Skip to content

Akka Persistence Storage Plugin Implementation for MapR-DB

Notifications You must be signed in to change notification settings

anicolaspp/akka-persistence-maprdb

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

akka-persistence-maprdb

Maven Central Build Status

This is a plugin for Akka Persistence that uses MapR-DB as backend. It implements a Journal store for saving the corresponding events for persistence entities and a Snapshot store.

Linking

Releases are pushed to Maven Central. Click the Maven Central back to check the latest version there.

<dependency>
  <groupId>com.github.anicolaspp</groupId>
  <artifactId>akka-persistence-maprdb_2.12</artifactId>
  <version>X.Y.Z</version>
</dependency>
libraryDependencies += "com.github.anicolaspp" % "akka-persistence-maprdb_2.12" % "X.Y.Z"

Activation

akka {
  extensions = [akka.persistence.Persistence]

  # This enables akka-persistence-maprdb plugin
  persistence {
    journal.plugin = "akka-persistence-maprdb.journal"
    snapshot-store.plugin = "akka-persistence-maprdb.snapshot"
  }
}

MapR-DB Configuration

we need some settings to be set up for MapR-DB.

  • maprdb.path is the base MFS path where our journals and snapshots live.
  • maprdb.pollingIntervalms is used by the query side for polling the new persistence entity ids.
  • maprb.driver.url can be used to configure what kind of MapR-DB implementation we want to use. ojai:mapr: points to the real MapR cluster. However, we could use an in-memory implementation through ojai-testing by indicating maprdb.driver.url = ojai:anicolaspp:mem. Notice that we package ojai-testing within akka-persistence-maprdb but it should not be used in production.

The following configuration shows that our journals and snapshots will live inside the folder /user/mapr/tables/akka in the MapR File System. We can indicate any valid location in the distributed file system within MFS. The polling interval is 5 seconds and we use a real MapR cluster through the MapR client and driver. For reference about OJAI, please see the related MapR Documentation

maprdb {
  path = "/user/mapr/tables/akka"
  
  pollingIntervalms = 5000
  
  driver {
    url = "ojai:mapr:"
  }
}

For each persistence entity, one MapR-DB table is created. For example, if we have the persistence entity user then two tables are automatically created.

/user/mapr/tables/akka/user.journal
/user/mapr/tables/akka/user.snapshot

These two tables are created automatically the first time the plugin is activated, after that they will consequently be used to read the initial state of the persistence entity when needed and to save new events and snapshots.

Persistence Entity Ids Table

One additional MapR-DB table is created along with your journals and snapshot. The table will have the following name:

/user/mapr/tables/akka/ids

Notice that the base path is what we indicated in the configuration. The table name is ids. This table is set of all persistence entity ids that is used in the query side. There are different ways to queries the persistence entity ids. One possible way is to obtain a handler to the MapR distributed file system (MFS) and enumerate the tables there. However, having an extra table (ids) makes all very easy.

MapR Client

akka-persistence-maprdb plugin uses OJAI and the MapR Client to communicate with the MapR Cluster. Make sure you have configured the MapR Client accordingly. In a secured cluster, make sure that the corresponding mapr ticket has been created so authentication happens correctly.

How data is stored in MapR-DB?

akka-persistence-maprdb plugin uses MapR-DB JSON to store the corresponding user-defined events and persistence entity snapshots into MapR-DB.

As mentioned above, each .journal table contains the events for the corresponding persistent entity and the following structure is used.

{
  "_id": {"$binary": sequenceNr in binary format},
  "persistentRepr": {"$binary": persistentRepr in binary format},
  "deleted": {"$boolean": true or false}
}

Each row is an even and they are sorted by MapR-DB based on the _id in ASC order.

Each .snapshot table represents the snapshots taken for a specific persistent entity and the following structure is used.

{
  "_id": "persistenceId_sequenceNr_timestamp", // this is String sorted ASC by default
  "meta": {
    "persistenceId": {"$long": persistenceId},
    "sequenceNr": {"$long": sequenceNr},
    "timestamp": timestamp 
  },
  "snapshot": {"$binary": binary representation of the Snapshot class}
}

Inspecting your Journals and Snapshots

At any moment in time, we could use one of the many available tools to inspect the corresponding structures created by this library.

