knows'), values('age').is(lt(30))). values('name')
The and()
-step can take an arbitrary number of traversals. All traversals must produce at least one output for the
original traverser to pass to the next step.
An infix notation can be used as well.
g.V().where(outE('created').and().outE('knows')).values('name')
Additional References
It is possible to filter list traversers using any()
-step (filter). All items in the list will be tested against
the supplied predicate and if any of the items pass then the traverser is passed along the stream, otherwise it is
filtered. Empty lists, null traversers, and non-iterable traversers are filtered out as well.
Python
|
The term |
g.V().values('age').fold().any(gt(25)) (1)
-
Return the list of ages if anyone’s age is greater than 25.
Additional References
The as()
-step is not a real step, but a "step modulator" similar to by()
and option()
.
With as()
, it is possible to provide a label to the step that can later be accessed by steps and data structures
that make use of such labels — e.g., select()
, match()
, and path.
Groovy
|
The term |
Python
|
The term |
g.V().as('a').out('created').as('b').select('a','b') (1) g.V().as('a').out('created').as('b').select('a','b').by('name') (2)
-
Select the objects labeled "a" and "b" from the path.
-
Select the objects labeled "a" and "b" from the path and, for each object, project its name value.
A step can have any number of labels associated with it. This is useful for referencing the same step multiple times in a future step.
g.V().hasLabel('software').as('a','b','c'). select('a','b','c'). by('name'). by('lang'). by(__.in('created').values('name').fold())
Additional References
The asString()
-step (map) returns the value of incoming traverser as strings. Null values are returned unchanged.
g.V().hasLabel('person').values('age').asString() (1) g.V().hasLabel('person').values('age').asString().concat(' years old') (2) g.V().hasLabel('person').values('age').fold().asString(local) (3)
-
Return ages as string.
-
Return ages as string and use concat to generate phrases.
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
Additional References
The asDate()
-step (map) converts string or numeric input to Date.
For string input only ISO-8601 format is supported. For numbers, the value is considered as the number of the milliseconds since "the epoch" (January 1, 1970, 00:00:00 GMT). Date input is passed without changes.
If the incoming traverser is not a string, number, Date or OffsetDateTime then an IllegalArgumentException
will be thrown.
g.inject(1690934400000).asDate() (1) g.inject("2023-08-02T00:00:00Z").asDate() (2) g.inject(datetime("2023-08-24T00:00:00Z")).asDate() (3)
-
Convert number to Date
-
Convert ISO-8601 string to Date
-
Pass Date without modification
Additional References
The barrier()
-step (barrier) turns the lazy traversal pipeline into a bulk-synchronous pipeline. This step is
useful in the following situations:
-
When everything prior to
barrier()
needs to be executed before moving onto the steps after thebarrier()
(i.e. ordering). -
When "stalling" the traversal may lead to a "bulking optimization" in traversals that repeatedly touch many of the same elements (i.e. optimizing).
g.V().sideEffect{println "first: ${it}"}.sideEffect{println "second: ${it}"}.iterate() g.V().sideEffect{println "first: ${it}"}.barrier().sideEffect{println "second: ${it}"}.iterate()
The theory behind a "bulking optimization" is simple. If there are one million traversers at vertex 1, then there is
no need to calculate one million both()
-computations. Instead, represent those one million traversers as a single
traverser with a Traverser.bulk()
equal to one million and execute both()
once. A bulking optimization example is
made more salient on a larger graph. Therefore, the example below leverages the Grateful Dead graph.
graph = TinkerGraph.open() g = traversal().with(graph) g.io('data/grateful-dead.xml').read().iterate() g = traversal().with(graph).withoutStrategies(LazyBarrierStrategy) (1) clockWithResult(1){g.V().both().both().both().count().next()} (2) clockWithResult(1){g.V().repeat(both()).times(3).count().next()} (3) clockWithResult(1){g.V().both().barrier().both().barrier().both().barrier().count().next()} (4)
-
Explicitly remove
LazyBarrierStrategy
which yields a bulking optimization. -
A non-bulking traversal where each traverser is processed.
-
Each traverser entering
repeat()
has its recursion bulked. -
A bulking traversal where implicit traversers are not processed.
If barrier()
is provided an integer argument, then the barrier will only hold n
-number of unique traversers in its
barrier before draining the aggregated traversers to the next step. This is useful in the aforementioned bulking
optimization scenario with the added benefit of reducing the risk of an out-of-memory exception.
LazyBarrierStrategy
inserts barrier()
-steps into a traversal where appropriate in order to gain the
"bulking optimization."
graph = TinkerGraph.open() g = traversal().with(graph) (1) g.io('data/grateful-dead.xml').read().iterate() clockWithResult(1){g.V().both().both().both().count().next()} g.V().both().both().both().count().iterate().toString() (2)
-
LazyBarrierStrategy
is a default strategy and thus, does not need to be explicitly activated. -
With
LazyBarrierStrategy
activated,barrier()
-steps are automatically inserted where appropriate.
Additional References
The branch()
step splits the traverser to all the child traversals provided to it. Please see the
General Steps section for more information, but also consider that branch()
is the basis for more
robust steps like choose() and union().
Additional References
The by()
-step is not an actual step, but instead is a "step-modulator" similar to as()
and
option()
. If a step is able to accept traversals, functions, comparators, etc. then by()
is the
means by which they are added. The general pattern is step().by()…by()
. Some steps can only accept one by()
while others can take an arbitrary amount.
g.V().group().by(bothE().count()) (1) g.V().group().by(bothE().count()).by('name') (2) g.V().group().by(bothE().count()).by(count()) (3)
-
by(outE().count())
will group the elements by their edge count (traversal). -
by('name')
will process the grouped elements by their name (element property projection). -
by(count())
will count the number of elements in each group (traversal).
When a by()
modulator does not produce a result, it is deemed "unproductive". An "unproductive" modulator will lead
to the filtering of the traverser it is currently working with. The filtering will manifest in various ways depending
on the step.
g.V().sample(1).by('age') (1)
-
The "age" property key is not present for all vertices, therefore
sample()
will ignore (i.e. filter) such vertices for consideration in the sampling.
The following steps all support by()
-modulation. Note that the semantics of such modulation should be understood
on a step-by-step level and thus, as discussed in their respective section of the documentation.
-
aggregate()
: aggregate all objects into a set but only store theirby()
-modulated values. -
cyclicPath()
: filter if the traverser’s path is cyclic givenby()
-modulation. -
dedup()
: dedup on the results of aby()
-modulation. -
format()
: transform a traverser provided to the step by way of theby()
modulator before it is processed by it. -
group()
: create group keys and values according toby()
-modulation. -
groupCount()
: count those groups where the group keys are the result ofby()
-modulation. -
math()
: transform a traverser provided to the step by way of theby()
modulator before it is processed by it. -
order()
: order the objects by the results of aby()
-modulation. -
path()
: get the path of the traverser where each path element isby()
-modulated. -
project()
: project a map of results given variousby()
-modulations off the current object. -
propertyMap()
: transform the result of the values in the resultingMap
using theby()
modulator. -
sack()
: provides the transformation for a traverser to a value to be stored in the sack. -
sample()
: sample using the value returned byby()
-modulation. -
select()
: select path elements and transform them viaby()
-modulation. -
simplePath()
: filter if the traverser’s path is simple givenby()
-modulation. -
tree()
: get a tree of traversers objects where the objects have beenby()
-modulated. -
valueMap()
: transform the result of the values in the resultingMap
using theby()
modulator. -
where()
: determine the predicate given the testing of the results ofby()
-modulation.
Additional References
The call()
step allows for custom, provider-specific service calls either at the start of a traversal or mid-traversal.
This allows Graph providers to expose operations not natively built into the Gremlin language, such as full text search,
custom analytics, notification triggers, etc.
When called with no arguments, call()
will produce a list of callable services available for the graph in use. This
no-argument version is equivalent to call('--list')
. This "directory service" is also capable of producing more
verbose output describing all the services or an individual service:
g.call() (1) g.call('--list') (1) g.call().with('verbose') (2) g.call().with('verbose').with('service', 'xyz-service') (3)
-
List available services by name
-
Produce a Map of detailed service information by name
-
Produce the detailed service information for the 'xyz-service'
The first argument to call()
is always the name of the service call. Additionally, service calls can accept both
static and dynamically produced parameters. Static parameters can be passed as a Map
to the call()
as the second
argument. Individual static parameters can also be added using the .with()
modulator. Dynamic parameters can be
passed as a Map
-producing Traversal
as the second argument (no static parameters) or third argument (static + dynamic
parameters). Additional individual dynamic parameters can be added using the .with()
modulator.
g.call('xyz-service') (1)
g.call('xyz-service', ['a':'b']) (2)
g.call('xyz-service').with('a', 'b') (2)
g.call('xyz-service', __.inject(['a':'b'])) (3)
g.call('xyz-service').with('a', __.inject('b')) (3)
g.call('xyz-service', ['a':'b'], __.inject(['c':'d'])) (4)
-
Call the 'xyz-service' with no parameters
-
Examples of static parameters (constants known before execution)
-
Examples of dynamic parameters (these will be computed at execution time)
-
Example of static + dynamic parameters (these will be computed and merged into one set of parameters at execution time)
Additional References
GraphTraversalSource:
GraphTraversal:
The cap()
-step (barrier) iterates the traversal up to itself and emits the sideEffect referenced by the provided
key. If multiple keys are provided, then a Map<String,Object>
of sideEffects is emitted.
g.V().groupCount('a').by(label).cap('a') (1) g.V().groupCount('a').by(label).groupCount('b').by(outE().count()).cap('a','b') (2)
-
Group and count vertices by their label. Emit the side effect labeled 'a', which is the group count by label.
-
Same as statement 1, but also emit the side effect labeled 'b' which groups vertices by the number of out edges.
Additional References
The choose()
-step (branch) routes the current traverser to a particular traversal branch option. With choose()
,
it is possible to implement if/then/else-semantics as well as more complicated selections.
g.V().hasLabel('person'). choose(values('age').is(lte(30)), __.in(), __.out()).values('name') (1) g.V().hasLabel('person'). choose(values('age')). option(27, __.in()). option(32, __.out()).values('name') (2)
-
If the traversal yields an element, then do
in
, else doout
(i.e. true/false-based option selection). -
Use the result of the traversal as a key to the map of traversal options (i.e. value-based option selection).
If the "false"-branch is not provided, then if/then-semantics are implemented.
g.V().choose(hasLabel('person'), out('created')).values('name') (1) g.V().choose(hasLabel('person'), out('created'), identity()).values('name') (2)
-
If the vertex is a person, emit the vertices they created, else emit the vertex.
-
If/then/else with an
identity()
on the false-branch is equivalent to if/then with no false-branch.
Note that choose()
can have an arbitrary number of options and moreover, can take an anonymous traversal as its choice function.
g.V().hasLabel('person'). choose(values('name')). option('marko', values('age')). option('josh', values('name')). option('vadas', elementMap()). option('peter', label())
The choose()
-step can leverage the Pick.none
option match. For anything that does not match a specified option, the none
-option is taken.
g.V().hasLabel('person'). choose(values('name')). option('marko', values('age')). option(none, values('name'))
Additional References
The coalesce()
-step evaluates the provided traversals in order and returns the first traversal that emits at
least one element.
g.V(1).coalesce(outE('knows'), outE('created')).inV().path().by('name').by(label) g.V(1).coalesce(outE('created'), outE('knows')).inV().path().by('name').by(label) g.V(1).property('nickname', 'okram') g.V().hasLabel('person').coalesce(values('nickname'), values('name'))
Additional References
To randomly filter out a traverser, use the coin()
-step (filter). The provided double argument biases the "coin toss."
g.V().coin(0.5) g.V().coin(0.0) g.V().coin(1.0)
Additional References
The combine()
-step (map) combines the elements of the incoming list traverser and the provided list argument into
one list. This is also known as appending or concatenating. This step only expects list data (array or Iterable) and
will throw an IllegalArgumentException
if any other type is encountered (including null
). This differs from the
merge()
-step in that it allows duplicates to exist.
g.V().values("name").fold().combine(["james","jen","marko","vadas"]) g.V().values("name").fold().combine(__.constant("stephen").fold())
Additional References
The concat()
-step (map) concatenates one or more String values together to the incoming String traverser. This step
can take either String varargs or Traversal varargs.
Any null
String values will be skipped when concatenated with non-null
String values. If two null
value are
concatenated, the null
value will be propagated and returned.
If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.addV(constant('prefix_').concat(__.V(1).label())).property(id, 10) (1) g.V(10).label() g.V().hasLabel('person').values('name').as('a'). constant('Mr.').concat(__.select('a')) (2) g.V().hasLabel('software').as('a').values('name'). concat(' uses '). concat(select('a').values('lang')) (3) g.V(1).outE().as('a').V(1).values('name'). concat(' '). concat(select('a').label()). concat(' '). concat(select("a").inV().values('name')) (4) g.V(1).outE().as('a').V(1).values('name'). concat(constant(' '), select("a").label(), constant(' '), select('a').inV().values('name')) (5) g.inject('hello', 'hi').concat(__.V().values('name')) (6) g.inject('This').concat(' ').concat('is a ', 'gremlin.') (7)
-
Add a new vertex with id 10 which should be labeled like an existing vertex but with some prefix attached
-
Attach the prefix "Mr." to all the names using the
constant()
-step -
Generate a string of software names and the language they use
-
Generate a string description for each of marko’s outgoing edges
-
Alternative way to generate the string description by using traversal varargs. Use the
constant()
step to add desired strings between arguments. -
The
concat()
step will append the first result from the child traversal to the incoming traverser -
A generic use of
concat()
to join strings together
Additional References
The conjoin()
-step (map) joins together the elements in the incoming list traverser together with the provided argument
as a delimiter. The resulting String
is added to the Traversal Stream. This step only expects list data (array or
Iterable) in the incoming traverser and will throw an IllegalArgumentException
if any other type is encountered
(including null
). Null values are skipped and not included in the result.
g.V().values("name").fold().conjoin("+")
Additional References
The connectedComponent()
step performs a computation to identify Connected Component
instances in a graph. When this step completes, the vertices will be labelled with a component identifier to denote
the component to which they are associated.
Important
|
The connectedComponent() -step is a VertexComputing -step and as such, can only be used against a graph
that supports GraphComputer (OLAP).
|
g = traversal().with(graph).withComputer() g.V(). connectedComponent(). with(ConnectedComponent.propertyName, 'component'). project('name','component'). by('name'). by('component') g.V().hasLabel('person'). connectedComponent(). with(ConnectedComponent.propertyName, 'component'). with(ConnectedComponent.edges, outE('knows')). project('name','component'). by('name'). by('component')
Note the use of the with()
modulating step which provides configuration options to the algorithm. It takes
configuration keys from the ConnectedComponent
class and is automatically imported to the Gremlin Console.
Additional References
To specify a constant value for a traverser, use the constant()
-step (map). This is often useful with conditional
steps like choose()
-step or coalesce()
-step.
g.V().choose(hasLabel('person'), values('name'), constant('inhuman')) (1) g.V().coalesce( hasLabel('person').values('name'), constant('inhuman')) (2)
-
Show the names of people, but show "inhuman" for other vertices.
-
Same as statement 1 (unless there is a person vertex with no name).
Additional References
The count()
-step (map) counts the total number of represented traversers in the streams (i.e. the bulk count).
g.V().count() g.V().hasLabel('person').count() g.V().hasLabel('person').outE('created').count().path() (1) g.V().hasLabel('person').outE('created').count().map {it.get() * 10}.path() (2)
-
count()
-step is a reducing barrier step meaning that all of the previous traversers are folded into a new traverser. -
The path of the traverser emanating from
count()
starts atcount()
.
Important
|
count(local) counts the current, local object (not the objects in the traversal stream). This works for
Collection - and Map -type objects. For any other object, a count of 1 is returned.
|
Additional References
Each traverser maintains its history through the traversal over the graph — i.e. its path.
If it is important that the traverser repeat its course, then cyclic()
-path should be used (filter). The step
analyzes the path of the traverser thus far and if there are any repeats, the traverser is filtered out over the
traversal computation. If non-cyclic behavior is desired, see simplePath()
.
g.V(1).both().both() g.V(1).both().both().cyclicPath() g.V(1).both().both().cyclicPath().path() g.V(1).both().both().cyclicPath().by('age').path() (1) g.V(1).as('a').out('created').as('b'). in('created').as('c'). cyclicPath(). path() g.V(1).as('a').out('created').as('b'). in('created').as('c'). cyclicPath().from('a').to('b'). path()
-
The "age" property is not productive for all vertices and therefore those traversers are filtered.
Additional References
The dateAdd()
-step (map) returns the value with the addition of the value number of units as specified by the DateToken.
If the incoming traverser is not a Date or OffsetDateTime, then an IllegalArgumentException
will be thrown.
g.inject("2023-08-02T00:00:00Z").asDate().dateAdd(DT.day, 7) (1) g.inject(["2023-08-02T00:00:00Z", "2023-08-03T00:00:00Z"]).unfold().asDate().dateAdd(DT.minute, 1) (2)
-
Add 7 days to Date
-
Add 1 minute to incoming dates
Additional References
The dateDiff()
-step (map) returns the difference between two Dates in epoch time.
If the incoming traverser is not a Date or OffsetDateTime, then an IllegalArgumentException
will be thrown.
g.inject("2023-08-02T00:00:00Z").asDate().dateDiff(constant("2023-08-03T00:00:00Z").asDate()) (1)
-
Find difference between two dates
Additional References
With dedup()
-step (filter), repeatedly seen objects are removed from the traversal stream. Note that if a
traverser’s bulk is greater than 1, then it is set to 1 before being emitted.
g.V().values('lang') g.V().values('lang').dedup() g.V(1).repeat(bothE('created').dedup().otherV()).emit().path() (1) g.V().bothE().properties().dedup() (2)
-
Traverse all
created
edges, but don’t touch any edge twice. -
Note that
Property
instances will compare on key and value, whereas aVertexProperty
will also include its element as it is a first-class citizen.
If a by-step modulation is provided to dedup()
, then the object is processed accordingly prior to determining if it
has been seen or not.
g.V().elementMap('name') g.V().dedup().by(label).values('name')
If dedup()
is provided an array of strings, then it will ensure that the de-duplication is not with respect to the
current traverser object, but to the path history of the traverser.
g.V().as('a').out('created').as('b').in('created').as('c').select('a','b','c') g.V().as('a').out('created').as('b').in('created').as('c').dedup('a','b').select('a','b','c') (1) g.V().as('a').both().as('b').both().as('c'). dedup('a','b').by('age'). (2) select('a','b','c').by('name')
-
If the current
a
andb
combination has been seen previously, then filter the traverser. -
The "age" property is not productive for all vertices and therefore those values are filtered.
The dedup()
step can work on many different types of objects. One object in particular can need a bit of explanation.
If you use dedup()
on a Path
object there is a chance that you may get some unexpected results. Consider the
following example which forcibly generates duplicate path results in the first traversal and in the second applies
dedup()
to remove them:
g.V().union(out().path(), out().path()) g.V().union(out().path(), out().path()).dedup()
The dedup()
step checks the equality of the paths by examining the equality of the objects on the Path
(in this case
vertices), but also on any path labels. In the prior example, there weren’t any path labels so dedup()
behaved as
expected. In the next example, note the difference in the results if a label is added for one Path
but not the other:
g.V().union(out().as('x').path(), out().path()) g.V().union(out().as('x').path(), out().path()).dedup()
The prior example shows how dedup()
does not have the same effect when a path label is in place. In this contrived
example the answer is simple: remove the as('x')
. If in the real world, it is not possible to remove the label, the
workaround is to deconstruct the Path
into a List
to drop the label. In this way, dedup()
is just comparing List
objects and the objects in the Path
.
g.V().union(out().as('x').path(), out().path()).map(unfold().fold()).dedup()
Additional References
The difference()
-step (map) calculates the difference between the incoming list traverser and the provided list
argument. More specifically, this provides the set operation A-B where A is the traverser and B is the argument. This
step only expects list data (array or Iterable) and will throw an IllegalArgumentException
if any other type is
encountered (including null
).
g.V().values("name").fold().difference(["lop","ripple"]) g.V().values("name").fold().difference(__.V().limit(2).values("name").fold())
Additional References
The discard()
-step (filter) filters all objects from a traversal stream. It is especially useful for traversals
that are executed remotely where returning results is not useful and the traversal is only meant to generate
side-effects. Choosing not to return results saves in serialization and network costs as the objects are filtered on
the remote end and not returned to the client side. Typically, this step does not need to be used directly and is
quietly used by the iterate()
terminal step which appends discard()
to the traversal before actually cycling through
results.
Additional References
The disjunct()
-step (map) calculates the disjunct set between the incoming list traverser and the provided list
argument. This step only expects list data (array or Iterable) and will throw an IllegalArgumentException
if any other
type is encountered (including null
).
g.V().values("name").fold().disjunct(["lop","peter","sam"]) (1) g.V().values("name").fold().disjunct(__.V().limit(3).values("name").fold())
-
Find the unique names between two group of names
Additional References
The drop()
-step (filter/sideEffect) is used to remove element and properties from the graph (i.e. remove). It
is a filter step because the traversal yields no outgoing objects.
g.V().outE().drop() g.E() g.V().properties('name').drop() g.V().elementMap() g.V().drop() g.V()
Additional References
The E()
-step is meant to read edges from the graph and is usually used to start a GraphTraversal
, but can also
be used mid-traversal.
g.E(11) (1) g.E().hasLabel('knows').has('weight', gt(0.75)) g.inject(1).coalesce(E().hasLabel("knows"), addE("knows").from(V().has("name","josh")).to(V().has("name","vadas"))) (2)
-
Find the edge by its unique identifier (i.e.
T.id
) - not all graphs will use a numeric value for their identifier. -
Get edges with label
knows
, if there is none then add new one betweenjosh
andvadas
.
Additional References
The element()
step is a no-argument step that traverses from a Property
to the Element
that owns it.
g.V().properties().element() (1) g.E().properties().element() (2) g.V().properties().properties().element() (3)
-
Traverse from
VertexProperty
toVertex
-
Traverse from
Property
(edge property) toEdge
-
Traverse from
Property
(meta property) toVertexProperty
Additional References
The elementMap()
-step yields a Map
representation of the structure of an element.
g.V().elementMap() g.V().elementMap('age') g.V().elementMap('age','blah') g.E().elementMap()
It is important to note that the map of a vertex assumes that cardinality for each key is single
and if it is list
then only the first item encountered will be returned. As single
is the more common cardinality for properties this
assumption should serve the greatest number of use cases.
g.V().elementMap() g.V().has('name','marko').properties('location') g.V().has('name','marko').properties('location').elementMap()
Important
|
The elementMap() -step does not return the vertex labels for incident vertices when using GraphComputer
as the id is the only available data to the star graph.
|
Additional References
The emit
-step is not an actual step, but is instead a step modulator for repeat()
(find more
documentation on the emit()
there).
Additional References
The explain()
-step (terminal) will return a TraversalExplanation
. A traversal explanation details how the
traversal (prior to explain()
) will be compiled given the registered traversal strategies.
A TraversalExplanation
has a toString()
representation with 3-columns. The first column is the
traversal strategy being applied. The second column is the traversal strategy category: [D]ecoration, [O]ptimization,
[P]rovider optimization, [F]inalization, and [V]erification. Finally, the third column is the state of the traversal
post strategy application. The final traversal is the resultant execution plan.
g.V().hasLabel('person').outE().identity().inV().count().is(gt(5)).explain()
For traversal profiling information, please see profile()
-step.
The fail()
-step provides a way to force a traversal to immediately fail with an exception. This feature is often
helpful during debugging purposes and for validating certain conditions prior to continuing with traversal execution.
gremlin> g.V().has('person','name','peter').fold().
......1> coalesce(unfold(),
......2> fail('peter should exist')).
......3> property('k',100)
==>v[6]
gremlin> g.V().has('person','name','stephen').fold().
......1> coalesce(unfold(),
......2> fail('stephen should exist')).
......3> property('k',100)
fail() Step Triggered
===========================================================================================================================
Message > stephen should exist
Traverser> []
Bulk > 1
Traversal> fail()
Parent > CoalesceStep [V().has("person","name","stephen").fold().coalesce(__.unfold(),__.fail()).property("k",(int) 100)]
Metadata > {}
===========================================================================================================================
The code example above exemplifies the latter use case where there is essentially an assertion that there is a vertex with a particular "name" value prior to updating the property "k" and explicitly failing when that vertex is not found.
Additional References
The filter()
step maps the traverser from the current object to either true
or false
where the latter will not
pass the traverser to the next step in the process. Please see the General Steps section for more
information.
Additional References
The flatMap()
step maps the traverser from the current object to an Iterator
of objects for the next step in the
process. Please see the General Steps section for more information.
Additional References
This step is designed to simplify some string operations. In general, it is similar to the string formatting function available in many programming languages. Variable values can be picked up from Element properties, maps and scope variables.
g.V().format("%{name} is %{age} years old") (1) g.V().hasLabel("person").as("a").values("name").as("p1").select("a").in("knows").format("%{p1} knows %{name}") (2) g.V().format("%{name} has %{_} connections").by(bothE().count()) (3) g.V().project("name","count").by(values("name")).by(bothE().count()).format("%{name} has %{count} connections") (4)
-
A
format()
will use property values from incoming Element to produce String result. -
A
format()
will use scope variablep1
and propertyname
to resolve variable values. -
A
format()
will use propertyname
and traversal product for positional argument to resolve variable values. -
A
format()
will use map produced byproject
step to resolve variable values.
Additional References
There are situations when the traversal stream needs a "barrier" to aggregate all the objects and emit a computation
that is a function of the aggregate. The fold()
-step (map) is one particular instance of this. Please see
unfold()
-step for the inverse functionality.
g.V(1).out('knows').values('name') g.V(1).out('knows').values('name').fold() (1) g.V(1).out('knows').values('name').fold().next().getClass() (2) g.V(1).out('knows').values('name').fold(0) {a,b -> a + b.length()} (3) g.V().values('age').fold(0) {a,b -> a + b} (4) g.V().values('age').fold(0, sum) (5) g.V().values('age').sum() (6) g.inject(["a":1],["b":2]).fold([], addAll) (7)
-
A parameterless
fold()
will aggregate all the objects into a list and then emit the list. -
A verification of the type of list returned.
-
fold()
can be provided two arguments — a seed value and a reduce bi-function ("vadas" is 5 characters + "josh" with 4 characters). -
What is the total age of the people in the graph?
-
The same as before, but using a built-in bi-function.
-
The same as before, but using the
sum()
-step. -
A mechanism for merging
Map
instances. If a key occurs in more than a singleMap
, the later occurrence will replace the earlier.
Additional References
The from()
-step is not an actual step, but instead is a "step-modulator" similar to as()
and
by()
. If a step is able to accept traversals or strings then from()
is the
means by which they are added. The general pattern is step().from()
. See to()
-step.
The list of steps that support from()
-modulation are: simplePath()
, cyclicPath()
,
path()
, and addE()
.
Javascript
|
The term |
Python
|
The term |
Additional References
As traversers propagate across a graph as defined by a traversal, sideEffect computations are sometimes required.
That is, the actual path taken or the current location of a traverser is not the ultimate output of the computation,
but some other representation of the traversal. The group()
-step (map/sideEffect) is one such sideEffect that
organizes the objects according to some function of the object. Then, if required, that organization (a list) is
reduced. An example is provided below.
g.V().group().by(label) (1) g.V().group().by(label).by('name') (2) g.V().group().by(label).by(count()) (3)
-
Group the vertices by their label.
-
For each vertex in the group, get their name.
-
For each grouping, what is its size?
The two projection parameters available to group()
via by()
are:
-
Key-projection: What feature of the object to group on (a function that yields the map key)?
-
Value-projection: What feature of the group to store in the key-list?
g.V().group().by('age').by('name') (1) g.V().group().by('name').by('age') (2)
-
The "age" property is not productive for all vertices and therefore those keys are filtered.
-
The "age" property is not productive for all vertices and therefore those values are filtered.
Additional References
When it is important to know how many times a particular object has been at a particular part of a traversal,
groupCount()
-step (map/sideEffect) is used.
"What is the distribution of ages in the graph?"
g.V().hasLabel('person').values('age').groupCount() g.V().hasLabel('person').groupCount().by('age') (1) g.V().groupCount().by('age') (2)
-
You can also supply a pre-group projection, where the provided
by()
-modulation determines what to group the incoming object by. -
The "age" property is not productive for all vertices and therefore those values are filtered.
There is one person that is 32, one person that is 35, one person that is 27, and one person that is 29.
"Iteratively walk the graph and count the number of times you see the second letter of each name."
g.V().repeat(both().groupCount('m').by(label)).times(10).cap('m')
The above is interesting in that it demonstrates the use of referencing the internal Map<Object,Long>
of
groupCount()
with a string variable. Given that groupCount()
is a sideEffect-step, it simply passes the object
it received to its output. Internal to groupCount()
, the object’s count is incremented.
Additional References
It is possible to filter vertices, edges, and vertex properties based on their properties using has()
-step
(filter). There are numerous variations on has()
including:
-
has(key,value)
: Remove the traverser if its element does not have the provided key/value property. -
has(label, key, value)
: Remove the traverser if its element does not have the specified label and provided key/value property. -
has(key,predicate)
: Remove the traverser if its element does not have a key value that satisfies the bi-predicate. For more information on predicates, please read A Note on Predicates. -
hasLabel(labels…)
: Remove the traverser if its element does not have any of the labels. -
hasId(ids…)
: Remove the traverser if its element does not have any of the ids. -
hasKey(keys…)
: Remove theProperty
traverser if it does not match one of the provided keys. -
hasValue(values…)
: Remove theProperty
traverser if it does not match one of the provided values. -
has(key)
: Remove the traverser if its element does not have a value for the key. -
hasNot(key)
: Remove the traverser if its element has a value for the key. -
has(key, traversal)
: Remove the traverser if its object does not yield a result through the traversal off the property value.
g.V().hasLabel('person') g.V().hasLabel('person','name','marko') g.V().hasLabel('person').out().has('name',within('vadas','josh')) g.V().hasLabel('person').out().has('name',within('vadas','josh')). outE().hasLabel('created') g.V().has('age',inside(20,30)).values('age') (1) g.V().has('age',outside(20,30)).values('age') (2) g.V().has('name',within('josh','marko')).elementMap() (3) g.V().has('name',without('josh','marko')).elementMap() (4) g.V().has('name',not(within('josh','marko'))).elementMap() (5) g.V().properties().hasKey('age').value() (6) g.V().hasNot('age').values('name') (7) g.V().has('person','name', startingWith('m')) (8) g.V().has(null, 'vadas') (9) g.V().has(label, __.is('person')) (10)
-
Find all vertices whose ages are between 20 (exclusive) and 30 (exclusive). In other words, the age must be greater than 20 and less than 30.
-
Find all vertices whose ages are not between 20 (inclusive) and 30 (inclusive). In other words, the age must be less than 20 or greater than 30.
-
Find all vertices whose names are exact matches to any names in the collection
[josh,marko]
, display all the key,value pairs for those vertices. -
Find all vertices whose names are not in the collection
[josh,marko]
, display all the key,value pairs for those vertices. -
Same as the prior example save using
not
onwithin
to yieldwithout
. -
Find all age-properties and emit their value.
-
Find all vertices that do not have an age-property and emit their name.
-
Find all "person" vertices that have a name property that starts with the letter "m".
-
Property key is always stored as
String
and therefore an equality check withnull
will produce no result. -
An example of
has()
where the argument is aTraversal
and does not quite behave the way most expect.
Item 10 in the above set of examples bears some discussion. The behavior is not such that the result of the Traversal
is used as the comparing value for has()
, but the current Traverser
, which in this case is the vertex label
, is
given to the Traversal
to behave as a filter itself. In other words, if the Traversal
(i.e. is('person')
) returns
a value then the has()
is effectively true
. A common mistake is to try to use select()
in this context where one
would do has('name', select('n'))
to try to inject the value of "n" into the step to get has('name', <value-of-n>)
,
but this would instead simply produce an always true
filter for has()
.
TinkerPop does not support a regular expression predicate, although specific graph databases that leverage TinkerPop may provide a partial match extension.
Additional References
has(String)
,
has(String,Object)
,
has(String,P)
,
has(String,String,Object)
,
has(String,String,P)
,
has(String,Traversal)
,
has(T,Object)
,
has(T,P)
,
has(T,Traversal)
,
hasId(Object,Object…)
,
hasId(P)
,
hasKey(P)
,
hasKey(String,String…)
,
hasLabel(P)
,
hasLabel(String,String…)
,
hasNot(String)
,
hasValue(Object,Object…)
,
hasValue(P)
,
P
,
TextP
,
T
,
Recipes - Anti-pattern
The id()
-step (map) takes an Element
and extracts its identifier from it.
g.V().id() g.V(1).out().id().is(2) g.V(1).outE().id() g.V(1).properties().id()
Additional References
The identity()
-step (map) is an identity function which maps
the current object to itself.
g.V().identity()
Additional References
The index()
-step (map) indexes each element in the current collection. If the current traverser’s value is not a collection, then it’s treated as a single-item collection. There are two indexers
available, which can be chosen using the with()
modulator. The list indexer (default) creates a list for each collection item, with the first item being the original element and the second element
being the index. The map indexer created a linked hash map in which the index represents the key and the original item is used as the value.
g.V().hasLabel("software").index() (1) g.V().hasLabel("software").values("name").fold(). order(Scope.local). index(). unfold(). order(). by(__.tail(Scope.local, 1)) (2) g.V().hasLabel("software").values("name").fold(). order(Scope.local). index(). with(WithOptions.indexer, WithOptions.list). unfold(). order(). by(__.tail(Scope.local, 1)) (3) g.V().hasLabel("person").values("name").fold(). order(Scope.local). index(). with(WithOptions.indexer, WithOptions.map) (4)
-
Indexing non-collection items results in multiple indexed single-item collections.
-
Index all software names in their alphabetical order.
-
Same as statement 1, but with an explicitely specified list indexer.
-
Index all person names in their alphabetical order and store the result in an ordered map.
Additional References
The concept of "injectable steps" makes it possible to insert objects arbitrarily into a traversal stream. In general,
inject()
-step (sideEffect) exists and a few examples are provided below.
g.V(4).out().values('name').inject('daniel') g.V(4).out().values('name').inject('daniel').map {it.get().length()} g.V(4).out().values('name').inject('daniel').map {it.get().length()}.path()
In the last example above, note that the path starting with daniel
is only of length 2. This is because the
daniel
string was inserted half-way in the traversal. Finally, a typical use case is provided below — when the
start of the traversal is not a graph object.
inject(1,2) inject(1,2).map {it.get() + 1} inject(1,2).map {it.get() + 1}.map {g.V(it.get()).next()}.values('name')
Additional References
The intersect()
-step (map) calculates the intersection between the incoming list traverser and the provided list
argument. This step only expects list data (array or Iterable) and will throw an IllegalArgumentException
if any other
type is encountered (including null
).
g.V().values("name").fold().intersect(["marko","josh","james","jen"]) g.V().values("name").fold().intersect(__.V().limit(2).values("name").fold())
Additional References
The task of importing and exporting the data of
Graph
instances is the
job of the io()
-step. By default, TinkerPop supports three formats for importing and exporting graph data in
GraphML, GraphSON, and Gryo.
Note
|
Additional documentation for TinkerPop IO formats can be found in the IO Reference. |
By itself the io()
-step merely configures the kind of importing and exporting that is going
to occur and it is the follow-on call to the read()
or write()
step that determines which of those actions will
execute. Therefore, a typical usage of the io()
-step would look like this:
g.io(someInputFile).read().iterate()
g.io(someOutputFile).write().iterate()
Important
|
The commands above are still traversals and therefore require iteration to be executed, hence the use of
iterate() as a termination step.
|
By default, the io()
-step will try to detect the right file format using the file name extension. To gain greater
control of the format use the with()
step modulator to provide further information to io()
. For example:
g.io(someInputFile).
with(IO.reader, IO.graphson).
read().iterate()
g.io(someOutputFile).
with(IO.writer,IO.graphml).
write().iterate()
The IO
class is a helper for the io()
-step that provides expressions that can be used to help configure it
and in this case it allows direct specification of the "reader" or "writer" to use. The "reader" actually refers to
a GraphReader
implementation and the "writer" refers to a GraphWriter
implementation. The implementations of
those interfaces provided by default are the standard TinkerPop implementations.
That default is an important point to consider for users. The default TinkerPop implementations are not designed with massive, complex, parallel bulk loading in mind. They are designed to do single-threaded, OLTP-style loading of data in the most generic way possible so as to accommodate the greatest number of graph databases out there. As such, from a reading perspective, they work best for small datasets (or perhaps medium datasets where memory is plentiful and time is not critical) that are loading to an empty graph - incremental loading is not supported. The story from the writing perspective is not that different in there are no parallel operations in play, however streaming the output to disk requires a single pass of the data without high memory requirements for larger datasets.
Important
|
Default graph formats don’t contain information about property cardinality, so it is up to the graph provider to choose the appropriate one. You will see a warning message if the chosen cardinality is SINGLE while your graph input contains multiple values for that property. |
In general, TinkerPop recommends that users examine the native bulk import/export tools of the graph implementation
that they choose. Those tools will often outperform the io()
-step and perhaps be easier to use with a greater
feature set. That said, graph providers do have the option to optimize io()
to back it with their own
import/export utilities and therefore the default behavior provided by TinkerPop described above might be overridden
by the graph.
An excellent example of this lies in HadoopGraph with SparkGraphComputer
which replaces the default single-threaded implementation with a more advanced OLAP style bulk import/export
functionality internally using CloneVertexProgram. With this model, graphs of arbitrary size
can be imported/exported assuming that there is a Hadoop InputFormat
or OutputFormat
to support it.
Important
|
Remote Gremlin Console users or Gremlin Language Variant (GLV) users (e.g. gremlin-python) who utilize
the io() -step should recall that their read() or write() operation will occur on the server and not locally
and therefore the file specified for import/export must be something accessible by the server.
|
GraphSON and Gryo formats are extensible allowing users and graph providers to extend supported serialization options.
These extensions are exposed through IoRegistry
implementations. To apply an IoRegistry
use the with()
option
and the IO.registry
key, where the value is either an actual IoRegistry
instance or the fully qualified class
name of one.
g.io(someInputFile).
with(IO.reader, IO.gryo).
with(IO.registry, TinkerIoRegistryV3d0.instance())
read().iterate()
g.io(someOutputFile).
with(IO.writer,IO.graphson).
with(IO.registry, "org.apache.tinkerpop.gremlin.tinkergraph.structure.TinkerIoRegistryV3d0")
write().iterate()
GLVs will obviously always be forced to use the latter form as they can’t explicitly create an instance of an
IoRegistry
to pass to the server (nor are IoRegistry
instances necessarily serializable).
The version of the formats (e.g. GraphSON 2.0 or 3.0) utilized by io()
is determined entirely by the IO.reader
and
IO.writer
configurations or their defaults. The defaults will always be the latest version for the current release
of TinkerPop. It is also possible for graph providers to override these defaults, so consult the documentation of the
underlying graph database in use for any details on that.
Note
|
The io() step will try to automatically detect the appropriate GraphReader or GraphWriter to use based on
the file extension. If the file has a different extension than the ones expected, use with() as shown above to set the
reader or writer explicitly.
|
For more advanced configuration of GraphReader
and GraphWriter
operations (e.g. normalized output for GraphSON,
disabling class registrations for Gryo, etc.) then construct the appropriate GraphReader
and GraphWriter
using
the build()
method on their implementations and use it directly. It can be passed directly to the IO.reader
or
IO.writer
options. Obviously, these are JVM based operations and thus not available to GLVs as portable features.
The GraphML file format is a
common XML-based representation of a graph. It is widely supported by graph-related tools and libraries making it a
solid interchange format for TinkerPop. In other words, if the intent is to work with graph data in conjunction with
applications outside of TinkerPop, GraphML may be the best choice to do that. Common use cases might be:
Warning
|
GraphML is a "lossy" format in that it only supports primitive values for properties and does not have
support for Graph variables. It will use toString to serialize property values outside of those primitives.
|
Warning
|
GraphML as a specification allows for <edge> and <node> elements to appear in any order. Most software
that writes GraphML (including as TinkerPop’s GraphMLWriter ) write <node> elements before <edge> elements.
However it is important to note that GraphMLReader will read this data in order and order can matter. This is because
TinkerPop does not allow the vertex label to be changed after the vertex has been created. Therefore, if an <edge>
element comes before the <node> , the label on the vertex will be ignored. It is thus better to order <node>
elements in the GraphML to appear before all <edge> elements if vertex labels are important to the graph.
|
// expects a file extension of .xml or .graphml to determine that
// a GraphML reader/writer should be used.
g.io("graph.xml").read().iterate();
g.io("graph.xml").write().iterate();
Note
|
If using GraphML generated from TinkerPop 2.x, read more about its incompatibilities in the Upgrade Documentation. |
GraphSON is a JSON-based format extended
from earlier versions of TinkerPop. It is important to note that TinkerPop’s GraphSON is not backwards compatible
with prior TinkerPop GraphSON versions. GraphSON has some support from graph-related application outside of TinkerPop,
but it is generally best used in two cases:
-
A text format of the graph or its elements is desired (e.g. debugging, usage in source control, etc.)
-
The graph or its elements need to be consumed by code that is not JVM-based (e.g. JavaScript, Python, .NET, etc.)