Using MapR DBShell

mapr dbshell

====================================================
*                  MapR-DB Shell                   *
* NOTE: This is a shell for JSON table operations. *
====================================================
Version: 6.1.0-mapr

MapR-DB Shell

maprdb mapr:> ls /user/mapr/tables/akka/
Found 17 items
tr--------   - mapr 5000          3 2019-08-19 23:46 /user/mapr/tables/akka/ids
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/non-existing-pid.journal
tr--------   - mapr 5000          3 2019-08-19 23:43 /user/mapr/tables/akka/p-1.journal
tr--------   - mapr 5000          3 2019-08-19 23:46 /user/mapr/tables/akka/p-10.journal
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/p-11.journal
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/p-12.journal
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/p-13.journal
tr--------   - mapr 5000          3 2019-08-19 23:46 /user/mapr/tables/akka/p-14.journal
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/p-15.journal
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/p-2.journal
tr--------   - mapr 5000          3 2019-08-19 23:37 /user/mapr/tables/akka/p-3.journal
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/p-4.journal
tr--------   - mapr 5000          3 2019-08-19 23:43 /user/mapr/tables/akka/p-5.journal
tr--------   - mapr 5000          3 2019-08-19 23:46 /user/mapr/tables/akka/p-6.journal
tr--------   - mapr 5000          3 2019-08-19 23:41 /user/mapr/tables/akka/p-7.journal
tr--------   - mapr 5000          3 2019-08-19 23:37 /user/mapr/tables/akka/p-8.journal

The previous example shows the journal tables and ids tables.

maprdb mapr:> find /user/mapr/tables/akka/ids

{"_id":"non-existing-pid","path":"/user/mapr/tables/akka/non-existing-pid.journal"}
{"_id":"p-1","path":"/user/mapr/tables/akka/p-1.journal"}
{"_id":"p-10","path":"/user/mapr/tables/akka/p-10.journal"}
{"_id":"p-11","path":"/user/mapr/tables/akka/p-11.journal"}
{"_id":"p-12","path":"/user/mapr/tables/akka/p-12.journal"}
{"_id":"p-13","path":"/user/mapr/tables/akka/p-13.journal"}
{"_id":"p-14","path":"/user/mapr/tables/akka/p-14.journal"}
{"_id":"p-15","path":"/user/mapr/tables/akka/p-15.journal"}
{"_id":"p-2","path":"/user/mapr/tables/akka/p-2.journal"}
{"_id":"p-3","path":"/user/mapr/tables/akka/p-3.journal"}
{"_id":"p-4","path":"/user/mapr/tables/akka/p-4.journal"}
{"_id":"p-5","path":"/user/mapr/tables/akka/p-5.journal"}
{"_id":"p-6","path":"/user/mapr/tables/akka/p-6.journal"}
{"_id":"p-7","path":"/user/mapr/tables/akka/p-7.journal"}
{"_id":"p-8","path":"/user/mapr/tables/akka/p-8.journal"}
{"_id":"p-9","path":"/user/mapr/tables/akka/p-9.journal"}

Notice that the ids table has each persistence entity id along with the path where it lives.

Now let's inspect one of the journals.

maprdb mapr:> find /user/mapr/tables/akka/p-1.journal

{"_id":{"$binary":"AQ=="},"deleted":false,"persistentRepr":{"$binary":"Cg4IARIKrO0ABXQAA2EtMRABGgNwLTFqJDdjYmUxNmQ4LWY1ZTktNGEyYy04ZjZiLWFlMWNmNWQxNzZkZg=="}}
{"_id":{"$binary":"Ag=="},"deleted":false,"persistentRepr":{"$binary":"Cg4IARIKrO0ABXQAA2EtMhACGgNwLTFqJDdjYmUxNmQ4LWY1ZTktNGEyYy04ZjZiLWFlMWNmNWQxNzZkZg=="}}
{"_id":{"$binary":"Aw=="},"deleted":false,"persistentRepr":{"$binary":"Cg4IARIKrO0ABXQAA2EtMxADGgNwLTFqJDdjYmUxNmQ4LWY1ZTktNGEyYy04ZjZiLWFlMWNmNWQxNzZkZg=="}}
{"_id":{"$binary":"BA=="},"deleted":false,"persistentRepr":{"$binary":"Cg4IARIKrO0ABXQAA2EtNBAEGgNwLTFqJDdjYmUxNmQ4LWY1ZTktNGEyYy04ZjZiLWFlMWNmNWQxNzZkZg=="}}
{"_id":{"$binary":"BQ=="},"deleted":false,"persistentRepr":{"$binary":"Cg4IARIKrO0ABXQAA2EtNRAFGgNwLTFqJDdjYmUxNmQ4LWY1ZTktNGEyYy04ZjZiLWFlMWNmNWQxNzZkZg=="}}
5 document(s) found.