// expects a file extension of .json to interpret that
// a GraphSON reader/writer should be used
g.io("graph.json").read().iterate();
g.io("graph.json").write().iterate();
Note
|
Additional documentation for GraphSON can be found in the IO Reference. |
Kryo is a popular
serialization package for the JVM. Gremlin-Kryo is a binary
Graph
serialization format for use on the JVM by JVM
languages. It is designed to be space efficient, non-lossy and is promoted as the standard format to use when working
with graph data inside of the TinkerPop stack. A list of common use cases is presented below:
-
Migration from one Gremlin Structure implementation to another (e.g.
TinkerGraph
toNeo4jGraph
) -
Serialization of individual graph elements to be sent over the network to another JVM.
-
Backups of in-memory graphs or subgraphs.
Warning
|
When migrating between Gremlin Structure implementations, Kryo may not lose data, but it is important to
consider the features of each Graph and whether or not the data types supported in one will be supported in the
other. Failure to do so, may result in errors.
|
// expects a file extension of .kryo to interpret that
// a GraphSON reader/writer should be used
g.io("graph.kryo").read().iterate()
g.io("graph.kryo").write().iterate()
Additional References
It is possible to filter scalar values using is()
-step (filter).
Python
|
The term |
g.V().values('age').is(32) g.V().values('age').is(lte(30)) g.V().values('age').is(inside(30, 40)) g.V().where(__.in('created').count().is(1)).values('name') (1) g.V().where(__.in('created').count().is(gte(2))).values('name') (2) g.V().where(__.in('created').values('age'). mean().is(inside(30d, 35d))).values('name') (3)
-
Find projects having exactly one contributor.
-
Find projects having two or more contributors.
-
Find projects whose contributors average age is between 30 and 35.
Additional References
is(Object)
,
is(P)
,
P
The key()
-step (map) takes a Property
and extracts the key from it.
g.V(1).properties().key() g.V(1).properties().properties().key()
Additional References
The label()
-step (map) takes an Element
and extracts its label from it.
g.V().label() g.V(1).outE().label() g.V(1).properties().label()
Additional References
The length()
-step (map) returns the length incoming string or list of string traverser. Null values are not processed and remain as null when returned.
If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.V().values('name').length() (1) g.V().values('name').fold().length(local) (2)
-
Return the string length of all vertex names.
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
Additional References
The limit()
-step is analogous to range()
-step save that the lower end range is set to 0.
g.V().limit(2) g.V().range(0, 2)
The limit()
-step can also be applied with Scope.local
, in which case it operates on the incoming collection.
The examples below use the The Crew toy data set.
g.V().valueMap().select('location').limit(local,2) (1) g.V().valueMap().limit(local, 1) (2)
-
List<String>
for each vertex containing the first two locations. -
Map<String, Object>
for each vertex, but containing only the first property value.
Additional References
A GraphTraversal
operates on a continuous stream of objects. In many situations, it is important to operate on a
single element within that stream. To do such object-local traversal computations, local()
-step exists (branch).
Note that the examples below use the The Crew toy data set.
g.V().as('person'). properties('location').order().by('startTime',asc).limit(2).value().as('location'). select('person','location').by('name').by() (1) g.V().as('person'). local(properties('location').order().by('startTime',asc).limit(2)).value().as('location'). select('person','location').by('name').by() (2)
-
Get the first two people and their respective location according to the most historic location start time.
-
For every person, get their two most historic locations.
The two traversals above look nearly identical save the inclusion of local()
which wraps a section of the traversal
in an object-local traversal. As such, the order().by()
and the limit()
refer to a particular object, not to the
stream as a whole.
Local Step is quite similar in functionality to Flat Map Step where it can often be confused.
local()
propagates the traverser through the internal traversal as is without splitting/cloning it. Thus, its
a “global traversal” with local processing. Its use is subtle and primarily finds application in compilation
optimizations (i.e. when writing TraversalStrategy
implementations. As another example consider:
g.V().both().barrier().flatMap(groupCount().by("name")) g.V().both().barrier().local(groupCount().by("name"))
Use of local()
is often a mistake. This is especially true when its argument contains a reducing step. For example,
let’s say the requirement was to count the number of properties per Vertex
in:
g.V().both().local(properties('name','age').count()) (1) g.V().both().map(properties('name','age').count()) (2)
-
The output here seems impossible because no single vertex in the "modern" graph can have more than two properties given the "name" and "age" filters, but because the counting is happening object-local the counting is occurring unique to each object rather than each global traverser.
-
Replacing
local()
withmap()
returns the result desired by the requirement.
Warning
|
The anonymous traversal of local() processes the current object "locally." In OLAP, where the atomic unit
of computing is the vertex and its local "star graph," it is important that the anonymous traversal does not leave
the confines of the vertex’s star graph. In other words, it can not traverse to an adjacent vertex’s properties or edges.
|
Additional References
The loops()
-step (map) extracts the number of times the Traverser
has gone through the current loop.
g.V().emit(__.has("name", "marko").or().loops().is(2)).repeat(__.out()).values("name")
Additional References
The lTrim()
-step (map) returns a string with leading whitespace removed. Null values are not processed and remain
as null when returned. If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.inject(" hello ", " world ", null).lTrim() g.inject([" hello ", " world ", null]).lTrim(local) (1)
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
The map()
step maps the traverser from the current object to the next step in the process. Please see the
General Steps section for more information.
Additional References
The match()
-step (map) provides a more declarative
form of graph querying based on the notion of pattern matching.
With match()
, the user provides a collection of "traversal fragments," called patterns, that have variables defined
that must hold true throughout the duration of the match()
. When a traverser is in match()
, a registered
MatchAlgorithm
analyzes the current state of the traverser (i.e. its history based on its
path data), the runtime statistics of the traversal patterns, and returns a traversal-pattern
that the traverser should try next. The default MatchAlgorithm
provided is called CountMatchAlgorithm
and it
dynamically revises the pattern execution plan by sorting the patterns according to their filtering capabilities
(i.e. largest set reduction patterns execute first). For very large graphs, where the developer is uncertain of the
statistics of the graph (e.g. how many knows
-edges vs. worksFor
-edges exist in the graph), it is advantageous to
use match()
, as an optimal plan will be determined automatically. Furthermore, some queries are much easier to
express via match()
than with single-path traversals.
"Who created a project named 'lop' that was also created by someone who is 29 years old? Return the two creators."
g.V().match( __.as('a').out('created').as('b'), __.as('b').has('name', 'lop'), __.as('b').in('created').as('c'), __.as('c').has('age', 29)). select('a','c').by('name')
Note that the above can also be more concisely written as below which demonstrates that standard inner-traversals can be arbitrarily defined.
g.V().match( __.as('a').out('created').has('name', 'lop').as('b'), __.as('b').in('created').has('age', 29).as('c')). select('a','c').by('name')
In order to improve readability, as()
-steps can be given meaningful labels which better reflect your domain. The
previous query can thus be written in a more expressive way as shown below.
g.V().match( __.as('creators').out('created').has('name', 'lop').as('projects'), (1) __.as('projects').in('created').has('age', 29).as('cocreators')). (2) select('creators','cocreators').by('name') (3)
-
Find vertices that created something and match them as 'creators', then find out what they created which is named 'lop' and match these vertices as 'projects'.
-
Using these 'projects' vertices, find out their creators aged 29 and remember these as 'cocreators'.
-
Return the name of both 'creators' and 'cocreators'.
MatchStep
brings functionality similar to SPARQL to Gremlin. Like SPARQL,
MatchStep conjoins a set of patterns applied to a graph. For example, the following traversal finds exactly those
songs which Jerry Garcia has both sung and written (using the Grateful Dead graph distributed in the data/
directory):
g = traversal().with(graph) g.io('data/grateful-dead.xml').read().iterate() g.V().match( __.as('a').has('name', 'Garcia'), __.as('a').in('writtenBy').as('b'), __.as('a').in('sungBy').as('b')). select('b').values('name')
Among the features which differentiate match()
from SPARQL are:
g.V().match( __.as('a').out('created').has('name','lop').as('b'), (1) __.as('b').in('created').has('age', 29).as('c'), __.as('c').repeat(out()).times(2)). (2) select('c').out('knows').dedup().values('name') (3)
-
Patterns of arbitrary complexity:
match()
is not restricted to triple patterns or property paths. -
Recursion support:
match()
supports the branch-based steps within a pattern, includingrepeat()
. -
Imperative/declarative hybrid: Before and after a
match()
, it is possible to leverage classic Gremlin traversals.
To extend point #3, it is possible to support going from imperative, to declarative, to imperative, ad infinitum.
g.V().match( __.as('a').out('knows').as('b'), __.as('b').out('created').has('name','lop')). select('b').out('created'). match( __.as('x').in('created').as('y'), __.as('y').out('knows').as('z')). select('z').values('name')
Important
|
The match() -step is stateless. The variable bindings of the traversal patterns are stored in the path
history of the traverser. As such, the variables used over all match() -steps within a traversal are globally unique.
A benefit of this is that subsequent where() , select() , match() , etc. steps can leverage the same variables in
their analysis.
|
Like all other steps in Gremlin, match()
is a function and thus, match()
within match()
is a natural consequence
of Gremlin’s functional foundation (i.e. recursive matching).
g.V().match( __.as('a').out('knows').as('b'), __.as('b').out('created').has('name','lop'), __.as('b').match( __.as('b').out('created').as('c'), __.as('c').has('name','ripple')). select('c').as('c')). select('a','c').by('name')
If a step-labeled traversal proceeds the match()
-step and the traverser entering the match()
is destined to bind
to a particular variable, then the previous step should be labeled accordingly.
g.V().as('a').out('knows').as('b'). match( __.as('b').out('created').as('c'), __.not(__.as('c').in('created').as('a'))). select('a','b','c').by('name')
There are three types of match()
traversal patterns.
-
as('a')…as('b')
: both the start and end of the traversal have a declared variable. -
as('a')…
: only the start of the traversal has a declared variable. -
…
: there are no declared variables.
If a variable is at the start of a traversal pattern it must exist as a label in the path history of the traverser
else the traverser can not go down that path. If a variable is at the end of a traversal pattern then if the variable
exists in the path history of the traverser, the traverser’s current location must match (i.e. equal) its historic
location at that same label. However, if the variable does not exist in the path history of the traverser, then the
current location is labeled as the variable and thus, becomes a bound variable for subsequent traversal patterns. If a
traversal pattern does not have an end label, then the traverser must simply "survive" the pattern (i.e. not be
filtered) to continue to the next pattern. If a traversal pattern does not have a start label, then the traverser
can go down that path at any point, but will only go down that pattern once as a traversal pattern is executed once
and only once for the history of the traverser. Typically, traversal patterns that do not have a start and end label
are used in conjunction with and()
, or()
, and where()
. Once the traverser has "survived" all the patterns (or at
least one for or()
), match()
-step analyzes the traverser’s path history and emits a Map<String,Object>
of the
variable bindings to the next step in the traversal.
g.V().as('a').out().as('b'). (1) match( (2) __.as('a').out().count().as('c'), (3) __.not(__.as('a').in().as('b')), (4) or( (5) __.as('a').out('knows').as('b'), __.as('b').in().count().as('c').and().as('c').is(gt(2)))). (6) dedup('a','c'). (7) select('a','b','c').by('name').by('name').by() (8)
-
A standard, step-labeled traversal can come prior to
match()
. -
If the traverser’s path prior to entering
match()
has requisite label values, then those historic values are bound. -
It is possible to use barrier steps though they are computed locally to the pattern (as one would expect).
-
It is possible to
not()
a pattern. -
It is possible to nest
and()
- andor()
-steps for conjunction matching. -
Both infix and prefix conjunction notation is supported.
-
It is possible to "distinct" the specified label combination.
-
The bound values are of different types — vertex ("a"), vertex ("b"), long ("c").
Match is typically used in conjunction with both select()
(demonstrated previously) and where()
(presented here).
A where()
-step allows the user to further constrain the result set provided by match()
.
g.V().match( __.as('a').out('created').as('b'), __.as('b').in('created').as('c')). where('a', neq('c')). select('a','c').by('name')
The where()
-step can take either a P
-predicate (example above) or a Traversal
(example below). Using
MatchPredicateStrategy
, where()
-clauses are automatically folded into match()
and thus, subject to the query
optimizer within match()
-step.
traversal = g.V().match( __.as('a').has(label,'person'), (1) __.as('a').out('created').as('b'), __.as('b').in('created').as('c')). where(__.as('a').out('knows').as('c')). (2) select('a','c').by('name'); null (3) traversal.toString() (4) traversal (5) (6) traversal.toString() (7)
-
Any
has()
-step traversal patterns that start with the match-key are pulled out ofmatch()
to enable the graph system to leverage the filter for index lookups. -
A
where()
-step with a traversal containing variable bindings declared inmatch()
. -
A useful trick to ensure that the traversal is not iterated by Gremlin Console.
-
The string representation of the traversal prior to its strategies being applied.
-
The Gremlin Console will automatically iterate anything that is an iterator or is iterable.
-
Both marko and josh are co-developers and marko knows josh.
-
The string representation of the traversal after the strategies have been applied (and thus,
where()
is folded intomatch()
)
Important
|
A where() -step is a filter and thus, variables within a where() clause are not globally bound to the
path of the traverser in match() . As such, where() -steps in match() are used for filtering, not binding.
|
Additional References
The math()
-step (math) enables scientific calculator functionality within Gremlin. This step deviates from the common
function composition and nesting formalisms to provide an easy to read string-based math processor. Variables within the
equation map to scopes in Gremlin — e.g. path labels, side-effects, or incoming map keys. This step supports
by()
-modulation where the by()
-modulators are applied in the order in which the variables are first referenced
within the equation. Note that the reserved variable _
refers to the current numeric traverser object incoming to the
math()
-step.
g.V().as('a').out('knows').as('b').math('a + b').by('age') g.V().as('a').out('created').as('b'). math('b + a'). by(both().count().math('_ + 100')). by('age') g.withSideEffect('x',10).V().values('age').math('_ / x') g.withSack(1).V(1).repeat(sack(sum).by(constant(1))).times(10).emit().sack().math('sin _') g.V().math('_+1').by('age') (1)
-
The "age" property is not productive for all vertices and therefore those values are filtered.
The operators supported by the calculator include: *
, +
, /
, ^
, and %
. Furthermore, the following built in
functions are provided:
-
abs
: absolute value -
acos
: arc cosine -
asin
: arc sine -
atan
: arc tangent -
cbrt
: cubic root -
ceil
: nearest upper integer -
cos
: cosine -
cosh
: hyperbolic cosine -
exp
: euler’s number raised to the power (e^x
) -
floor
: nearest lower integer -
log
: logarithmus naturalis (base e) -
log10
: logarithm (base 10) -
log2
: logarithm (base 2) -
sin
: sine -
sinh
: hyperbolic sine -
sqrt
: square root -
tan
: tangent -
tanh
: hyperbolic tangent -
signum
: signum function
Additional References
The max()
-step (map) operates on a stream of comparable objects and determines which is the last object according
to its natural order in the stream.
g.V().values('age').max() g.V().repeat(both()).times(3).values('age').max() g.V().values('name').max()
When called as max(local)
it determines the maximum value of the current, local object (not the objects in the
traversal stream). This works for Collection
and Comparable
-type objects.
g.V().values('age').fold().max(local)
When there are null
values being evaluated the null
objects are ignored, but if all values are recognized as null
the return value is null
.
g.inject(null,10, 9, null).max() g.inject([null,null,null]).max(local)
Additional References
The mean()
-step (map) operates on a stream of numbers and determines the average of those numbers.
g.V().values('age').mean() g.V().repeat(both()).times(3).values('age').mean() (1) g.V().repeat(both()).times(3).values('age').dedup().mean()
-
Realize that traversers are being bulked by
repeat()
. There may be more of a particular number than another, thus altering the average.
When called as mean(local)
it determines the mean of the current, local object (not the objects in the traversal
stream). This works for Collection
and Number
-type objects.
g.V().values('age').fold().mean(local)
If mean()
encounters null
values, they will be ignored (i.e. their traversers not counted toward toward the
divisor). If all traversers are null
then the stream will return null
.
g.inject(null,10, 9, null).mean() g.inject([null,null,null]).mean(local)
Additional References
The merge()
-step (map) combines collections like lists and maps. It expects an incoming traverser to contain a
collection objection and will combine that object with its specified argument which must be of a matching type. This is
also known as the union operation. If the incoming traverser or its associated argument do not meet the expected type,
the step will throw an IllegalArgumentException
if any other type is encountered (including null
). This step differs
from the combine()
-step in that it doesn’t allow duplicates.
g.V().values("name").fold().merge(["james","jen","marko","vadas"]) g.V().values("name").fold().merge(__.constant("james").fold()) g.V().hasLabel('software').elementMap().merge([year:2009])
Additional References
The mergeE()
step is used to add edges and their properties to a graph in a "create
if not exist" fashion. The mergeE()
step can also be used to find edges matching a given
pattern. The input passed to mergeE()
can be either a Map
, or a child traversal that
produces a Map
.
Note
|
There is a corresponding mergeV() step that can be used when creating vertices.
|
Additionally, option()
modulators may be combined with mergeE()
to take action depending on
whether a vertex was created, or already existed. There are various ways that mergeE()
can
be used. The simplest being to provide a single Map
of keys and values, along with the
source and target vertex IDs, as a parameter. A T.id
and a T.label
may also be provided but
this is optional. The mergeE()
step can be used directly from the GraphTraversalSource
- g
,
or in the middle of a traversal. For a match with an existing vertex to occur, all values
in the Map
must exist on a vertex; otherwise, a new vertex will be created. The examples
that follow show how mergeE()
can be used to add relationships between dogs in the graph.
g.mergeV([(T.id):1,(T.label):'Dog',name:'Toby']) g.mergeV([(T.id):2,(T.label):'Dog',name:'Brandy']) (1) g.mergeE([(T.label):'Sibling',created:'2022-02-07',(Direction.from):1,(Direction.to):2]) (2) g.E().elementMap()
-
Create two vertices with ID values of 1 and 2.
-
Create a "Sibling" relationship between the vertices.
Note
|
The example above is written with gremlin-groovy and evaluated in Gremlin Console as a Groovy script thus
allowing Groovy syntax for initializing a Map .
|
For a mergeE()
step to succeed, both the from
and to
vertices must already exist. It
is not possible to create new vertices directly using mergeE()
, but mergeV()
and mergeE()
steps can be combined, in a single query, to achieve that goal.
Note
|
The mergeE() step will not create vertices that do not exist. In those cases an
error will be returned.
|
If the Direction
enum has been statically included, its explicit use can be omitted from
the query.
g.mergeV([(T.id):1,(T.label):'Dog',name:'Toby']) g.mergeV([(T.id):2,(T.label):'Dog',name:'Brandy']) g.mergeE([(T.label):'Sibling',created:'2022-02-07',(from):1,(to):2]) g.E().elementMap()
One or more option()
steps can be used to control the behavior when an edge is created or
updated. Similar to mergeV()
, the onCreate Map
inherits from the main merge argument - any
existence criteria in the main merge argument (T.id
, T.label
, Direction.OUT
, Direction.IN
)
will be automatically carried over to the onCreate action, and these existence criteria cannot be overriden
in the onCreate Map
.
g.mergeV([(T.id):1,(T.label):'Dog',name:'Toby']) g.mergeV([(T.id):2,(T.label):'Dog',name:'Brandy']) g.withSideEffect('map',[(T.label):'Sibling',(from):1,(to):2]). mergeE(select('map')). option(Merge.onCreate,[created:'2022-02-07']). (1) option(Merge.onMatch,[updated:'2022-02-07']) g.E().elementMap() g.withSideEffect('map',[(T.label):'Sibling',(from):1,(to):2]). mergeE(select('map')). option(Merge.onCreate,[created:'2022-02-07']). option(Merge.onMatch,[updated:'2022-02-07']) (2) g.E().elementMap()
-
The edge did not exist - set the created date.
-
The edge did exist - set the updated date.
More than one edge can be created by a single mergeE()
operation. This is done by
injecting a list of maps into the traversal and letting them stream into the mergeE()
step.
maps = [[(T.label):'Siblings',(from):1,(to):2], [(T.label):'Siblings',(from):1,(to):3]] g.mergeV([(T.id):1,(T.label):'Dog',name:'Toby']) (1) g.mergeV([(T.id):2,(T.label):'Dog',name:'Brandy']) g.mergeV([(T.id):3,(T.label):'Dog',name:'Dax']) g.inject(maps).unfold().mergeE() (2) g.E().elementMap()
-
Create three dogs.
-
Stream the edge maps into
mergeE()
steps.
The mergeE
step can be combined with the mergeV
step (or any other step producing a Vertex
) using the
Merge.outV
and Merge.inV
option modulators. These options can be used to "late-bind" the OUT
and IN
vertices in the main merge argument and in the onCreate
argument:
g.mergeV([(T.id):1,(T.label):'Dog',name:'Toby']).as('Toby'). mergeV([(T.id):2,(T.label):'Dog',name:'Brandy']).as('Brandy'). mergeE([(T.label):'Sibling',created:'2022-02-07',(from):Merge.outV,(to):Merge.inV]). option(Merge.outV, select('Toby')). option(Merge.inV, select('Brandy')) g.E().elementMap()
The Merge.outV
and Merge.inV
tokens can be used as placeholders for values for Direction.OUT
and Direction.IN
respectively in the mergeE
arguments. These options can produce Vertices
, as in the example above, or they can
specify Maps
, which will be used to search for Vertices
in the graph. This is useful when the exact T.id
of
the from/to vertices is not known in advance:
g.mergeV([(T.label):'Dog',name:'Toby']) g.mergeV([(T.label):'Dog',name:'Brandy']) g.mergeE([(T.label):'Sibling',created:'2022-02-07',(from):Merge.outV,(to):Merge.inV]). option(Merge.outV, [(T.label):'Dog',name:'Toby']). option(Merge.inV, [(T.label):'Dog',name:'Brandy']) g.E().elementMap()
Additional References
The mergeV()
-step is used to add vertices and their properties to a graph in a "create
if not exist" fashion. The mergeV()
step can also be used to find vertices matching a given
pattern. The input passed to mergeV()
can be either a Map
, or a child Traversal
that
produces a Map
.
Note
|
There is a corresponding mergeE() step that can be used when creating edges.
|
Additionally, option()
modulators may be combined with mergeV()
to take action depending on
whether a vertex was created, or already existed. There are various ways mergeV()
can
be used. The simplest being to provide a single Map
of keys and values as a parameter. A T.id
and a T.label
may also be provided but this is optional. The mergeV()
step can be used directly
from the GraphTraversalSource
- g
, or in the middle of a traversal. For a match with an
existing vertex to occur, all values in the Map
must exist on a vertex; otherwise, a new
vertex will be created. The examples that follow show how mergeV()
can be used to add some
dogs to the graph.
g.mergeV([name: 'Brandy']) (1) g.V().has('name','Brandy') g.mergeV([(T.label):'Dog',name:'Scamp', age:12]) (2) g.V().hasLabel('Dog').valueMap() g.mergeV([(T.id):300, (T.label):'Dog', name:'Toby', age:10]) (3) g.V().hasLabel('Dog').valueMap().with(WithOptions.tokens)
-
Create a vertex for Brandy as no other matching ones exist yet.
-
Create a vertex for Scamp and also add a Dog label his age.
-
Create a vertex for Toby with an
T.id
of 300.
Note
|
The example above is written with gremlin-groovy and evaluated in Gremlin Console as a Groovy script thus
allowing Groovy syntax for initializing a Map .
|
If a vertex already exists that matches the map passed to mergeV()
, the existing
vertex will be returned, otherwise a new one will be created. In this way, mergeV()
provides "get or create" semantics.
g.mergeV([name: 'Brandy']) (1)
-
A vertex for Brandy already exists so return it. A new one is not created.
It’s important to note that every key/value pair passed to mergeV()
must already exist on
one or more vertices for there to be a match. If a match is found, the vertex, or
vertices, representing that match will be returned. If a vertex representing a dog called
Brandy already exists, but it does not have an "age" property, the mergeV()
below will not
find a match and a new vertex will be created.
g.addV('Dog').property('name','Brandy') (1) g.mergeV([(T.label):'Dog',name:'Brandy',age:13]) (2)
-
Create a vertex for Brandy with no age property.
-
A new vertex is created as there is no exact match to any existing vertices.
A common scenario is to search for a vertex with a known T.id
and if it exists return that
vertex. If it does not exist, create it. As we have seen, one way to do this is to pass
the T.id
and all properties directly to mergeV()
. Another is to use Merge.onCreate
. Note
that the Map
specified for Match.onCreate
does not need to include the T.id
already present
in the original search. The values provided to the mergeV()
Map
are inherited by the onCreate
action and combined with the Map
provided to Merge.onCreate
. Overrides of the T.id
or T.label
in the onCreate Map
are prohibited.
g.mergeV([(T.id):300]). option(Merge.onCreate,[(T.label):'Dog', name:'Toby', age:10])
To take specific action when the vertex already exists, Merge.onMatch
can be used. The
second parameter to the option
step can be either a Map
whose values are used to update
the vertex or another Gremlin traversal that generates a Map
.
Note
|
If mergeV() is given an empty Map ; such as mergeV([:]) , it will match, and
return, every vertex in the graph. This is the same behavior seen with V([]) .
|
g.mergeV([(T.id):300]). option(Merge.onCreate,[(T.label):'Dog', name:'Toby', age:10]). (1) option(Merge.onMatch,[age:11]) (2) g.withSideEffect('new-data',[age:11]). mergeV([(T.id):300]). option(Merge.onCreate,[(T.label):'Dog', name:'Toby', age:10]). option(Merge.onMatch,select('new-data')) (3) g.V(300).valueMap().with(WithOptions.tokens)
-
If no match found create the vertex using these values.
-
If a match is found, change the age property value.
-
Change the age property by selecting from the
new-data
map.
It is sometimes helpful to incorporate fail()
step into scenarios where there is a need to stop the traversal
for one event or the other:
gremlin> g.mergeV([(T.id): 1]).
......1> option(onCreate, fail("vertex did not exist")).
......2> option(onMatch, [modified: 2022])
fail() Step Triggered
======================================================================================================================================================================
Message > vertex did not exist
Traverser> false
Bulk > 1
Traversal> fail("vertex did not exist")
Parent > TinkerMergeVertexStep [mergeV([(T.id):((int) 1)]).option(Merge.onCreate,__.fail("vertex did not exist")).option(Merge.onMatch,[("modified"):((int) 2022)])]
Metadata > {}
======================================================================================================================================================================
When working with multi-properties, there are two ways to specify them for mergeV()
. First, you can specify them
individually using a CardinalityValue
as the value in the Map
. The CardinalityValue
allows you to specify the
value as well as the Cardinality
for that value. Note that it is only possible to specify one value with this syntax
even if you are using set
or list
.
g.mergeV([(T.label):'Dog', name:'Max']). (1) option(onCreate, [alias: set('Maximus')]). (2) property(set,'alias','Maxamillion') (3) g.V().has('name','Max').valueMap().with(WithOptions.tokens)
-
Find or create a vertex for Max.
-
If Max is not found then add an alias of
set
cardinality. -
Whether Max was found or created, add another alias with
set
cardinality.
The second option is to specify Cardinality
for the entire range of values as follows:
g.mergeV([(T.label):'Dog', name:'Max']). option(onCreate, [alias: 'Maximus', city: 'Boston'], set) (1) g.mergeV([(T.label):'Dog', name:'Max']). option(onCreate, [alias: 'Maximus', city: single('Boston')], set) (2)
-
If Max is created then set the alias and city with cardinality of
set
. -
If Max is created then set the alias with cardinality of
set
and city with cardinalitysingle
.
More than one vertex can be created by a single mergeV()
operation. This is done by
injecting a List
of Map
objects into the traversal and letting them stream into the mergeV()
step.
maps = [[(T.label) : 'Dog', name: 'Toby' , breed: 'Golden Retriever'], [(T.label) : 'Dog', name: 'Brandy', breed: 'Golden Retriever'], [(T.label) : 'Dog', name: 'Scamp' , breed: 'King Charles Spaniel'], [(T.label) : 'Dog', name: 'Shadow', breed: 'Mixed'], [(T.label) : 'Dog', name: 'Rocket', breed: 'Golden Retriever'], [(T.label) : 'Dog', name: 'Dax' , breed: 'Mixed'], [(T.label) : 'Dog', name: 'Baxter', breed: 'Mixed'], [(T.label) : 'Dog', name: 'Zoe' , breed: 'Corgi'], [(T.label) : 'Dog', name: 'Pixel' , breed: 'Mixed']] g.inject(maps).unfold().mergeV() g.V().hasLabel('Dog').valueMap().with(WithOptions.tokens)
Another useful pattern that can be used with mergeV()
involves putting multiple maps in a
list and selecting different maps based on the action being taken. The examples below use
a list containing three maps. The first containing just the ID to be searched for. The
second map contains all the information to use when the vertex is created. The third map
contains additional information that will be applied if an existing vertex is found.
g.inject([[(T.id):400],[(T.label):'Dog',name:'Pixel',age:1],[updated:'2022-02-1']]).as('m'). mergeV(select('m').limit(local,1)). (1) option(Merge.onCreate, select('m').range(local,1,2)). (2) option(Merge.onMatch, select('m').tail(local)) (3) g.V(400).valueMap().with(WithOptions.tokens) g.inject([[(T.id):400],[(T.label):'Dog',name:'Pixel',age:1],[updated:'2022-02-1']]).as('m'). mergeV(select('m').limit(local,1)). option(Merge.onCreate, select('m').range(local,1,2)). option(Merge.onMatch, select('m').tail(local)) (4) g.V(400).valueMap().with(WithOptions.tokens) (5)
-
Use the first map to search for a vertex with an ID of 400.
-
If the vertex was not found, use the second map to create it.
-
If the vertex was found, add an
updated
property. -
Pixel exists now, so we will take this option.
-
The
updated
property has now been added.
Additional References
The min()
-step (map) operates on a stream of comparable objects and determines which is the first object according
to its natural order in the stream.
g.V().values('age').min() g.V().repeat(both()).times(3).values('age').min() g.V().values('name').min()
When called as min(local)
it determines the minimum value of the current, local object (not the objects in the
traversal stream). This works for Collection
and Comparable
-type objects.
g.V().values('age').fold().min(local)
When there are null
values being evaluated the null
objects are ignored, but if all values are recognized as null
the return value is null
.
g.inject(null,10, 9, null).min() g.inject([null,null,null]).min(local)
Additional References
It is possible to filter list traversers using none()
-step (filter). Every item in the list will be tested against
the supplied predicate and if none of the items pass then the traverser is passed along the stream, otherwise it is
filtered. Empty lists are passed along but null or non-iterable traversers are filtered out.
Note
|
Prior to release 4.0.0, none() was a traversal discarding step primarily used by iterate() . This step has
since been renamed to discard()
|
g.V().values('age').fold().none(gt(25)) (1)
-
Return the list of ages only if no one’s age is greater than 25.
Additional References
The not()
-step (filter) removes objects from the traversal stream when the traversal provided as an argument
returns an object.
Groovy
|
The term |
Python
|
The term |
g.V().not(hasLabel('person')).elementMap() g.V().hasLabel('person'). not(out('created').count().is(gt(1))).values('name') (1)
-
josh created two projects and vadas none
Additional References
Additional References
The optional()
-step (branch/flatMap) returns the result of the specified traversal if it yields a result else it returns the calling
element, i.e. the identity()
.
g.V(2).optional(out('knows')) (1) g.V(2).optional(__.in('knows')) (2)
-
vadas does not have an outgoing knows-edge so vadas is returned.
-
vadas does have an incoming knows-edge so marko is returned.
optional
is particularly useful for lifting entire graphs when used in conjunction with path
or tree
.
g.V().hasLabel('person').optional(out('knows').optional(out('created'))).path() (1)
-
Returns the paths of everybody followed by who they know followed by what they created.
Additional References
The or()
-step ensures that at least one of the provided traversals yield a result (filter). Please see
and()
for and-semantics.
Python
|
The term |
g.V().or( __.outE('created'), __.inE('created').count().is(gt(1))). values('name')
The or()
-step can take an arbitrary number of traversals. At least one of the traversals must produce at least one
output for the original traverser to pass to the next step.
An infix notation can be used as well.
g.V().where(outE('created').or().outE('knows')).values('name')
Additional References
When the objects of the traversal stream need to be sorted, order()
-step (map) can be leveraged.
g.V().values('name').order() g.V().values('name').order().by(desc) g.V().hasLabel('person').order().by('age', asc).values('name')
One of the most traversed objects in a traversal is an Element
. An element can have properties associated with it
(i.e. key/value pairs). In many situations, it is desirable to sort an element traversal stream according to a
comparison of their properties.
g.V().values('name') g.V().order().by('name',asc).values('name') g.V().order().by('name',desc).values('name') g.V().both().order().by('age') (1)
-
The "age" property is not productive for all vertices and therefore those values are filtered.
The order()
-step allows the user to provide an arbitrary number of comparators for primary, secondary, etc. sorting.
In the example below, the primary ordering is based on the outgoing created-edge count. The secondary ordering is
based on the age of the person.
g.V().hasLabel('person').order().by(outE('created').count(), asc). by('age', asc).values('name') g.V().hasLabel('person').order().by(outE('created').count(), asc). by('age', desc).values('name')
Randomizing the order of the traversers at a particular point in the traversal is possible with Order.shuffle
.
g.V().hasLabel('person').order().by(shuffle) g.V().hasLabel('person').order().by(shuffle)
It is possible to use order(local)
to order the current local object and not the entire traversal stream. This works for
Collection
- and Map
-type objects. For any other object, the object is returned unchanged.
g.V().values('age').fold().order(local).by(desc) (1) g.V().values('age').order(local).by(desc) (2) g.V().groupCount().by(inE().count()).order(local).by(values, desc) (3) g.V().groupCount().by(inE().count()).order(local).by(keys, asc) (4)
-
The ages are gathered into a list and then that list is sorted in decreasing order.
-
The ages are not gathered and thus
order(local)
is "ordering" single integers and thus, does nothing. -
The
groupCount()
map is ordered by its values in decreasing order. -
The
groupCount()
map is ordered by its keys in increasing order.
Note
|
The values and keys enums are from Column which is used to select "columns" from a Map , Map.Entry , or Path .
|
If a property key does not exist, then it will be treated as null
which will sort it first for Order.asc
and last
for Order.desc
.
g.V().order().by("age").elementMap()
Note
|
Prior to version 3.3.4, ordering was defined by Order.incr for ascending order and Order.decr for descending
order. Those tokens were deprecated and eventually removed in 3.5.0.
|
Additional References
The pageRank()
-step (map/sideEffect) calculates PageRank using
PageRankVertexProgram
.
Important
|
The pageRank() -step is a VertexComputing -step and as such, can only be used against a graph that
supports GraphComputer (OLAP).
|
g = traversal().with(graph).withComputer() g.V().pageRank().with(PageRank.propertyName, 'friendRank').values('pageRank') g.V().hasLabel('person'). pageRank(). with(PageRank.edges, __.outE('knows')). with(PageRank.propertyName, 'friendRank'). order().by('friendRank',desc). elementMap('name','friendRank')
Note the use of the with()
modulating step which provides configuration options to the algorithm. It takes
configuration keys from the PageRank
and is automatically imported to the Gremlin Console.
The explain()
-step can be used to understand how the traversal is compiled into multiple
GraphComputer
jobs.
g = traversal().with(graph).withComputer() g.V().hasLabel('person'). pageRank(). with(PageRank.edges, __.outE('knows')). with(PageRank.propertyName, 'friendRank'). order().by('friendRank',desc). elementMap('name','friendRank').explain()
Additional References
A traverser is transformed as it moves through a series of steps within a traversal. The history of the traverser is
realized by examining its path with path()
-step (map).
g.V().out().out().values('name') g.V().out().out().values('name').path() g.V().both().path().by('age') (1)
-
The "age" property is not productive for all vertices and therefore those values are filtered.
If edges are required in the path, then be sure to traverse those edges explicitly.
g.V().outE().inV().outE().inV().path()
It is possible to post-process the elements of the path in a round-robin fashion via by()
.
g.V().out().out().path().by('name').by('age')
Finally, because by()
-based post-processing, nothing prevents triggering yet another traversal. In the traversal
below, for each element of the path traversed thus far, if its a person (as determined by having an age
-property),
then get all of their creations, else if its a creation, get all the people that created it.
g.V().out().out().path().by( choose(hasLabel('person'), out('created').values('name'), __.in('created').values('name')).fold())
g.V().has('person','name','vadas').as('e'). in('knows'). out('knows').where(neq('e')). path().by('name') (1) g.V().has('person','name','vadas').as('e'). in('knows').as('m'). out('knows').where(neq('e')). path().to('m').by('name') (2) g.V().has('person','name','vadas').as('e'). in('knows').as('m'). out('knows').where(neq('e')). path().from('m').by('name') (3)
-
Obtain the full path from vadas to josh.
-
Save the middle node, marko, and use the
to()
modulator to show only the path from vadas to marko -
Use the
from()
mdoulator to show only the path from marko to josh
Warning
|
Generating path information is expensive as the history of the traverser is stored into a Java list. With
numerous traversers, there are numerous lists. Moreover, in an OLAP GraphComputer environment
this becomes exceedingly prohibitive as there are traversers emanating from all vertices in the graph in parallel.
In OLAP there are optimizations provided for traverser populations, but when paths are calculated (and each traverser
is unique due to its history), then these optimizations are no longer possible.
|
The Path
data structure is an ordered list of objects, where each object is associated to a Set<String>
of
labels. An example is presented below to demonstrate both the Path
API as well as how a traversal yields labeled paths.
path = g.V(1).as('a').has('name').as('b'). out('knows').out('created').as('c'). has('name','ripple').values('name').as('d'). identity().as('e').path().next() path.size() path.objects() path.labels() path.a path.b path.c path.d == path.e
Additional References
The peerPressure()
-step (map/sideEffect) clusters vertices using PeerPressureVertexProgram
.
Important
|
The peerPressure() -step is a VertexComputing -step and as such, can only be used against a graph that supports GraphComputer (OLAP).
|
g = traversal().with(graph).withComputer() g.V().peerPressure().with(PeerPressure.propertyName, 'cluster').values('cluster') g.V().hasLabel('person'). peerPressure(). with(PeerPressure.propertyName, 'cluster'). group(). by('cluster'). by('name')
Note the use of the with()
modulating step which provides configuration options to the algorithm. It takes
configuration keys from the PeerPressure
class and is automatically imported to the Gremlin Console.
Additional References
The product()
-step (map) calculates the cartesian product between the incoming list traverser and the provided list
argument. This step only expects list data (array or Iterable) and will throw an IllegalArgumentException
if any
other type is encountered (including null
).
g.V().values("name").fold().product(["james","jen"]) g.V().values("name").fold().product(__.V().has("age").limit(1).values("age").fold())
Additional References
The profile()
-step (sideEffect) exists to allow developers to profile their traversals to determine statistical
information like step runtime, counts, etc.
Warning
|
Profiling a Traversal will impede the Traversal’s performance. This overhead is mostly excluded from the profile results, but durations are not exact. Thus, durations are best considered in relation to each other. |
g.V().out('created').repeat(both()).times(3).hasLabel('person').values('age').sum().profile()
The profile()
-step generates a TraversalMetrics
sideEffect object that contains the following information:
-
Step
: A step within the traversal being profiled. -
Count
: The number of represented traversers that passed through the step. -
Traversers
: The number of traversers that passed through the step. -
Time (ms)
: The total time the step was actively executing its behavior. -
% Dur
: The percentage of total time spent in the step.
It is important to understand the difference between "Count"
and "Traversers". Traversers can be merged and as such, when two traversers are "the same" they may be aggregated
into a single traverser. That new traverser has a
Traverser.bulk()
that is the sum of the two merged traverser
bulks. On the other hand, the Count
represents the sum of all Traverser.bulk()
results and thus, expresses the
number of "represented" (not enumerated) traversers. Traversers
will always be less than or equal to Count
.
For traversal compilation information, please see explain()
-step.
Additional References
The project()
-step (map) projects the current object into a Map<String,Object>
keyed by provided labels. It is similar
to select()
-step, save that instead of retrieving and modulating historic traverser state, it modulates
the current state of the traverser.
g.V().has('name','marko'). project('id', 'name', 'out', 'in'). by(id). by('name'). by(outE().count()). by(inE().count()) g.V().has('name','marko'). project('name', 'friendsNames'). by('name'). by(out('knows').values('name').fold()) g.V().out('created'). project('a','b'). by('name'). by(__.in('created').count()). order().by(select('b'),desc). select('a') g.V().project('n','a').by('name').by('age') (1)
-
The "age" property is not productive for all vertices and therefore those values are filtered and the key not present in the
Map
.
Additional References
The program()
-step (map/sideEffect) is the "lambda" step for GraphComputer
jobs. The step takes a
VertexProgram
as an argument and will process the incoming graph accordingly. Thus, the user
can create their own VertexProgram
and have it execute within a traversal. The configuration provided to the
vertex program includes:
-
gremlin.vertexProgramStep.rootTraversal
is a serialization of aPureTraversal
form of the root traversal. -
gremlin.vertexProgramStep.stepId
is the step string id of theprogram()
-step being executed.
The user supplied VertexProgram
can leverage that information accordingly within their vertex program. Example uses
are provided below.