The previous example shows the content of p-1.journal. Notice that the structure matches to the journal's structure we explained before.

Journal Tests

All tests for the journal passed. However, since we don't have a MapR Cluster in Travis we are going to ignore the test. One can run the test locally against a configured MapR Cluster

Tests Output

[info] MyJournalSpec:
[info] A journal
[info] - must replay all messages
[info] - must replay messages using a lower sequence number bound
[info] - must replay messages using an upper sequence number bound
[info] - must replay messages using a count limit
[info] - must replay messages using a lower and upper sequence number bound
[info] - must replay messages using a lower and upper sequence number bound and a count limit
[info] - must replay a single if lower sequence number bound equals upper sequence number bound
[info] - must replay a single message if count limit equals 1
[info] - must not replay messages if count limit equals 0
[info] - must not replay messages if lower  sequence number bound is greater than upper sequence number bound
[info] - must not replay messages if the persistent actor has not yet written messages
[info] - must not replay permanently deleted messages (range deletion)
[info] - must not reset highestSequenceNr after message deletion
[info] - must not reset highestSequenceNr after journal cleanup
[info] A Journal optionally
[info] + CapabilityFlag `supportsRejectingNonSerializableObjects` was turned `off`. To enable the related tests override it with `CapabilityFlag.on` (or `true` in Scala). 
[info] + CapabilityFlag `supportsSerialization` was turned `on`.  
[info] - may serialize events
[info] Run completed in 52 seconds, 904 milliseconds.
[info] Total number of tests run: 15
[info] Suites: completed 1, aborted 0
[info] Tests: succeeded 15, failed 0, canceled 0, ignored 0, pending 0
[info] All tests passed.
[success] Total time: 63 s, completed Aug 19, 2019 2:02:01 AM

Query Side

The current version supports Persistence Read Side. The following queries have been added.

  • currentPersistenceIds() gives back the persistence ids in a bounded stream that is closed after completion
  • persistenceIds() gives back the persistence ids in an unbounded stream that keeps open. New persistence ids will be pushed into this stream.
  • currentEventsByPersistenceId(...) queries events of the especified persistence entity id.
  • eventsByPersistenceId(...) is the stream of events of the especified persistence entity id. New events will be pushed into this stream as it keeps live.
object QueryExample extends App {

  implicit val system = ActorSystem("example")

  implicit val mat = ActorMaterializer()


  val readJournal =
    PersistenceQuery(system).readJournalFor[MapRDBScalaReadJournal]("akka.persistence.query.journal")

  val events = readJournal.currentEventsByPersistenceId("p-1", 3, Long.MaxValue)
  
  val liveEvents = readJournal.eventsByPersistenceId("p-1", 3, Long.MaxValue)

  val boundedStream = readJournal.currentPersistenceIds().runForeach(println)

  val unboundedStream = readJournal.persistenceIds().runForeach(println)

  Await.result(boundedStream, scala.concurrent.duration.Duration.Inf)
  Await.result(unboundedStream, scala.concurrent.duration.Duration.Inf)
  Await.result(events.runForeach(println), scala.concurrent.duration.Duration.Inf)
  Await.result(liveEvents.runForeach(println), scala.concurrent.duration.Duration.Inf)
}

Tagging Events and Events by Tags

In general, Tagged events are just a especial kind of events. All along you could send a Tagged event to the persistence entity and it ends up being saved in the corresponding Journal for the entity. Queries that are based on events by persistence id will retrieve all the events of the entity (including any Tagged event) back to you.

However, querying Tagged events across multiple persistence entities was not possible before version 1.0.6.

Since version 1.0.6 we added support for querying Tagged events.