Warning
|
Developing a VertexProgram is for expert users. Moreover, developing one that can be used effectively within
a traversal requires yet more expertise. This information is recommended to advanced users with a deep understanding of the
mechanics of Gremlin OLAP (GraphComputer ).
|
private TraverserSet<Object> haltedTraversers;
public void loadState(Graph graph, Configuration configuration) {
VertexProgram.super.loadState(graph, configuration);
this.traversal = PureTraversal.loadState(configuration, VertexProgramStep.ROOT_TRAVERSAL, graph);
this.programStep = new TraversalMatrix<>(this.traversal.get()).getStepById(configuration.getString(ProgramVertexProgramStep.STEP_ID));
// if the traversal sideEffects will be used in the computation, add them as memory compute keys
this.memoryComputeKeys.addAll(MemoryTraversalSideEffects.getMemoryComputeKeys(this.traversal.get()));
// if master-traversal traversers may be propagated, create a memory compute key
this.memoryComputeKeys.add(MemoryComputeKey.of(TraversalVertexProgram.HALTED_TRAVERSERS, Operator.addAll, false, false));
// returns an empty traverser set if there are no halted traversers
this.haltedTraversers = TraversalVertexProgram.loadHaltedTraversers(configuration);
}
public void storeState(Configuration configuration) {
VertexProgram.super.storeState(configuration);
// if halted traversers is null or empty, it does nothing
TraversalVertexProgram.storeHaltedTraversers(configuration, this.haltedTraversers);
}
public void setup(Memory memory) {
if(!this.haltedTraversers.isEmpty()) {
// do what you like with the halted master traversal traversers
}
// once used, no need to keep that information around (master)
this.haltedTraversers = null;
}
public void execute(Vertex vertex, Messenger messenger, Memory memory) {
// once used, no need to keep that information around (workers)
if(null != this.haltedTraversers)
this.haltedTraversers = null;
if(vertex.property(TraversalVertexProgram.HALTED_TRAVERSERS).isPresent()) {
// haltedTraversers in execute() represent worker-traversal traversers
// for example, from a traversal of the form g.V().out().program(...)
TraverserSet<Object> haltedTraversers = vertex.value(TraversalVertexProgram.HALTED_TRAVERSERS);
// create a new halted traverser set that can be used by the next OLAP job in the chain
// these are worker-traversers that are distributed throughout the graph
TraverserSet<Object> newHaltedTraversers = new TraverserSet<>();
haltedTraversers.forEach(traverser -> {
newHaltedTraversers.add(traverser.split(traverser.get().toString(), this.programStep));
});
vertex.property(VertexProperty.Cardinality.single, TraversalVertexProgram.HALTED_TRAVERSERS, newHaltedTraversers);
// it is possible to create master-traversers that are localized to the master traversal (this is how results are ultimately delivered back to the user)
memory.add(TraversalVertexProgram.HALTED_TRAVERSERS,
new TraverserSet<>(this.traversal().get().getTraverserGenerator().generate("an example", this.programStep, 1l)));
}
public boolean terminate(Memory memory) {
// the master-traversal will have halted traversers
assert memory.exists(TraversalVertexProgram.HALTED_TRAVERSERS);
TraverserSet<String> haltedTraversers = memory.get(TraversalVertexProgram.HALTED_TRAVERSERS);
// it will only have the traversers sent to the master traversal via memory
assert haltedTraversers.stream().map(Traverser::get).filter(s -> s.equals("an example")).findAny().isPresent();
// it will not contain the worker traversers distributed throughout the vertices
assert !haltedTraversers.stream().map(Traverser::get).filter(s -> !s.equals("an example")).findAny().isPresent();
return true;
}
Note
|
The test case ProgramTest in gremlin-test has an example vertex program called TestProgram that demonstrates
all the various ways in which traversal and traverser information is propagated within a vertex program and ultimately
usable by other vertex programs (including TraversalVertexProgram ) down the line in an OLAP compute chain.
|
Finally, an example is provided using PageRankVertexProgram
which doesn’t use pageRank()
-step.
g = traversal().with(graph).withComputer() g.V().hasLabel('person'). program(PageRankVertexProgram.build().property('rank').create(graph)). order().by('rank', asc). elementMap('name', 'rank')
The properties()
-step (map) extracts properties from an Element
in the traversal stream.
g.V(1).properties() g.V(1).properties('location').valueMap() g.V(1).properties('location').has('endTime').valueMap()
Additional References
The property()
-step is used to add properties to the elements of the graph (sideEffect). Unlike addV()
and
addE()
, property()
is a full sideEffect step in that it does not return the property it created, but the element
that streamed into it. Moreover, if property()
follows an addV()
or addE()
, then it is "folded" into the
previous step to enable vertex and edge creation with all its properties in one creation operation.
g.V(1).property('country','usa') g.V(1).property('city','santa fe').property('state','new mexico').valueMap() g.V(1).property(['city': 'santa fe', 'state': 'new mexico']) (1) g.V(1).property(list,'age',35) (2) g.V(1).property(list, ['city': 'santa fe', 'state': 'new mexico']) (3) g.V(1).valueMap() g.V(1).property(list, ['age': single(36), 'city': 'wilmington', 'state': 'delaware']) (4) g.V(1).valueMap() g.V(1).property('friendWeight',outE('knows').values('weight').sum(),'acl','private') (5) g.V(1).properties('friendWeight').valueMap() (6) g.addV().property(T.label,'person').valueMap().with(WithOptions.tokens) (7) g.addV().property(null) (8) g.addV().property(set, null)
-
Properties can also take a
Map
as an argument. -
For vertices, a cardinality can be provided for vertex properties.
-
If a cardinality is specified for a
Map
then that cardinality will be used for all properties in the map. -
Assign the
Cardinality
individually to override the specifiedlist
or the default cardinality if not specified. -
It is possible to select the property value (as well as key) via a traversal.
-
For vertices, the
property()
-step can add meta-properties. -
The label value can be specified as a property only at the time a vertex is added and if one is not specified in the addV()
-
If you pass a
null
value for the Map this will be treated as a no-op and the input will be returned
Additional References
The propertiesMap()
-step yields a Map representation of the properties of an element.
g.V().propertyMap() g.V().propertyMap('age') g.V().propertyMap('age','blah') g.E().propertyMap()
Additional References
As traversers propagate through the traversal, it is possible to only allow a certain number of them to pass through
with range()
-step (filter). When the low-end of the range is not met, objects are continued to be iterated. When
within the low (inclusive) and high (exclusive) range, traversers are emitted. When above the high range, the traversal
breaks out of iteration. Finally, the use of -1
on the high range will emit remaining traversers after the low range
begins.
g.V().range(0,3) g.V().range(1,3) g.V().range(1, -1) g.V().repeat(both()).times(1000000).emit().range(6,10)
The range()
-step can also be applied with Scope.local
, in which case it operates on the incoming collection.
For example, it is possible to produce a Map<String, String>
for each traversed path, but containing only the second
property value (the "b" step).
g.V().as('a').out().as('b').in().as('c').select('a','b','c').by('name').range(local,1,2)
The next example uses the The Crew toy data set. It produces a List<String>
containing the
second and third location for each vertex.
g.V().valueMap().select('location').range(local, 1, 3)
Additional References
The read()
-step is not really a "step" but a step modulator in that it modifies the functionality of the io()
-step.
More specifically, it tells the io()
-step that it is expected to use its configuration to read data from some
location. Please see the documentation for io()
-step for more complete details on usage.
Additional References
The repeat()
-step (branch) is used for looping over a traversal given some break predicate. Below are some
examples of repeat()
-step in action.
g.V(1).repeat(out()).times(2).path().by('name') (1) g.V().until(has('name','ripple')). repeat(out()).path().by('name') (2)
-
do-while semantics stating to do
out()
2 times. -
while-do semantics stating to break if the traverser is at a vertex named "ripple".
Important
|
There are two modulators for repeat() : until() and emit() . If until() comes after repeat() it is
do/while looping. If until() comes before repeat() it is while/do looping. If emit() is placed after repeat() ,
it is evaluated on the traversers leaving the repeat-traversal. If emit() is placed before repeat() , it is
evaluated on the traversers prior to entering the repeat-traversal.
|
The repeat()
-step also supports an "emit predicate", where the predicate for an empty argument emit()
is
true
(i.e. emit() == emit{true}
). With emit()
, the traverser is split in two — the traverser exits the code
block as well as continues back within the code block (assuming until()
holds true).
g.V(1).repeat(out()).times(2).emit().path().by('name') (1) g.V(1).emit().repeat(out()).times(2).path().by('name') (2)
-
The
emit()
comes afterrepeat()
and thus, emission happens after therepeat()
traversal is executed. Thus, no one vertex paths exist. -
The
emit()
comes beforerepeat()
and thus, emission happens prior to therepeat()
traversal being executed. Thus, one vertex paths exist.
The emit()
-modulator can take an arbitrary predicate.
g.V(1).repeat(out()).times(2).emit(has('lang')).path().by('name')
g.V(1).repeat(out()).times(2).emit().path().by('name')
The first time through the repeat()
, the vertices lop, vadas, and josh are seen. Given that loops==1
, the
traverser repeats. However, because the emit-predicate is declared true, those vertices are emitted. The next time through
repeat()
, the vertices traversed are ripple and lop (Josh’s created projects, as lop and vadas have no out edges).
Given that loops==2
, the until-predicate fails and ripple and lop are emitted.
Therefore, the traverser has seen the vertices: lop, vadas, josh, ripple, and lop.
repeat()
-steps may be nested inside each other or inside the emit()
or until()
predicates and they can also be 'named' by passing a string as the first parameter to repeat()
. The loop counter of a named repeat step can be accessed within the looped context with loops(loopName)
where loopName
is the name set whe creating the repeat()
-step.
g.V(1). repeat(out("knows")). until(repeat(out("created")).emit(has("name", "lop"))) (1) g.V(6). repeat('a', both('created').simplePath()). emit(repeat('b', both('knows')). until(loops('b').as('b').where(loops('a').as('b'))). hasId(2)).dedup() (2)
-
Starting from vertex 1, keep going taking outgoing 'knows' edges until the vertex was created by 'lop'.
-
Starting from vertex 6, keep taking created edges in either direction until the vertex is same distance from vertex 2 over knows edges as it is from vertex 6 over created edges.
Finally, note that both emit()
and until()
can take a traversal and in such, situations, the predicate is
determined by traversal.hasNext()
. A few examples are provided below.
g.V(1).repeat(out()).until(hasLabel('software')).path().by('name') (1) g.V(1).emit(hasLabel('person')).repeat(out()).path().by('name') (2) g.V(1).repeat(out()).until(outE().count().is(0)).path().by('name') (3)
-
Starting from vertex 1, keep taking outgoing edges until a software vertex is reached.
-
Starting from vertex 1, and in an infinite loop, emit the vertex if it is a person and then traverser the outgoing edges.
-
Starting from vertex 1, keep taking outgoing edges until a vertex is reached that has no more outgoing edges.
Warning
|
The anonymous traversal of emit() and until() (not repeat() ) process their current objects "locally."
In OLAP, where the atomic unit of computing is the vertex and its local "star graph," it is important that the
anonymous traversals do not leave the confines of the vertex’s star graph. In other words, they can not traverse to
an adjacent vertex’s properties or edges.
|
Additional References
The replace()
-step (map) returns a string with the specified characters in the original string replaced with the new
characters. Any null arguments will be a no-op and the original string is returned. Null values from the incoming
traversers are not processed and remain as null when returned. If the incoming traverser is a non-String value then
an IllegalArgumentException
will be thrown.
g.inject('that', 'this', 'test', null).replace('h', 'j') (1) g.inject('hello world').replace(null, 'j') (2) g.V().hasLabel("software").values("name").replace("p", "g") (3) g.V().hasLabel("software").values("name").fold().replace(local, "p", "g") (4)
-
Replace "h" in the strings with "j".
-
Null inputs are ignored and the original string is returned.
-
Return software names with "p" replaced by "g".
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
Additional References
replace(String,String)
replace(Scope,String,String)
The reverse()
-step (map) returns the reverse of the incoming list traverser. Single values (including null
) are not
processed and are added back to the Traversal Stream unchanged. If the incoming traverser is a String value then the
reversed String will be returned.
g.V().values("name").reverse() (1) g.V().values("name").order().fold().reverse() (2)
-
Reverse the order of the characters in each name.
-
Fold all the names into a list in ascending order and then reverse the list’s ordering (into descending).
The rTrim()
-step (map) returns a string with trailing whitespace removed. Null values are not processed and remain
as null when returned. If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.inject(" hello ", " world ", null).rTrim() g.inject([" hello ", " world ", null]).rTrim(local) (1)
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
A traverser can contain a local data structure called a "sack".
The
sack()
-step is used to read and write sacks (sideEffect or map). Each sack of each traverser is created
when using GraphTraversal.withSack(initialValueSupplier,splitOperator?,mergeOperator?)
.
-
Initial value supplier: A
Supplier
providing the initial value of each traverser’s sack. -
Split operator: a
UnaryOperator
that clones the traverser’s sack when the traverser splits. If no split operator is provided, thenUnaryOperator.identity()
is assumed. -
Merge operator: A
BinaryOperator
that unites two traverser’s sack when they are merged. If no merge operator is provided, then traversers with sacks can not be merged.
Two trivial examples are presented below to demonstrate the initial value supplier. In the first example below, a
traverser is created at each vertex in the graph (g.V()
), with a 1.0 sack (withSack(1.0f)
), and then the sack
value is accessed (sack()
). In the second example, a random float supplier is used to generate sack values.
g.withSack(1.0f).V().sack() rand = new Random() g.withSack {rand.nextFloat()}.V().sack()
A more complicated initial value supplier example is presented below where the sack values are used in a running
computation and then emitted at the end of the traversal. When an edge is traversed, the edge weight is multiplied
by the sack value (sack(mult).by('weight')
). Note that the by()
-modulator can be any arbitrary traversal.
g.withSack(1.0f).V().repeat(outE().sack(mult).by('weight').inV()).times(2) g.withSack(1.0f).V().repeat(outE().sack(mult).by('weight').inV()).times(2).sack() g.withSack(1.0f).V().repeat(outE().sack(mult).by('weight').inV()).times(2).path(). by().by('weight') g.V().sack(assign).by('age').sack() (1)
-
The "age" property is not productive for all vertices and therefore those values are filtered during the assignment.
When complex objects are used (i.e. non-primitives), then a
split operator should be defined to ensure that each traverser gets a clone of its parent’s sack. The first example
does not use a split operator and as such, the same map is propagated to all traversers (a global data structure). The
second example, demonstrates how
Map.clone()
ensures that each traverser’s sack contains a unique, local sack.
g.withSack {[:]}.V().out().out(). sack {m,v -> m[v.value('name')] = v.value('lang'); m}.sack() // BAD: single map g.withSack {[:]}{it.clone()}.V().out().out(). sack {m,v -> m[v.value('name')] = v.value('lang'); m}.sack() // GOOD: cloned map
Note
|
For primitives (i.e. integers, longs, floats, etc.), a split operator is not required as a primitives are encoded in the memory address of the sack, not as a reference to an object. |
If a merge operator is not provided, then traversers with sacks can not be bulked. However, in many situations,
merging the sacks of two traversers at the same location is algorithmically sound and good to provide so as to gain
the bulking optimization. In the examples below, the binary merge operator is Operator.sum
. Thus, when two traverser
merge, their respective sacks are added together.
g.withSack(1.0d).V(1).out('knows').in('knows') (1) g.withSack(1.0d).V(1).out('knows').in('knows').sack() (2) g.withSack(1.0d, sum).V(1).out('knows').in('knows').sack() (3) g.withSack(1.0d).V(1).local(outE('knows').barrier(normSack).inV()).in('knows').barrier() (4) g.withSack(1.0d).V(1).local(outE('knows').barrier(normSack).inV()).in('knows').barrier().sack() (5) g.withSack(1.0d,sum).V(1).local(outE('knows').barrier(normSack).inV()).in('knows').barrier().sack() (6) g.withBulk(false).withSack(1.0f,sum).V(1).local(outE('knows').barrier(normSack).inV()).in('knows').barrier().sack() (7) g.withBulk(false).withSack(1.0f).V(1).local(outE('knows').barrier(normSack).inV()).in('knows').barrier().sack()(8)
-
We find vertex 1 twice because he knows two other people
-
Without a merge operation the sack values are 1.0.
-
When specifying
sum
as the merge operation, the sack values are 2.0 because of bulking -
Like 1, but using barrier internally
-
The
local(…barrier(normSack)…)
ensures that all traversers leaving vertex 1 have an evenly distributed amount of the initial 1.0 "energy" (50-50), i.e. the sack is 0.5 on each result -
Like 3, but using
sum
as merge operator leads to the expected 1.0 -
There is now a single traverser with bulk of 2 and sack of 1.0 and thus, setting
withBulk(false)`
yields the expected 1.0 -
Like 7, but without the
sum
operator
Additional References
The sample()
-step is useful for sampling some number of traversers previous in the traversal.
g.V().outE().sample(1).values('weight') g.V().outE().sample(1).by('weight').values('weight') g.V().outE().sample(2).by('weight').values('weight') g.V().both().sample(2).by('age') (1)
-
The "age" property is not productive for all vertices and therefore those values are not considered when sampling.
One of the more interesting use cases for sample()
is when it is used in conjunction with local()
.
The combination of the two steps supports the execution of random walks.
In the example below, the traversal starts are vertex 1 and selects one edge to traverse based on a probability
distribution generated by the weights of the edges. The output is always a single path as by selecting a single edge,
the traverser never splits and continues down a single path in the graph.
g.V(1). repeat(local(bothE().sample(1).by('weight').otherV())). times(5) g.V(1). repeat(local(bothE().sample(1).by('weight').otherV())). times(5). path() g.V(1). repeat(local(bothE().sample(1).by('weight').otherV())). times(10). path()
As a clarification, note that in the above example local()
is not strictly required as it only does the random walk
over a single vertex, but note what happens without it if multiple vertices are traversed:
g.V().repeat(bothE().sample(1).by('weight').otherV()).times(5).path()
The use of local()
ensures that the traversal over bothE()
occurs once per vertex traverser that passes through,
thus allowing one random walk per vertex.
g.V().repeat(local(bothE().sample(1).by('weight').otherV())).times(5).path()
So, while not strictly required, it is likely better to be explicit with the use of local()
so that the proper intent
of the traversal is expressed.
Additional References
Functional languages make use of function composition and
lazy evaluation to create complex computations from primitive operations. This is exactly what Traversal
does. One
of the differentiating aspects of Gremlin’s data flow approach to graph processing is that the flow need not always go
"forward," but in fact, can go back to a previously seen area of computation. Examples include path()
as well as the select()
-step (map). There are two general ways to use select()
-step.
-
Select labeled steps within a path (as defined by
as()
in a traversal). -
Select objects out of a
Map<String,Object>
flow (i.e. a sub-map).
The first use case is demonstrated via example below.
g.V().as('a').out().as('b').out().as('c') // no select g.V().as('a').out().as('b').out().as('c').select('a','b','c') g.V().as('a').out().as('b').out().as('c').select('a','b') g.V().as('a').out().as('b').out().as('c').select('a','b').by('name') g.V().as('a').out().as('b').out().as('c').select('a') (1) g.V(1).as('a').both().as('b').select('a','b').by('age')
-
If the selection is one step, no map is returned.
-
The "age" property is not productive for all vertices and therefore those values are filtered.
When there is only one label selected, then a single object is returned. This is useful for stepping back in a computation and easily moving forward again on the object reverted to.
g.V().out().out() g.V().out().out().path() g.V().as('x').out().out().select('x') g.V().out().as('x').out().select('x') g.V().out().out().as('x').select('x') // pointless
Note
|
When executing a traversal with select() on a standard traversal engine (i.e. OLTP), select() will do its
best to avoid calculating the path history and instead, will rely on a global data structure for storing the currently
selected object. As such, if only a subset of the path walked is required, select() should be used over the more
resource intensive path() -step.
|
When the set of keys or values (i.e. columns) of a path or map are needed, use select(keys)
and select(values)
,
respectively. This is especially useful when one is only interested in the top N elements in a groupCount()
ranking.
g = traversal().with(graph) g.io('data/grateful-dead.xml').read().iterate() g.V().hasLabel('song').out('followedBy').groupCount().by('name'). order(local).by(values,desc).limit(local, 5) g.V().hasLabel('song').out('followedBy').groupCount().by('name'). order(local).by(values,desc).limit(local, 5).select(keys) g.V().hasLabel('song').out('followedBy').groupCount().by('name'). order(local).by(values,desc).limit(local, 5).select(keys).unfold()
Similarly, for extracting the values from a path or map.
g = traversal().with(graph) g.io('data/grateful-dead.xml').read().iterate() g.V().hasLabel('song').out('sungBy').groupCount().by('name') (1) g.V().hasLabel('song').out('sungBy').groupCount().by('name').select(values) (2) g.V().hasLabel('song').out('sungBy').groupCount().by('name').select(values).unfold(). groupCount().order(local).by(values,desc).limit(local, 5) (3)
-
Which artist sung how many songs?
-
Get an anonymized set of song repertoire sizes.
-
What are the 5 most common song repertoire sizes?
Warning
|
Note that by() -modulation is not supported with select(keys) and select(values) .
|
There is also an option to supply a Pop
operation to select()
to manipulate List
objects in the Traverser
:
g.V(1).as("a").repeat(out().as("a")).times(2).select(first, "a") g.V(1).as("a").repeat(out().as("a")).times(2).select(last, "a") g.V(1).as("a").repeat(out().as("a")).times(2).select(all, "a")
In addition to the previously shown examples, where select()
was used to select an element based on a static key, select()
can also accept a traversal
that emits a key.
Warning
|
Since the key used by select(<traversal>) cannot be determined at compile time, the TraversalSelectStep enables full path tracking.
|
g.withSideEffect("alias", ["marko":"okram"]).V(). (1) values("name").sack(assign). (2) optional(select("alias").select(sack())) (3)
-
Inject a name alias map and start the traversal from all vertices.
-
Select all
name
values and store them as the current traverser’s sack value. -
Optionally select the alias for the current name from the injected map.
Like match()
-step, it is possible to use where()
, as where is a filter that processes
Map<String,Object>
streams.
g.V().as('a').out('created').in('created').as('b').select('a','b').by('name') (1) g.V().as('a').out('created').in('created').as('b'). select('a','b').by('name').where('a',neq('b')) (2) g.V().as('a').out('created').in('created').as('b'). select('a','b'). (3) where('a',neq('b')). where(__.as('a').out('knows').as('b')). select('a','b').by('name')
-
A standard
select()
that generates aMap<String,Object>
of variables bindings in the path (i.e.a
andb
) for the sake of a running example. -
The
select().by('name')
projects each binding vertex to their name property value andwhere()
operates to ensure respectivea
andb
strings are not the same. -
The first
select()
projects a vertex binding set. A binding is filtered ifa
vertex equalsb
vertex. A binding is filtered ifa
doesn’t knowb
. The second and finalselect()
projects the name of the vertices.
Additional References
The shortestPath()
-step provides an easy way to find shortest non-cyclic paths in a graph. It is configurable
using the with()
-modulator with the options given below.
Important
|
The shortestPath() -step is a VertexComputing -step and as such, can only be used against a graph
that supports GraphComputer (OLAP).
|
Key | Type | Description | Default |
---|---|---|---|
|
|
Sets a filter traversal for the end vertices (e.g. |
all vertices ( |
|
|
Sets a |
|
|
|
Sets the |
|
|
|
Sets the distance limit for all shortest paths. |
none |
|
|
Whether to include edges in the result or not. |
|
g = g.withComputer() g.V().shortestPath() (1) g.V().has('person','name','marko').shortestPath() (2) g.V().shortestPath().with(ShortestPath.target, __.has('name','peter')) (3) g.V().shortestPath(). with(ShortestPath.edges, Direction.IN). with(ShortestPath.target, __.has('name','josh')) (4) g.V().has('person','name','marko'). shortestPath(). with(ShortestPath.target, __.has('name','josh')) (5) g.V().has('person','name','marko'). shortestPath(). with(ShortestPath.target, __.has('name','josh')). with(ShortestPath.distance, 'weight') (6) g.V().has('person','name','marko'). shortestPath(). with(ShortestPath.target, __.has('name','josh')). with(ShortestPath.includeEdges, true) (7)
-
Find all shortest paths.
-
Find all shortest paths from
marko
. -
Find all shortest paths to
peter
. -
Find all in-directed paths to
josh
. -
Find all shortest paths from
marko
tojosh
. -
Find all shortest paths from
marko
tojosh
using a custom distance property. -
Find all shortest paths from
marko
tojosh
and include edges in the result.
g.inject(g.withComputer().V().shortestPath(). with(ShortestPath.distance, 'weight'). with(ShortestPath.includeEdges, true). with(ShortestPath.maxDistance, 1).toList().toArray()). map(unfold().values('name','weight').fold()) (1)
-
Find all shortest paths using a custom distance property and limit the distance to 1. Inject the result into a OLTP
GraphTraversal
in order to be able to select properties from all elements in all paths.
Additional References
The sideEffect()
step performs some operation on the traverser and passes it to the next step in the process. Please
see the General Steps section for more information.
Additional References
When it is important that a traverser not repeat its path through the graph, simplePath()
-step should be used
(filter). The path information of the traverser is analyzed and if the path has repeated
objects in it, the traverser is filtered. If cyclic behavior is desired, see cyclicPath()
.
g.V(1).both().both() g.V(1).both().both().simplePath() g.V(1).both().both().simplePath().path() g.V(1).both().both().simplePath().by('age') (1) g.V().out().as('a').out().as('b').out().as('c'). simplePath().by(label). path() g.V().out().as('a').out().as('b').out().as('c'). simplePath(). by(label). from('b'). to('c'). path(). by('name')
-
The "age" property is not productive for all vertices and therefore those values are filtered.
By using the from()
and to()
modulators traversers can ensure that only certain sections of the path are acyclic.
g.addV().property(id, 'A').as('a'). addV().property(id, 'B').as('b'). addV().property(id, 'C').as('c'). addV().property(id, 'D').as('d'). addE('link').from('a').to('b'). addE('link').from('b').to('c'). addE('link').from('c').to('d').iterate() g.V('A').repeat(both().simplePath()).times(3).path() (1) g.V('D').repeat(both().simplePath()).times(3).path() (2) g.V('A').as('a'). repeat(both().simplePath().from('a')).times(3).as('b'). repeat(both().simplePath().from('b')).times(3).path() (3)
-
Traverse all acyclic 3-hop paths starting from vertex
A
-
Traverse all acyclic 3-hop paths starting from vertex
D
-
Traverse all acyclic 3-hop paths starting from vertex
A
and from there again all 3-hop paths. The second path may cross the vertices from the first path.
Additional References
The skip()
-step is analogous to range()
-step save that the higher end range is set to -1.
g.V().values('age').order() g.V().values('age').order().skip(2) g.V().values('age').order().range(2, -1)
The skip()
-step can also be applied with Scope.local
, in which case it operates on the incoming collection.
g.V().hasLabel('person').filter(outE('created')).as('p'). (1) map(out('created').values('name').fold()). project('person','primary','other'). by(select('p').by('name')). by(limit(local, 1)). (2) by(skip(local, 1)) (3)
-
For each person who created something…
-
…select the first project (random order) as
primary
and… -
…select all other projects as
other
.
Additional References
The split()
-step (map) returns a list of strings created by splitting the incoming string traverser around the
matches of the given separator. A null separator will split the string by whitespaces. An empty string separator will split on each character.
Null values from the incoming traversers are not processed and remain as null when returned. If the incoming traverser is a non-String value then an
IllegalArgumentException will be thrown.
g.inject("that", "this", "test", null).split("h") (1) g.V().hasLabel("person").values("name").split("a") (2) g.inject("helloworld", "hello world", "hello world").split(null) (3) g.inject("hello", "world", null).split("") (4) g.V().hasLabel("person").values("name").fold().split(local, "a") (5)
-
Split the strings by "h".
-
Split person names by "a".
-
Splitting by null will split by whitespaces.
-
Splitting by "" will split by each character.
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list of results.
Additional References
split(String)
split(Scope, String)
Extracting a portion of a graph from a larger one for analysis, visualization or other purposes is a fairly common
use case for graph analysts and developers. The subgraph()
-step (sideEffect) provides a way to produce an
edge-induced subgraph from virtually any traversal.
The following example demonstrates how to produce the "knows" subgraph:
subGraph = g.E().hasLabel('knows').subgraph('subGraph').cap('subGraph').next() (1) sg = traversal().with(subGraph) sg.E() (2)
-
As this function produces "edge-induced" subgraphs,
subgraph()
must be called at edge steps. -
The subgraph contains only "knows" edges.
A more common subgraphing use case is to get all of the graph structure surrounding a single vertex:
subGraph = g.V(3).repeat(__.inE().subgraph('subGraph').outV()).times(3).cap('subGraph').next() (1) sg = traversal().with(subGraph) sg.E()
-
Starting at vertex
3
, traverse 3 steps away on in-edges, outputting all of that into the subgraph.
The above example is purposely brief so as to focus on subgraph()
usage, however, it may not be the most optimal
method for constructing the subgraph. For instance, if the graph had cycles, it would attempt to reconstruct parts
of the subgraph which are already present. The duplicates would not be created, but it would involve some unnecessary
processing. If the only interest of the traversal was to populate the subgraph, it would be better to include
simplePath()
to filter out those cycles, as in .inE().subgraph('subGraph').outV().simplePath()
. From another
perspective, it might also make some sense to use dedup()
to avoid traversing the same vertices repeatedly where
two vertices shared the multiple edges between them, as in .inE().dedup().subgraph('subGraph').outV().dedup()
.
There can be multiple subgraph()
calls within the same traversal. Each operating against either the same graph
(i.e. same side-effect key) or different graphs (i.e. different side-effect keys).
t = g.V().outE('knows').subgraph('knowsG').inV().outE('created').subgraph('createdG'). inV().inE('created').subgraph('createdG').iterate() traversal().with(t.sideEffects.get('knowsG')).E() traversal().with(t.sideEffects.get('createdG')).E()
TinkerGraph is the ideal (and default) Graph
into which a subgraph is extracted as it’s fast, in-memory, and supports
user-supplied identifiers which can be any Java object. It is this last feature that needs some focus as many
TinkerPop-enabled graphs have complex identifier types and TinkerGraph’s ability to consume those makes it a perfect
host for an incoming subgraph. However care needs to be taken when using the elements of the TinkerGraph subgraph.
The original graph’s identifiers may be preserved, but the elements of the graph are now TinkerGraph objects like,
TinkerVertex
and TinkerEdge
. As a result, they can not be used directly in Gremlin running against the original
graph. For example, the following traversal would likely return an error:
Vertex v = sg.V().has('name','marko').next(); (1)
List<Vertex> vertices = g.V(v).out().toList(); (2)
-
Here "sg" is a reference to a TinkerGraph subgraph and "v" is a
TinkerVertex
. -
The
g.V(v)
has the potential to fail as "g" is the originalGraph
instance and not a TinkerGraph - it could reject theTinkerVertex
instance as it will not recognize it.
It is safer to wrap the TinkerVertex
in a ReferenceVertex
or simply reference the id()
as follows:
Vertex v = sg.V().has('name','marko').next();
List<Vertex> vertices = g.V(v.id()).out().toList();
// OR
Vertex v = new ReferenceVertex(sg.V().has('name','marko').next());
List<Vertex> vertices = g.V(v).out().toList();
Additional References
The substring()
-step (map) returns a substring with a 0-based start index (inclusive) and optionally an end index (exclusive) specified.
If the start index is negative then it will begin at the specified index counted from the end of the string, or 0 if exceeding the string length.
Likewise, if the end index is negative then it will end at the specified index counted from the end of the string, or 0 if exceeding the string length.
End index is optional, if it is not specified or if it exceeds the length of the string then all remaining characters will
be returned. End index ≤ start index will return the empty string. Null values are not processed and remain as null when returned.
If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.inject("test", "hello world", null).substring(1, 8) g.inject("hello world").substring(-4) (1) g.inject("hello world").substring(2, 0) (2) g.V().hasLabel("software").values("name").substring(2) g.V().hasLabel("software").values("name").fold().substring(local, 2) (3)
-
Negative start index, the first character is read by counting from the end of the string
-
Length of 0 specified will return the empty string
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
Additional References
substring(int)
substring(Scope,int)
substring(int,int)
substring(Scope,int,int)
The sum()
-step (map) operates on a stream of numbers and sums the numbers together to yield a result. Note that
the current traverser number is multiplied by the traverser bulk to determine how many such numbers are being
represented.
g.V().values('age').sum() g.V().repeat(both()).times(3).values('age').sum()
When called as sum(local)
it determines the sum of the current, local object (not the objects in the traversal
stream). This works for Collection
-type objects.
g.V().values('age').fold().sum(local)
When there are null
values being evaluated the null
objects are ignored, but if all values are recognized as null
the return value is null
.
g.inject(null,10, 9, null).sum() g.inject([null,null,null]).sum(local)
Additional References
The tail()
-step is analogous to limit()
-step, except that it emits the last n
-objects instead of
the first n
-objects.
g.V().values('name').order() g.V().values('name').order().tail() (1) g.V().values('name').order().tail(1) (2) g.V().values('name').order().tail(3) (3)
-
Last name (alphabetically).
-
Same as statement 1.
-
Last three names.
The tail()
-step can also be applied with Scope.local
, in which case it operates on the incoming collection.
g.V().as('a').out().as('a').out().as('a').select('a').by(tail(local)).values('name') (1) g.V().as('a').out().as('a').out().as('a').select('a').by(unfold().values('name').fold()).tail(local) (2) g.V().as('a').out().as('a').out().as('a').select('a').by(unfold().values('name').fold()).tail(local, 2) (3) g.V().elementMap().tail(local) (4)
-
Only the most recent name from the "a" step (
List<Vertex>
becomesVertex
). -
Same result as statement 1 (
List<String>
becomesString
). -
List<String>
for each path containing the last two names from the 'a' step. -
Map<String, Object>
for each vertex, but containing only the last property value.
Additional References
In many situations, a graph traversal is not about getting an exact answer as its about getting a relative ranking.
A classic example is recommendation. What is desired is a
relative ranking of vertices, not their absolute rank. Next, it may be desirable to have the traversal execute for
no more than 2 milliseconds. In such situations, timeLimit()
-step (filter) can be used.
Note
|
The method clock(int runs, Closure code) is a utility preloaded in the Gremlin Console
that can be used to time execution of a body of code.
|
g.V().repeat(both().groupCount('m')).times(16).cap('m').order(local).by(values,desc).next() clock(1) {g.V().repeat(both().groupCount('m')).times(16).cap('m').order(local).by(values,desc).next()} g.V().repeat(timeLimit(2).both().groupCount('m')).times(16).cap('m').order(local).by(values,desc).next() clock(1) {g.V().repeat(timeLimit(2).both().groupCount('m')).times(16).cap('m').order(local).by(values,desc).next()}
In essence, the relative order is respected, even through the number of traversers at each vertex is not. The primary
benefit being that the calculation is guaranteed to complete at the specified time limit (in milliseconds). Finally,
note that the internal clock of timeLimit()
-step starts when the first traverser enters it. When the time limit is
reached, any next()
evaluation of the step will yield a NoSuchElementException
and any hasNext()
evaluation will
yield false
.
Additional References
The times
-step is not an actual step, but is instead a step modulator for repeat()
(find more
documentation on the times()
there).
Additional References
The to()
-step is not an actual step, but instead is a "step-modulator" similar to as()
and
by()
. If a step is able to accept traversals or strings then to()
is the
means by which they are added. The general pattern is step().to()
. See from()
-step.
The list of steps that support to()
-modulation are: simplePath()
, cyclicPath()
,
path()
, and addE()
.
Additional References
The toLower()
-step (map) returns the lowercase representation of incoming string or list of string traverser. Null values are not processed and remain as null when returned.
If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.inject("HELLO", "wORlD", null).toLower() g.inject(["HELLO", "wORlD", null]).toLower(Scope.local) (1)
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
Additional References
The toUpper()
-step (map) returns the uppercase representation of incoming string or list of string traverser. Null values are not processed and remain as null when returned.
If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.inject("hello", "wORlD", null).toUpper() g.V().values("name").toUpper() (1) g.V().values("name").fold().toUpper(local) (2)
-
Returns the upper case representation of all vertex names.
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
Additional References
From any one element (i.e. vertex or edge), the emanating paths from that element can be aggregated to form a
tree. Gremlin provides tree()
-step (sideEffect) for such
this situation.
tree = g.V().out().out().tree().next()
It is important to see how the paths of all the emanating traversers are united to form the tree.
The resultant tree data structure can then be manipulated (see Tree
JavaDoc).
tree = g.V().out().out().tree().by('name').next() tree['marko'] tree['marko']['josh'] tree.getObjectsAtDepth(3)
Note that when using by()
-modulation, tree nodes are combined based on projection uniqueness, not on the
uniqueness of the original objects being projected. For instance:
g.V().has('name','josh').out('created').values('name').tree() (1) g.V().has('name','josh').out('created').values('name'). tree().by('name').by(label).by() (2)
-
When the
tree()
is created, vertex 3 and 5 are unique and thus, form unique branches in the tree structure. -
When the
tree()
isby()
-modulated bylabel
, then vertex 3 and 5 are both "software" and thus are merged to a single node in the tree.
Additional References
The trim()
-step (map) returns a string with leading and leading whitespace removed. Null values are not processed and remain
as null when returned. If the incoming traverser is a non-String value then an IllegalArgumentException
will be thrown.
g.inject(" hello ", " world ", null).trim() g.inject([" hello ", " world ", null]).trim(Scope.local) (1)
-
Use
Scope.local
to operate on individual string elements inside incoming list, which will return a list.
If the object reaching unfold()
(flatMap) is an iterator, iterable, or map, then it is unrolled into a linear
form. If not, then the object is simply emitted. Please see fold()
step for the inverse behavior.
g.V(1).out().fold().inject('gremlin',[1.23,2.34]) g.V(1).out().fold().inject('gremlin',[1.23,2.34]).unfold()
Note that unfold()
does not recursively unroll iterators. Instead, repeat()
can be used to for recursive unrolling.
inject(1,[2,3,[4,5,[6]]]) inject(1,[2,3,[4,5,[6]]]).unfold() inject(1,[2,3,[4,5,[6]]]).repeat(unfold()).until(count(local).is(1)).unfold()
Additional References
The union()
-step (branch) supports the merging of the results of an arbitrary number of traversals. When a
traverser reaches a union()
-step, it is copied to each of its internal steps. The traversers emitted from union()
are the outputs of the respective internal traversals.
g.V(4).union( __.in().values('age'), out().values('lang')) g.V(4).union( __.in().values('age'), out().values('lang')).path() g.union(V().has('person','name','vadas'), V().has('software','name','lop').in('created'))
Additional References
The until
-step is not an actual step, but is instead a step modulator for repeat()
(find more
documentation on the until()
there).
Additional References
The V()
-step is meant to read vertices from the graph and is usually used to start a GraphTraversal
, but can also
be used mid-traversal.
g.V(1) (1) g.V().has('name', within('marko', 'vadas', 'josh')).as('person'). V().has('name', within('lop', 'ripple')).addE('uses').from('person') (2)
-
Find the vertex by its unique identifier (i.e.
T.id
) - not all graphs will use a numeric value for their identifier. -
An example where
V()
is used both as a start step and in the middle of a traversal.
Note
|
Whether a mid-traversal V() uses an index or not, depends on a) whether suitable index exists and b) if the
particular graph system provider implemented this functionality.
|
g.V().has('name', within('marko', 'vadas', 'josh')).as('person'). V().has('name', within('lop', 'ripple')).addE('uses').from('person').toString() (1) g.V().has('name', within('marko', 'vadas', 'josh')).as('person'). V().has('name', within('lop', 'ripple')).addE('uses').from('person').iterate().toString() (2)
-
Normally the
V()
-step will iterate over all vertices. However, graph strategies can foldHasContainer
's into aGraphStep
to allow index lookups. -
Whether the graph system provider supports mid-traversal
V()
index lookups or not can easily be determined by inspecting thetoString()
output of the iterated traversal. Ifhas
conditions were folded into theV()
-step, an index - if one exists - will be used.
Additional References
The value()
-step (map) takes a Property
and extracts the value from it.
g.V(1).properties().value() g.V(1).properties().properties().value()
Additional References
The valueMap()
-step yields a Map
representation of the properties of an element.
Important
|
This step is the precursor to the elementMap()-step. Users should typically
choose elementMap() unless they utilize multi-properties. elementMap() effectively mimics the functionality of
valueMap(true).by(unfold()) as a single step.
|
g.V().valueMap() g.V().valueMap('age') g.V().valueMap('age','blah') g.E().valueMap()
It is important to note that the map of a vertex maintains a list of values for each key. The map of an edge or vertex-property represents a single property (not a list). The reason is that vertices in TinkerPop leverage vertex properties which support multiple values per key. Using the "The Crew" toy graph, the point is made explicit.
g.V().valueMap() g.V().has('name','marko').properties('location') g.V().has('name','marko').properties('location').valueMap()
To turn list of values into single items, the by()
modulator can be used as shown below.
g.V().valueMap().by(unfold()) g.V().valueMap('name','location').by().by(unfold())
If the id
, label
, key
, and value
of the Element
is desired, then the with()
modulator can be used to
trigger its insertion into the returned map.
g.V().hasLabel('person').valueMap().with(WithOptions.tokens) g.V().hasLabel('person').valueMap('name').with(WithOptions.tokens, WithOptions.labels) g.V().hasLabel('person').properties('location').valueMap().with(WithOptions.tokens, WithOptions.values)
Additional References
The values()
-step (map) extracts the values of properties from an Element
in the traversal stream.
g.V(1).values() g.V(1).values('location') g.V(1).properties('location').values()
Additional References
The vertex steps (flatMap) are fundamental to the Gremlin language. Via these steps, its possible to "move" on the graph — i.e. traverse.
-
out(string…)
: Move to the outgoing adjacent vertices given the edge labels. -
in(string…)
: Move to the incoming adjacent vertices given the edge labels. -
both(string…)
: Move to both the incoming and outgoing adjacent vertices given the edge labels. -
outE(string…)
: Move to the outgoing incident edges given the edge labels. -
inE(string…)
: Move to the incoming incident edges given the edge labels. -
bothE(string…)
: Move to both the incoming and outgoing incident edges given the edge labels. -
outV()
: Move to the outgoing vertex. -
inV()
: Move to the incoming vertex. -
bothV()
: Move to both vertices. -
otherV()
: Move to the vertex that was not the vertex that was moved from.