Crating a tag event

The following Actor code shows how we create Tagged events.

       final case class ExampleState(received: List[String] = Nil) {
           def updated(s: String): ExampleState = copy(s :: received)
       
           override def toString = received.reverse.toString
         }
       
         class ExamplePersistentActor extends PersistentActor {
           override def persistenceId: String = "sample-id-3"
       
           var state = ExampleState()
       
           override def receiveCommand: Receive = {
             case "print" => println("current state = " + state)
             case "snap" => saveSnapshot(state)
             case SaveSnapshotSuccess(metadata) => // ...
             case SaveSnapshotFailure(metadata, reason) => // ...
             case s: String =>
               persist(s) { evt => state = state.updated(evt) }
       
             case (s: String, tags: Seq[String]) =>
               persist(Tagged(s, tags.toSet)) { evt => state = state.updated(evt.payload.toString) }
           }
       
           override def receiveRecover: Receive = {
             case SnapshotOffer(_, s: ExampleState) =>
               println("offered state = " + s)
               state = s
             case evt: String =>
               state = state.updated(evt)
           }
         }

Notice the case (s: String, tags: Seq[String]) since it will create a Tagged event and send it to the Journal to be persisted.

The following code uses the defined Actor to create Tagged events.

val system = ActorSystem("example")
val persistentActor = system.actorOf(Props(classOf[ExamplePersistentActor]), "persistentActor-3-scala")

persistentActor ! ("nick", Seq("boy", "30"))

Querying Tagged Events

The idea of Tagged events is that they can be queried across multiple persistence entities. In other words, when retrieving events with a specific tag, they might not all com from the same persistence entity id, any event in the entire system that was tagged with the given tag will be retrieved.

The following code shows how to retrieve Tagged events.

val readJournal =
    PersistenceQuery(system).readJournalFor[MapRDBScalaReadJournal]("akka.persistence.query.journal")

val currentEventsByTag = readJournal.currentEventsByTag("boy", Offset.noOffset).runForeach(println)
  
readJournal.eventsByTag("", Offset.sequence(5)).runWith(Sink.asPublisher(publisher))
  • currentEventsByTag is the sequence of current Tagged event by the given tag.
  • eventsByTag is an open stream of Tagged event by the given tag that is kept alive waiting for more events in live fashion scenario.

Ordering and Offsets

It is important to describe the semantics offered by the library in regard of ordering and offsets for Tagged events.

Tagged events are sorted by the timestamp at the moment they are written by the Journal so they are retrieved in the exact same order. In highly async contexts, some Tagged events might take longer to persist than others. However, the sorting happens using the creation timestamp and it does not matter in what order they are actually written to the storage.

When retrieving Tagged events, the only supported Offset is Offset.sequence any other being passed will be treated as Offset.noOffset meaning that the offset will start at Offset.sequence(0). Since we are retrieving events by tag, the given offset refers to the offset of the Tagged stream after the events for the given tag is calculated.

Let's look at an example that might clarify how it works.

taggedEvents: [(t1, e1), (t2, e1), (t1, e2), (t3, e3), (t1, e4), (t1, e5)]

Notice that tag t1 has events e1, e2, e4 and e5 associated to it.

Now, if we want the events associated with tag t1 and Offset.noOffset, the events e1, e2, e4 and e5 are retrieved since offset was effectively 0.

Changing the query by moving the offset will result in a different set of events. Using Offset.sequence(2) for instance, will return only events e4 and e5 since for the given tag t1 it is skipping the first 2 events.

It is important to notice that Offsets in the events from these queries are stable. In the previous example where Offset.noOffset, e1 will have Offset 0 and e5 will have Offset 3. When Offset.sequence(2), e4 will have Offset 2 and e5 will have Offset 3.

In general, the offset is applied to the resultant stream of events associated with the given tag. Notice that the given offsets is applied only once, meaning that is your query is eventsByTag it will move the cursor the offset, but all new events after the offset will be pushed in the stream as they arrive.

MapR-DB Tagged Structure.

The internal structure being used for storing Tagged events is very simplistic. However, the number of Tagged events can be overwhelming. Even though, MapR-DB is insanely fast for the queries being used, it might be a good idea to create a secondary index on taggedEvents table. The following command shows how it can be done, but the MapR Control System can also be used.

maprcli table index add -path /user/mapr/tables/akka/taggedEvents -index tagged_events_idx -indexedfields tags  

Notice that the based path /user/mapr/tables/akka/ is where you persistence entities live (configurable) and the taggedEvents in the table where Tagged events are stored.

MapR-DB Performance

It is known that MapR-DB is quite faster than Cassandra read paper here while keeping hight consistency with zero tunning. The following ESG Lab Analysis shows that MapR-DB yielded an average performance improvement of 2.5x more operations/sec than Cassandra and 5.5x more than HBase.

akka-persistence-maprdb plugin has been designed to take advantage of the especific contructs that MapR-DB offers in order to yield supperior performance. There is no need to for tunning other than the explained in MapR-DB Tagged Structure