Groovy
|
The term |
Javascript
|
The term |
Python
|
The term |
g.V(4) g.V(4).outE() (1) g.V(4).inE('knows') (2) g.V(4).inE('created') (3) g.V(4).bothE('knows','created','blah') g.V(4).bothE('knows','created','blah').otherV() g.V(4).both('knows','created','blah') g.V(4).outE().inV() (4) g.V(4).out() (5) g.V(4).inE().outV() g.V(4).inE().bothV()
-
All outgoing edges.
-
All incoming knows-edges.
-
All incoming created-edges.
-
Moving forward touching edges and vertices.
-
Moving forward only touching vertices.
Additional References
The where()
-step filters the current object based on either the object itself (Scope.local
) or the path history
of the object (Scope.global
) (filter). This step is typically used in conjunction with either
match()
-step or select()
-step, but can be used in isolation.
g.V(1).as('a').out('created').in('created').where(neq('a')) (1) g.withSideEffect('a',['josh','peter']).V(1).out('created').in('created').values('name').where(within('a')) (2) g.V(1).out('created').in('created').where(out('created').count().is(gt(1))).values('name') (3)
-
Who are marko’s collaborators, where marko can not be his own collaborator? (predicate)
-
Of the co-creators of marko, only keep those whose name is josh or peter. (using a sideEffect)
-
Which of marko’s collaborators have worked on more than 1 project? (using a traversal)
Important
|
Please see match().where() and select().where()
for how where() can be used in conjunction with Map<String,Object> projecting steps — i.e. Scope.local .
|
A few more examples of filtering an arbitrary object based on a anonymous traversal is provided below.
g.V().where(out('created')).values('name') (1) g.V().out('knows').where(out('created')).values('name') (2) g.V().where(out('created').count().is(gte(2))).values('name') (3) g.V().where(out('knows').where(out('created'))).values('name') (4) g.V().where(__.not(out('created'))).where(__.in('knows')).values('name') (5) g.V().where(__.not(out('created')).and().in('knows')).values('name') (6) g.V().as('a').out('knows').as('b'). where('a',gt('b')). by('age'). select('a','b'). by('name') (7) g.V().as('a').out('knows').as('b'). where('a',gt('b').or(eq('b'))). by('age'). by('age'). by(__.in('knows').values('age')). select('a','b'). by('name') (8) g.V().as('a').both().both().as('b'). where('a',eq('b')).by('age') (9)
-
What are the names of the people who have created a project?
-
What are the names of the people that are known by someone one and have created a project?
-
What are the names of the people how have created two or more projects?
-
What are the names of the people who know someone that has created a project? (This only works in OLTP — see the
WARNING
below) -
What are the names of the people who have not created anything, but are known by someone?
-
The concatenation of
where()
-steps is the same as a singlewhere()
-step with an and’d clause. -
Marko knows josh and vadas but is only older than vadas.
-
Marko is younger than josh, but josh knows someone equal in age to marko (which is marko).
-
The "age" property is not productive for all vertices and therefore those values are filtered.
Warning
|
The anonymous traversal of where() processes the current object "locally". In OLAP, where the atomic unit
of computing is the vertex and its local "star graph," it is important that the anonymous traversal does not leave
the confines of the vertex’s star graph. In other words, it can not traverse to an adjacent vertex’s properties or
edges.
|
Additional References
The with()
-step is not an actual step, but is instead a "step modulator" which modifies the behavior of the step
prior to it. The with()
-step provides additional "configuration" information to steps that implement the Configuring
interface. Steps that allow for this type of modulation will explicitly state so in their documentation.
Javascript
|
The term |
Python
|
The term |
The write()
-step is not really a "step" but a step modulator in that it modifies the functionality of the io()
-step.
More specifically, it tells the io()
-step that it is expected to use its configuration to write data to some
location. Please see the documentation for io()
-step for more complete details on usage.
Additional References
A subset of gremlin steps are able to accept parameterized arguments also known as GValues. GValues can be used to
provide protection against gremlin-injection attacks in cases where untrusted and unsanitized inputs must be passed as
step arguments. Additionally, use of GValues may offer performance benefits in certain environments by making use of
some query caching capabilities. Note that the reference implementation of the gremlin language and gremlin-server
do
not have such a query caching mechanism, and thus will not see any performance improvements through parameterization. Users
should consult the documentation of their specific graph system details of potential performance benefits via parameterization.
Note
|
There are unique considerations regarding parameters when using gremlin-groovy scripts. Groovy allows for parameterization
at arbitrary points in the query in addition to the subset of parameterizable steps documented here. Groovy is also bound by
a comparatively slow script compilation, which makes parameterization essential for performant execution of gremlin-groovy scripts.
|
Step | Parameterizable arguments |
---|---|
String edgeLabel |
|
String vertexLabel |
|
String…edgeLabels |
|
String…edgeLabels |
|
Map params |
|
double probability |
|
Object values |
|
String delimiter |
|
E e/value |
|
Object values |
|
Object values |
|
Vertex fromVertex |
|
String label |
|
String label, String… labels |
|
String…edgeLabels |
|
String…edgeLabels |
|
Object values |
|
Long limit |
|
Object values |
|
Map searchCreate |
|
Map searchCreate |
|
M token, Map m |
|
String…edgeLabels |
|
String…edgeLabels |
|
Object values |
|
Long low, Long high |
|
Long limit |
|
Long limit |
|
String… edgeLabels, Vertex toVertex |
|
String…edgeLabels |
Additional References
A P
is a predicate of the form Function<Object,Boolean>
. That is, given some object, return true or false. As of
the release of TinkerPop 3.4.0, Gremlin also supports simple text predicates, which only work on String
values. The TextP
text predicates extend the P
predicates, but are specialized in that they are of the form Function<String,Boolean>
.
The provided predicates are outlined in the table below and are used in various steps such as has()
-step,
where()
-step, is()
-step, etc. Two new additional TextP
predicate members were added in the
TinkerPop 3.6.0 release that allow working with regular expressions. These are TextP.regex
and TextP.notRegex
Predicate | Description |
---|---|
|
Is the incoming object equal to the provided object? |
|
Is the incoming object not equal to the provided object? |
|
Is the incoming number less than the provided number? |
|
Is the incoming number less than or equal to the provided number? |
|
Is the incoming number greater than the provided number? |
|
Is the incoming number greater than or equal to the provided number? |
|
Is the incoming number greater than the first provided number and less than the second? |
|
Is the incoming number less than the first provided number or greater than the second? |
|
Is the incoming number greater than or equal to the first provided number and less than the second? |
|
Is the incoming object in the array of provided objects? |
|
Is the incoming object not in the array of the provided objects? |
|
Does the incoming |
|
Does the incoming |
|
Does the incoming |
|
Does the incoming |
|
Does the incoming |
|
Does the incoming |
|
Does the incoming |
|
Does the incoming |
Note
|
The TinkerPop reference implementation uses the Java Pattern and Matcher classes for it regular expression
engine. Other implementations may decide to use a different regular expression engine. It’s a good idea to check
the documentation for the implementation you are using to verify the allowed regular expression syntax.
|
eq(2) not(neq(2)) (1) not(within('a','b','c')) not(within('a','b','c')).test('d') (2) not(within('a','b','c')).test('a') within(1,2,3).and(not(eq(2))).test(3) (3) inside(1,4).or(eq(5)).test(3) (4) inside(1,4).or(eq(5)).test(5) between(1,2) (5) not(between(1,2))
-
The
not()
of aP
-predicate is anotherP
-predicate. -
P
-predicates are arguments to various steps which internallytest()
the incoming value. -
P
-predicates can be and’d together. -
P
-predicates can be or' together. -
and()
is aP
-predicate and thus, aP
-predicate can be composed of multipleP
-predicates.
Tip
|
To reduce the verbosity of predicate expressions, it is good to
import static org.apache.tinkerpop.gremlin.process.traversal.P.* .
|
The following example demonstrates how the regex()
predicate is used and it demonstrates an important point. When
using regex()
, the string is considered a match to the pattern if any substring matches the pattern. It is therefore
important to use the appropriate boundary matchers (e.g. $
for end of a line) to ensure a proper match.
g.V().has('person', 'name', regex('peter')).values('name') g.V().has('person', 'name', regex('r')).values('name') g.V().has('person', 'name', regex('r$')).values('name')
Finally, note that where()
-step takes a P<String>
. The provided string value refers to a variable
binding, not to the explicit string value.
g.V().as('a').both().both().as('b').count() g.V().as('a').both().both().as('b').where('a',neq('b')).count()
Many steps in Gremlin return Map
-based results. Commonly used steps like project()
,
'group()', and select()
are just some examples of steps that fall into this category.
When working with Map
results there are a couple of important things to know.
First, it is important to recognize that there is a bit of a difference in behavior that occurs when using
unfold() on a Map
in embedded contexts versus remote contexts. In embedded contexts, an unfolded Map
becomes its composite Map.Entry
objects as is typical in Java. The following example demonstrates the basic name/value
pairs that returned:
g.V().valueMap('name','age').unfold()
In remote contexts, an unfolded Map
becomes Map.Entry
on the server as in the embedded case, but is returned to the
application as a Map
with one entry. The slight difference in notation in Gremlin Console is shown in the following
remote example:
gremlin> g.V().valueMap('name','age').unfold()
==>[name:[marko]]
==>[age:[29]]
==>[name:[vadas]]
==>[age:[27]]
==>[name:[lop]]
==>[name:[josh]]
==>[age:[32]]
==>[name:[ripple]]
==>[name:[peter]]
==>[age:[35]]
The primary reason for this difference lies in the fact that Gremlin Language Variants, like Python and Go, do not have
a native Map.Entry
concept that can be used. The most universal data structure across programming languages is the
Map
itself. It is important to note that this transformation from Map.Entry
to Map
only applies to results
received on the client-side. In other words, if a step was to follow unfold()
in the prior example, it would be
dealing with Map.Entry
and not a Map
, so Gremlin semantics should remain consistent on the server side.
The second issues to consider with steps that return a Map
is that access keys on a Map
is not always as consistent
as expected. The issue is best demonstrated in some examples:
// note that elements can be grouped by(id), but that same pattern can't be applied to get
// a T.id in a Map
gremlin> g.V().hasLabel('person').both().group().by(id)
==>[1:[v[1],v[1]],2:[v[2]],3:[v[3],v[3],v[3]],4:[v[4]],5:[v[5]]]
gremlin> g.V().hasLabel('person').both().elementMap().group().by(id)
TokenTraversal support of java.util.LinkedHashMap does not allow selection by id
Type ':help' or ':h' for help.
Display stack trace? [yN]
// note that select() can't be used if the key is a non-string
gremlin> g.V().hasLabel('person').both().group().by('age').select(32)
No signature of method: org.apache.tinkerpop.gremlin.process.traversal.dsl.graph.DefaultGraphTraversal.select() is applicable for argument types: (Integer) values: [32]
Possible solutions: reset(), collect(), sleep(long), collect(groovy.lang.Closure), inject(groovy.lang.Closure), split(groovy.lang.Closure)
Type ':help' or ':h' for help.
Display stack trace? [yN]
While this problem might be solved in future versions, the workaround for both cases is to use constant() as shown in the following example:
g.V().hasLabel('person').both().group().by(constant(id)) g.V().hasLabel('person').both().group().by('age').select(constant(32))
Gremlin is primarily a
lazy, stream processing language. This means that Gremlin fully
processes (to the best of its abilities) any traversers currently in the traversal pipeline before getting more data
from the start/head of the traversal. However, there are numerous situations in which a completely lazy computation
is not possible (or impractical). When a computation is not lazy, a "barrier step" exists. There are three types of
barriers:
-
CollectingBarrierStep
: All of the traversers prior to the step are put into a collection and then processed in some way (e.g. ordered) prior to the collection being "drained" one-by-one to the next step. Examples include:order()
,sample()
,aggregate()
,barrier()
. -
ReducingBarrierStep
: All of the traversers prior to the step are processed by a reduce function and once all the previous traversers are processed, a single "reduced value" traverser is emitted to the next step. Note that the path history leading up to a reducing barrier step is destroyed given its many-to-one nature. Examples include:fold()
,count()
,sum()
,max()
,min()
. -
SupplyingBarrierStep
: All of the traversers prior to the step are iterated (no processing) and then some provided supplier yields a single traverser to continue to the next step. Examples include:cap()
.
In Gremlin OLAP (see TraversalVertexProgram
), a barrier is introduced at the end of
every adjacent vertex step. This means that the traversal does its best to compute as much as
possible at the current, local vertex. What it can’t compute without referencing an adjacent vertex is aggregated
into a barrier collection. When there are no more traversers at the local vertex, the barriered traversers are the
messages that are propagated to remote vertices for further processing.
The Scope
enum has two constants: Scope.local
and Scope.global
. Scope determines whether the particular step
being scoped is with respects to the current object (local
) at that step or to the entire stream of objects up to that
step (global
).
Python
|
The term |
g.V().has('name','marko').out('knows').count() (1) g.V().has('name','marko').out('knows').fold().count() (2) g.V().has('name','marko').out('knows').fold().count(local) (3) g.V().has('name','marko').out('knows').fold().count(global) (4)
-
Marko knows 2 people.
-
A list of Marko’s friends is created and thus, one object is counted (the single list).
-
A list of Marko’s friends is created and a
local
-count yields the number of objects in that list. -
count(global)
is the same ascount()
as the default behavior for most scoped steps isglobal
.
The steps that support scoping are:
-
count()
: count the local collection or global stream. -
dedup()
: dedup the local collection of global stream. -
max()
: get the max value in the local collection or global stream. -
mean()
: get the mean value in the local collection or global stream. -
min()
: get the min value in the local collection or global stream. -
order()
: order the objects in the local collection or global stream. -
range()
: clip the local collection or global stream. -
limit()
: clip the local collection or global stream. -
sample()
: sample objects from the local collection or global stream. -
tail()
: get the tail of the objects in the local collection or global stream.
A few more examples of the use of Scope
are provided below:
g.V().both().group().by(label).select('software').dedup(local) g.V().groupCount().by(label).select(values).min(local) g.V().groupCount().by(label).order(local).by(values,desc) g.V().fold().sample(local,2)
Finally, note that local()
-step is a "hard-scoped step" that transforms any internal traversal into a
locally-scoped operation. A contrived example is provided below:
g.V().fold().local(unfold().count()) g.V().fold().count(local)
A lambda is a function
that can be referenced by software and thus, passed around like any other piece of data. In Gremlin, lambdas make it
possible to generalize the behavior of a step such that custom steps can be created (on-the-fly) by the user. However,
it is advised to avoid using lambdas if possible.
g.V().filter{it.get().value('name') == 'marko'}. flatMap{it.get().vertices(OUT,'created')}. map {it.get().value('name')} (1) g.V().has('name','marko').out('created').values('name') (2)
-
A lambda-rich Gremlin traversal which should and can be avoided. (bad)
-
The same traversal (result), but without using lambdas. (good)
Gremlin attempts to provide the user a comprehensive collection of steps in the hopes that the user will never need to leverage a lambda in practice. It is advised that users only leverage a lambda if and only if there is no corresponding lambda-less step that encompasses the desired functionality. The reason being, lambdas can not be optimized by Gremlin’s compiler strategies as they can not be programmatically inspected (see traversal strategies). It is also not currently possible to send a natively written lambda for remote execution to Gremlin-Server or a driver that supports remote execution.
In many situations where a lambda could be used, either a corresponding step exists or a traversal can be provided in
its place. A TraversalLambda
behaves like a typical lambda, but it can be optimized and it yields less objects than
the corresponding pure-lambda form.
g.V().out().out().path().by {it.value('name')}. by {it.value('name')}. by {g.V(it).in('created').values('name').fold().next()} (1) g.V().out().out().path().by('name'). by('name'). by(__.in('created').values('name').fold()) (2)
-
The length-3 paths have each of their objects transformed by a lambda. (bad)
-
The length-3 paths have their objects transformed by a lambda-less step and a traversal lambda. (good)
A
TraversalStrategy
analyzes a Traversal
and, if the traversal
meets its criteria, can mutate it accordingly. Traversal strategies are executed at compile-time and form the foundation
of the Gremlin traversal machine’s compiler. There are 5 categories of strategies which are itemized below:
-
There is an application-level feature that can be embedded into the traversal logic (decoration).
-
There is a more efficient way to express the traversal at the TinkerPop level (optimization).
-
There is a more efficient way to express the traversal at the graph system/language/driver level (provider optimization).
-
There are some final adjustments/cleanups/analyses required before executing the traversal (finalization).
-
There are certain traversals that are not legal for the application or traversal engine (verification).
Note
|
The explain() -step shows the user how each registered strategy mutates the traversal.
|
TinkerPop ships with a generous number of TraversalStrategy
definitions, most of which are applied implicitly when
executing a gremlin traversal. Users and providers can add TraversalStrategy
definitions for particular needs. The
following sections detail how traversal strategies are applied and defined and describe a collection of traversal
strategies that are generally useful to end-users.
One can explicitly add or remove TraversalStrategy
strategies on the GraphTraversalSource
with the withStrategies()
and withoutStrategies()
start steps, see the ReadOnlyStrategy and the
barrier() step for examples. End users typically do this as part of issuing a gremlin traversal, either
on a locally opened graph or a remotely accessed graph. However, when configuring Gremlin Server, traversal strategies
can also be applied on exposed GraphTraversalSource
instances and as part of an Authorizer
implementation, see
Gremlin Server Authorization.
Therefore, one should keep the following in mind when modifying the list of TraversalStrategy
strategies:
-
A
TraversalStrategy
added to the traversal can be removed again later on. An example is theconf/gremlin-server-modern-readonly.yaml
file from the Gremlin Server distribution, which applies theReadOnlyStrategy
to theGraphTraversalSource
that remote clients can connect to. However, a remote client can remove it on its turn by applying thewithoutStrategies()
step with theReadOnlyStrategy
. -
When a
TraversalStrategy
of a particular type is added, it replaces any instances of its type that exist prior to it. Multiple instances of aTraversalStrategy
can therefore not be registered and their functionality is no way merged automatically. Therefore, if there is a particular strategy registered whose functionality needs to be changed it is important to either find and modify the existing instance or construct a new one copying the options to keep from the old to the new instance.
A simple OptimizationStrategy
is the IdentityRemovalStrategy
.
public final class IdentityRemovalStrategy extends AbstractTraversalStrategy<TraversalStrategy.OptimizationStrategy> implements TraversalStrategy.OptimizationStrategy {
private static final IdentityRemovalStrategy INSTANCE = new IdentityRemovalStrategy();
private IdentityRemovalStrategy() {
}
@Override
public void apply(Traversal.Admin<?, ?> traversal) {
if (traversal.getSteps().size() <= 1)
return;
for (IdentityStep<?> identityStep : TraversalHelper.getStepsOfClass(IdentityStep.class, traversal)) {
if (identityStep.getLabels().isEmpty() || !(identityStep.getPreviousStep() instanceof EmptyStep)) {
TraversalHelper.copyLabels(identityStep, identityStep.getPreviousStep(), false);
traversal.removeStep(identityStep);
}
}
}
public static IdentityRemovalStrategy instance() {
return INSTANCE;
}
}
This strategy simply removes any IdentityStep
steps in the Traversal as aStep().identity().identity().bStep()
is equivalent to aStep().bStep()
. For those traversal strategies that require other strategies to execute prior or
post to the strategy, then the following two methods can be defined in TraversalStrategy
(with defaults being an
empty set). If the TraversalStrategy
is in a particular traversal category (i.e. decoration, optimization,
provider-optimization, finalization, or verification), then priors and posts are only possible within the respective category.
public Set<Class<? extends S>> applyPrior();
public Set<Class<? extends S>> applyPost();
Important
|
TraversalStrategy categories are sorted within their category and the categories are then executed in
the following order: decoration, optimization, provider optimization, finalization, and verification. If a designed strategy
does not fit cleanly into these categories, then it can implement TraversalStrategy and its prior and posts can reference
strategies within any category. However, such generalization are strongly discouraged.
|
An example of a GraphSystemOptimizationStrategy
is provided below.
g.V().has('name','marko')
The expression above can be executed in a O(|V|)
or O(log(|V|)
fashion in TinkerGraph
depending on whether there is or is not an index defined for "name."
public final class TinkerGraphStepStrategy extends AbstractTraversalStrategy<TraversalStrategy.ProviderOptimizationStrategy> implements TraversalStrategy.ProviderOptimizationStrategy {
private static final TinkerGraphStepStrategy INSTANCE = new TinkerGraphStepStrategy();
private TinkerGraphStepStrategy() {
}
@Override
public void apply(Traversal.Admin<?, ?> traversal) {
if (TraversalHelper.onGraphComputer(traversal))
return;
for (GraphStep originalGraphStep : TraversalHelper.getStepsOfClass(GraphStep.class, traversal)) {
TinkerGraphStep<?, ?> tinkerGraphStep = new TinkerGraphStep<>(originalGraphStep);
TraversalHelper.replaceStep(originalGraphStep, tinkerGraphStep, traversal);
Step<?, ?> currentStep = tinkerGraphStep.getNextStep();
while (currentStep instanceof HasStep || currentStep instanceof NoOpBarrierStep) {
if (currentStep instanceof HasStep) {
for (HasContainer hasContainer : ((HasContainerHolder) currentStep).getHasContainers()) {
if (!GraphStep.processHasContainerIds(tinkerGraphStep, hasContainer))
tinkerGraphStep.addHasContainer(hasContainer);
}
TraversalHelper.copyLabels(currentStep, currentStep.getPreviousStep(), false);
traversal.removeStep(currentStep);
}
currentStep = currentStep.getNextStep();
}
}
}
public static TinkerGraphStepStrategy instance() {
return INSTANCE;
}
}
The traversal is redefined by simply taking a chain of has()
-steps after g.V()
(TinkerGraphStep
) and providing
their HasContainers
to TinkerGraphStep
. Then its up to TinkerGraphStep
to determine if an appropriate index exists.
Given that the strategy uses non-TinkerPop provided steps, it should go into the ProviderOptimizationStrategy
category
to ensure the added step does not interfere with the assumptions of the OptimizationStrategy
strategies.
t = g.V().has('name','marko'); null t.toString() t.iterate(); null t.toString()
Warning
|
The reason that OptimizationStrategy and ProviderOptimizationStrategy are two different categories is
that optimization strategies should only rewrite the traversal using TinkerPop steps. This ensures that the
optimizations executed at the end of the optimization strategy round are TinkerPop compliant. From there, provider
optimizations can analyze the traversal and rewrite the traversal as desired using graph system specific steps (e.g.
replacing GraphStep.HasStep…HasStep with TinkerGraphStep ). If provider optimizations use graph system specific
steps and implement OptimizationStrategy , then other TinkerPop optimizations may fail to optimize the traversal or
mis-understand the graph system specific step behaviors (e.g. ProviderVertexStep extends VertexStep ) and yield
incorrect semantics.
|
Finally, here is a complicated traversal that has various components that are optimized by the default TinkerPop strategies.
g.V().hasLabel('person'). (1) and(has('name'), (2) has('name','marko'), filter(has('age',gt(20)))). (3) match(__.as('a').has('age',lt(32)), (4) __.as('a').repeat(outE().inV()).times(2).as('b')). (5) where('a',neq('b')). (6) where(__.as('b').both().count().is(gt(1))). (7) select('b'). (8) groupCount(). by(out().count()). (9) explain()
-
TinkerGraphStepStrategy
pulls inhas()
-step predicates for global, graph-centric index lookups. -
FilterRankStrategy
sorts filter steps by their time/space execution costs. -
InlineFilterStrategy
de-nests filters to increase the likelihood of filter concatenation and aggregation. -
InlineFilterStrategy
pulls out named predicates frommatch()
-step to more easily allow provider strategies to use indices. -
RepeatUnrollStrategy
will unroll loops andIncidentToAdjacentStrategy
will turnoutE().inV()
-patterns intoout()
. -
MatchPredicateStrategy
will pull inwhere()
-steps so that they can be subjected tomatch()
-steps runtime query optimizer. -
CountStrategy
will limit the traversal to only the number of traversers required for thecount().is(x)
-check. -
PathRetractionStrategy
will remove paths from the traversers and increase the likelihood of bulking as path data is not required afterselect('b')
. -
AdjacentToIncidentStrategy
will turnout()
intooutE()
to increase data access locality.
EdgeLabelVerificationStrategy
prevents traversals from writing traversals that do not explicitly specify and edge
label when using steps like out()
, 'in()', 'both()' and their related E
oriented steps, providing the
option to throw an exception, log a warning or do both when one of these keys is encountered in a mutating step.
EdgeLabelVerificationStrategy verificationStrategy = EdgeLabelVerificationStrategy.build()
.throwException().create()
// results in VerificationException - as out() does not have a label specified
g.withStrategies(verificationStrategy).V(1).out().iterate();
// results in VerificationException - as out() does not have a label specified
g.withStrategies(new EdgeLabelVerificationStrategy(throwException: true))
.V(1).out().iterate()
// results in VerificationException - as out() does not have a label specified
g.WithStrategies(new EdgeLabelVerificationStrategy(throwException: true))
.V(1).Out().Iterate();
// results in Error - as out() does not have a label specified
g.withStrategies(new EdgeLabelVerificationStrategy(throwException: true))
.V(1).out().iterate();
// results in Error - as out() does not have a label specified
g.withStrategies(EdgeLabelVerificationStrategy(throwException=true))
.V(1).out().iterate()
ElementIdStrategy
provides control over element identifiers. Some Graph implementations, such as TinkerGraph
,
allow specification of custom identifiers when creating elements:
g = traversal().withEmbedded(TinkerGraph.open()) v = g.addV().property(id,'42a').next() g.V('42a')
Other Graph
implementations generate element identifiers automatically and cannot be assigned.
As a helper, ElementIdStrategy
can be used to make identifier assignment possible by using vertex and edge indices
under the hood. Note that this demonstration is using TinkerGraph
for convenience, however in practice
ElementIdStrategy
offers little value in a Graph
which allows custom identifiers.
strategy = ElementIdStrategy.build().create() g = traversal().withEmbedded(TinkerGraph.open()).withStrategies(strategy) g.addV().property(id, '42a').id() g.V().elementMap()
Important
|
The key that is used to store the assigned identifier should be indexed in the underlying graph database. If it is not indexed, then lookups for the elements that use these identifiers will perform a linear scan. |
The purpose of the EventStrategy
is to raise events to one or more MutationListener
objects as changes to the
underlying Graph
occur within a Traversal
. Such a strategy is useful for logging changes, triggering certain
actions based on change, or any application that needs notification of some mutating operation during a Traversal
.
If the transaction is rolled back, the event queue is reset.
The following events are raised to the MutationListener
:
-
New vertex
-
New edge
-
Vertex property changed
-
Edge property changed
-
Vertex property removed
-
Edge property removed
-
Vertex removed
-
Edge removed
To start processing events from a Traversal
first implement the MutationListener
interface. An example of this
implementation is the ConsoleMutationListener
which writes output to the console for each event. The following
console session displays the basic usage:
import org.apache.tinkerpop.gremlin.process.traversal.step.util.event.* graph = TinkerFactory.createModern() l = new ConsoleMutationListener(graph) strategy = EventStrategy.build().addListener(l).create() g = traversal().with(graph).withStrategies(strategy) g.addV().property('name','stephen') g.V().has('name','stephen'). property(list, 'location', 'centreville', 'startTime', 1990, 'endTime', 2000). property(list, 'location', 'dulles', 'startTime', 2000, 'endTime', 2006). property(list, 'location', 'purcellville', 'startTime', 2006) g.V().has('name','stephen'). property(set, 'location', 'purcellville', 'startTime', 2006, 'endTime', 2019) g.E().drop()
By default, the EventStrategy
is configured with an EventQueue
that raises events as they occur within execution
of a Step
. As such, the final line of Gremlin execution that drops all edges shows a bit of an inconsistent count,
where the removed edge count is accounted for after the event is raised. The strategy can also be configured with a
TransactionalEventQueue
that captures the changes within a transaction and does not allow them to fire until the
transaction is committed.
Warning
|
EventStrategy is not meant for usage in tracking global mutations across separate processes. In other
words, a mutation in one JVM process is not raised as an event in a different JVM process. In addition, events are
not raised when mutations occur outside of the Traversal context.
|
Another default configuration for EventStrategy
revolves around the concept of "detachment". Graph elements are
detached from the graph as copies when passed to referring mutation events. Therefore, when adding a new Vertex
in
TinkerGraph, the event will not contain a TinkerVertex
but will instead include a DetachedVertex
. This behavior
can be modified with the detach()
method on the EventStrategy.Builder
which accepts the following inputs: null
meaning no detachment and the return of the original element, DetachedFactory
which is the same as the default
behavior, and ReferenceFactory
which will return "reference" elements only with no properties.
Important
|
If setting the detach() configuration to null , be aware that transactional graphs will likely create a
new transaction immediately following the commit() that raises the events. The graph elements raised in the events
may also not behave as "snapshots" at the time of their creation as they are "live" references to actual database
elements.
|
PartitionStrategy
partitions the vertices and edges of a graph into String
named partitions (i.e. buckets,
subgraphs, etc.). The idea behind PartitionStrategy
is presented in the image above where each element is in a
single partition (represented by its color). Partitions can be read from, written to, and linked/joined by edges
that span one or two partitions (e.g. a tail vertex in one partition and a head vertex in another).
There are three primary configurations in PartitionStrategy
:
-
Partition Key - The property key that denotes a String value representing a partition.
-
Write Partition - A
String
denoting what partition all future written elements will be in. -
Read Partitions - A
Set<String>
of partitions that can be read from.
The best way to understand PartitionStrategy
is via example.
graph = TinkerFactory.createModern() strategyA = new PartitionStrategy(partitionKey: "_partition", writePartition: "a", readPartitions: ["a"]) strategyB = new PartitionStrategy(partitionKey: "_partition", writePartition: "b", readPartitions: ["b"]) gA = traversal().with(graph).withStrategies(strategyA) gA.addV() // this vertex has a property of {_partition:"a"} gB = traversal().with(graph).withStrategies(strategyB) gB.addV() // this vertex has a property of {_partition:"b"} gA.V() gB.V()
The following examples demonstrate the above PartitionStrategy
definition for "strategyA" in other programming
languages:
PartitionStrategy strategyA = PartitionStrategy.build().partitionKey("_partition")
.writePartition("a")
.readPartitions("a").create();
PartitionStrategy strategyA = new PartitionStrategy(
partitionKey: "_partition", writePartition: "a",
readPartitions: new List<string>(){"a"});
const strategyA = new PartitionStrategy(partitionKey: "_partition", writePartition: "a", readPartitions: ["a"])
strategyA = PartitionStrategy(partitionKey="_partition", writePartition="a", readPartitions=["a"])
Partitions may also extend to VertexProperty
elements if the Graph
can support meta-properties and if the
includeMetaProperties
value is set to true
when the PartitionStrategy
is built. The partitionKey
will be
stored in the meta-properties of the VertexProperty
and blind the traversal to those properties. Please note that
the VertexProperty
will only be hidden by way of the Traversal
itself. For example, calling Vertex.property(k)
bypasses the context of the PartitionStrategy
and will thus allow all properties to be accessed.
By writing elements to particular partitions and then restricting read partitions, the developer is able to create multiple graphs within a single address space. Moreover, by supporting references between partitions, it is possible to merge those multiple graphs (i.e. join partitions).
ReadOnlyStrategy
is largely self-explanatory. A Traversal
that has this strategy applied will throw an
IllegalStateException
if the Traversal
has any mutating steps within it.
ReadOnlyStrategy verificationStrategy = ReadOnlyStrategy.instance();
// results in VerificationException
g.withStrategies(verificationStrategy).addV('person').iterate();
// results in VerificationException
g.withStrategies(ReadOnlyStrategy).addV('person').iterate();
// results in VerificationException
g.WithStrategies(new ReadOnlyStrategy()).addV("person").Iterate();
// results in Error
g.withStrategies(new ReadOnlyStrategy()).addV("person").iterate();
// results in Error
g.withStrategies(ReadOnlyStrategy).addV("person").iterate()
ReservedKeysVerificationStrategy
prevents traversals from adding property keys that are protected, providing the
option to throw an exception, log a warning or do both when one of these keys is encountered in a mutating step. By
default "id" and "label" are considered "reserved" but the default can be changed by building with the
reservedKeys()
options and supply a Set
of keys to trigger the VerificationException
.
ReservedKeysVerificationStrategy verificationStrategy = ReservedKeysVerificationStrategy.build()
.throwException().create()
// results in VerificationException
g.withStrategies(verificationStrategy).addV('person').property("id",123).iterate();
// results in VerificationException
g.withStrategies(new ReservedKeysVerificationStrategy(throwException: true))
.addV('person').property("id",123).iterate()
// results in VerificationException
g.WithStrategies(new ReservedKeysVerificationStrategy(throwException: true))
.AddV('person').Property("id",123).Iterate();
// results in Error
g.withStrategies(new ReservedKeysVerificationStrategy(throwException: true))
.addV('person').property("id",123).iterate();
// results in Error
g.withStrategies(ReservedKeysVerificationStrategy(throwException=true))
.addV('person').property("id",123).iterate()
There are number of components of the Gremlin language that, by design, can produce non-deterministic results:
To get these steps to return deterministic results, SeedStrategy
allows assignment of a seed value to the Random
operations of the steps. The following example demonstrates the random nature of shuffle
:
g.V().values('name').fold().order(local).by(shuffle) g.V().values('name').fold().order(local).by(shuffle) g.V().values('name').fold().order(local).by(shuffle) g.V().values('name').fold().order(local).by(shuffle) g.V().values('name').fold().order(local).by(shuffle)
With SeedStrategy
in place, however, the same order is applied each time:
seedStrategy = SeedStrategy.build().seed(999998L).create() g.withStrategies(seedStrategy).V().values('name').fold().order(local).by(shuffle) g.withStrategies(seedStrategy).V().values('name').fold().order(local).by(shuffle) g.withStrategies(seedStrategy).V().values('name').fold().order(local).by(shuffle) g.withStrategies(seedStrategy).V().values('name').fold().order(local).by(shuffle) g.withStrategies(seedStrategy).V().values('name').fold().order(local).by(shuffle)
Important
|
SeedStrategy only makes specific steps behave in a deterministic fashion and does not necessarily make
the entire traversal deterministic itself. If the underlying graph database or processing engine happens to not
guarantee iteration order, then it is possible that the final result of the traversal will appear to be
non-deterministic. In these cases, it would be necessary to enforce a deterministic iteration with order() prior to
these steps that make use of randomness to return results.
|
SubgraphStrategy
is similar to PartitionStrategy
in that it constrains a Traversal
to certain vertices, edges,
and vertex properties as determined by a Traversal
-based criterion defined individually for each.
graph = TinkerFactory.createTheCrew() g = traversal().with(graph) g.V().as('a').values('location').as('b'). (1) select('a','b').by('name').by() g = g.withStrategies(new SubgraphStrategy(vertexProperties: hasNot('endTime'))) (2) g.V().as('a').values('location').as('b'). (3) select('a','b').by('name').by() g.V().as('a').values('location').as('b'). select('a','b').by('name').by().explain()
-
Get all vertices and their vertex property locations.
-
Create a
SubgraphStrategy
where vertex properties must not have anendTime
-property (thus, the current location). -
Get all vertices and their current vertex property locations.
The following examples demonstrate the above SubgraphStrategy
definition in other programming languages:
g.withStrategies(SubgraphStrategy.build().vertexProperties(hasNot("endTime")).create());
g.WithStrategies(new SubgraphStrategy(vertexProperties: HasNot("endTime")));
g.withStrategies(new SubgraphStrategy(vertexProperties: hasNot("endTime")));
g.withStrategies(new SubgraphStrategy(vertexProperties=hasNot("endTime")))
Important
|
This strategy is implemented such that the vertices attached to an Edge must both satisfy the vertex
criterion (if present) in order for the Edge to be considered a part of the subgraph.
|
The example below uses all three filters: vertex, edge, and vertex property. People vertices must have lived in more than three places, edges must be labeled "develops," and vertex properties must be the persons current location or a non-location property.
graph = TinkerFactory.createTheCrew() g = traversal().with(graph).withStrategies(SubgraphStrategy.build(). vertices(or(hasNot('location'),properties('location').count().is(gt(3)))). edges(hasLabel('develops')). vertexProperties(or(hasLabel(neq('location')),hasNot('endTime'))).create()) g.V().elementMap() g.E().elementMap() g.V().outE().inV(). path(). by('name'). by(). by('name')
Like the ReadOnlyStrategy
, the VertexProgramDenyStrategy
denies the execution of specific traversals. A Traversal
that has the VertexProgramDenyStrategy
applied will throw an IllegalStateException
if it uses the
withComputer()
step. This TraversalStrategy
can be useful for configuring GraphTraversalSource
instances in
Gremlin Server with the ScriptFileGremlinPlugin
.
gremlin> oltpOnly = g.withStrategies(VertexProgramDenyStrategy.instance())
==>graphtraversalsource[tinkergraph[vertices:5 edges:7], standard]
gremlin> oltpOnly.withComputer().V().elementMap()
The TraversalSource does not allow the use of a GraphComputer
Type ':help' or ':h' for help.
Display stack trace? [yN]
Gremlin is a domain specific language (DSL) for traversing graphs. It operates in the language of vertices, edges and properties. Typically, applications built with Gremlin are not of the graph domain, but instead model their domain within a graph. For example, the "modern" toy graph models software and person domain objects with the relationships between them (i.e. a person "knows" another person and a person "created" software).
An analyst who wanted to find out if "marko" knows "josh" could write the following Gremlin:
g.V().hasLabel('person').has('name','marko').
out('knows').hasLabel('person').has('name','josh').hasNext()
While this method achieves the desired answer, it requires the analyst to traverse the graph in the domain language of the graph rather than the domain language of the social network. A more natural way for the analyst to write this traversal might be:
g.persons('marko').knows('josh').hasNext()
In the statement above, the traversal is written in the language of the domain, abstracting away the underlying graph structure from the query. The two traversal results are equivalent and, indeed, the "Social DSL" produces the same set of traversal steps as the "Graph DSL" thus producing equivalent strategy application and performance runtimes.
To further the example of the Social DSL consider the following:
// Graph DSL - find the number of persons who created at least 2 projects
g.V().hasLabel('person').
where(outE("created").count().is(P.gte(2))).count()
// Social DSL - find the number of persons who created at least 2 projects
social.persons().where(createdAtLeast(2)).count()
// Graph DSL - determine the age of the youngest friend "marko" has
g.V().hasLabel('person').has('name','marko').
out("knows").hasLabel("person").values("age").min()
// Social DSL - determine the age of the youngest friend "marko" has
social.persons("marko").youngestFriendsAge()
Learn more about how to implement these DSLs in the Gremlin Language Variants section specific to the programming language of interest.
There are times when is helpful to translate Gremlin from one programming language to another. Perhaps a large Gremlin
example is found on StackOverflow written in Java, but the programming language the developer has chosen is Python.
Fortunately, TinkerPop has developed Translator
infrastructure that will convert Gremlin from one programming
language syntax to another.
The GremlinTranslator
class is capable of translating gremlin-lang
syntax scripts into any of the following
gremlin language syntaxes: Java, Python, .Net, Go, JavaScript, Groovy. There is additionally a special anonymizing
translator which strips out arguments which could potentially contain sensitive information.
import org.apache.tinkerpop.gremlin.language.translator.GremlinTranslator;
import org.apache.tinkerpop.gremlin.language.translator.Translator;
// Java
GremlinTranslator.translate("g.V().hasLabel('person', 'software', 'class')", Translator.JAVA).getTranslated();
// Returns: "g.V().hasLabel(\"person\", \"software\", \"class\")"
// Python
GremlinTranslator.translate("g.V().hasLabel('person', 'software', 'class')", Translator.PYTHON).getTranslated();
// Returns: "g.V().has_label('person', 'software', 'class')"
// .Net
GremlinTranslator.translate("g.V().hasLabel('person', 'software', 'class')", Translator.DOTNET).getTranslated();
// Returns: "g.V().HasLabel(\"person\", \"software\", \"class\")"
// Go
GremlinTranslator.translate("g.V().hasLabel('person', 'software', 'class')", Translator.GO).getTranslated();
// Returns: "g.V().HasLabel(\"person\", \"software\", \"class\")"
// JavaScript
GremlinTranslator.translate("g.V().hasLabel('person', 'software', 'class')", Translator.JAVASCRIPT).getTranslated();
// Returns: "g.V().hasLabel(\"person\", \"software\", \"class\")"
// Groovy
GremlinTranslator.translate("g.V().hasLabel('person', 'software', 'class')", Translator.GROOVY).getTranslated();
// Returns: "g.V().hasLabel('person', 'software', 'class')"
// Anonymized
GremlinTranslator.translate("g.V().hasLabel('person', 'software', 'class')", Translator.ANONYMIZED).getTranslated();
// Returns: "g.V().hasLabel(string0, string1, string2)"
Each GLV is capable of producing a gremlin-lang
script from a Traversal, which can be fed into GremlinTranslator
to
convert into any of the supported syntaxes. Each GLV has a unique interface, but the pattern is quite similar from one
to the next:
GraphTraversal t = g.V().hasLabel("person");
String script = t.asAdmin().getGremlinLang().getGremlin();
// Returns: "g.V().hasLabel(\"person\")"
t = g.V().has_label('person')
script = t.gremlin_lang.get_gremlin()
# Returns: "g.V().hasLabel('person')"