Find file History
Latest commit 3ba950b Jan 11, 2017 @jsternberg jsternberg Fix for subqueries to use the parallel iterator correctly
Also, fix the `Iterators.Merge(IteratorOptions)` function so it consults
the `Ordered` attribute to determine which iterator it should use to
merge the input iterators.
Permalink
..
Failed to load latest commit information.
internal Support subquery execution in the query language Jan 7, 2017
neldermead Update godoc for package influxql Dec 31, 2016
README.md Support subquery execution in the query language Jan 7, 2017
ast.go Support subquery execution in the query language Jan 7, 2017
ast_test.go Support subquery execution in the query language Jan 7, 2017
call_iterator.go Support subquery execution in the query language Jan 7, 2017
call_iterator_test.go Add sample function to query language Oct 6, 2016
cast.go Cast number fill values for the fill iterator Feb 11, 2016
doc.go Update godoc for package influxql Dec 31, 2016
emitter.go Support subquery execution in the query language Jan 7, 2017
emitter_test.go Introduce syntax for marking a partial response with chunking Nov 22, 2016
functions.gen.go Support subquery execution in the query language Jan 7, 2017
functions.gen.go.tmpl Support subquery execution in the query language Jan 7, 2017
functions.go Removed ineffective assignments, and added checks for errors that pre… Dec 29, 2016
functions_test.go Use local RNG in SampleReducer Dec 15, 2016
influxql.go Refactored call iterators to make them public and more usable as a li… Mar 2, 2016
iterator.gen.go Support subquery execution in the query language Jan 7, 2017
iterator.gen.go.tmpl Support subquery execution in the query language Jan 7, 2017
iterator.go Fix for subqueries to use the parallel iterator correctly Jan 11, 2017
iterator_mapper.go Support subquery execution in the query language Jan 7, 2017
iterator_mapper_test.go Support subquery execution in the query language Jan 7, 2017
iterator_test.go Support subquery execution in the query language Jan 7, 2017
linear.go Add fill(linear) to query language Oct 4, 2016
monitor.go Add config option to messages when limits exceeded Oct 28, 2016
parser.go Support subquery execution in the query language Jan 7, 2017
parser_test.go Support subquery execution in the query language Jan 7, 2017
point.gen.go Support subquery execution in the query language Jan 7, 2017
point.gen.go.tmpl Support subquery execution in the query language Jan 7, 2017
point.go Support subquery execution in the query language Jan 7, 2017
point_test.go Optimize the mean() call by moving the calculation into the shard ite… Feb 16, 2016
query_executor.go Use proper uber-go/zap import path Dec 15, 2016
query_executor_test.go Resolved PR comments on test files Dec 30, 2016
result.go Introduce syntax for marking a partial response with chunking Nov 22, 2016
sanitize.go Improve query sanitization to prevent a password leak in the logs Apr 22, 2016
sanitize_test.go Improve query sanitization to prevent a password leak in the logs Apr 22, 2016
scanner.go Update godoc for package influxql Dec 31, 2016
scanner_test.go Fix the dollar sign so it properly handles reserved keywords Sep 2, 2016
select.go Support subquery execution in the query language Jan 7, 2017
select_test.go Support subquery execution in the query language Jan 7, 2017
statement_rewriter.go Support the ON syntax in SHOW TAG VALUES Nov 1, 2016
statement_rewriter_test.go Support ON and use default database for SHOW commands Sep 13, 2016
task_manager.go Use proper uber-go/zap import path Dec 15, 2016
tmpldata remove NaN from query engine Feb 17, 2016
token.go remove IF EXISTS/IF NOT EXISTS from influxql language Jul 29, 2016

README.md

The Influx Query Language Specification

Introduction

This is a reference for the Influx Query Language ("InfluxQL").

InfluxQL is a SQL-like query language for interacting with InfluxDB. It has been lovingly crafted to feel familiar to those coming from other SQL or SQL-like environments while providing features specific to storing and analyzing time series data.

Notation

The syntax is specified using Extended Backus-Naur Form ("EBNF"). EBNF is the same notation used in the Go programming language specification, which can be found here. Not so coincidentally, InfluxDB is written in Go.

Production  = production_name "=" [ Expression ] "." .
Expression  = Alternative { "|" Alternative } .
Alternative = Term { Term } .
Term        = production_name | token [ "…" token ] | Group | Option | Repetition .
Group       = "(" Expression ")" .
Option      = "[" Expression "]" .
Repetition  = "{" Expression "}" .

Notation operators in order of increasing precedence:

|   alternation
()  grouping
[]  option (0 or 1 times)
{}  repetition (0 to n times)

Query representation

Characters

InfluxQL is Unicode text encoded in UTF-8.

newline             = /* the Unicode code point U+000A */ .
unicode_char        = /* an arbitrary Unicode code point except newline */ .

Letters and digits

Letters are the set of ASCII characters plus the underscore character _ (U+005F) is considered a letter.

Only decimal digits are supported.

letter              = ascii_letter | "_" .
ascii_letter        = "A" … "Z" | "a" … "z" .
digit               = "0" … "9" .

Identifiers

Identifiers are tokens which refer to database names, retention policy names, user names, measurement names, tag keys, and field keys.

The rules:

  • double quoted identifiers can contain any unicode character other than a new line
  • double quoted identifiers can contain escaped " characters (i.e., \")
  • double quoted identifiers can contain InfluxQL keywords
  • unquoted identifiers must start with an upper or lowercase ASCII character or "_"
  • unquoted identifiers may contain only ASCII letters, decimal digits, and "_"
identifier          = unquoted_identifier | quoted_identifier .
unquoted_identifier = ( letter ) { letter | digit } .
quoted_identifier   = `"` unicode_char { unicode_char } `"` .

Examples:

cpu
_cpu_stats
"1h"
"anything really"
"1_Crazy-1337.identifier>NAME👍"

Keywords

ALL           ALTER         ANY           AS            ASC           BEGIN
BY            CREATE        CONTINUOUS    DATABASE      DATABASES     DEFAULT
DELETE        DESC          DESTINATIONS  DIAGNOSTICS   DISTINCT      DROP
DURATION      END           EVERY         EXPLAIN       FIELD         FOR
FROM          GRANT         GRANTS        GROUP         GROUPS        IN
INF           INSERT        INTO          KEY           KEYS          KILL
LIMIT         SHOW          MEASUREMENT   MEASUREMENTS  NAME          OFFSET
ON            ORDER         PASSWORD      POLICY        POLICIES      PRIVILEGES
QUERIES       QUERY         READ          REPLICATION   RESAMPLE      RETENTION
REVOKE        SELECT        SERIES        SET           SHARD         SHARDS
SLIMIT        SOFFSET       STATS         SUBSCRIPTION  SUBSCRIPTIONS TAG
TO            USER          USERS         VALUES        WHERE         WITH
WRITE

Literals

Integers

InfluxQL supports decimal integer literals. Hexadecimal and octal literals are not currently supported.

int_lit             = ( "1" … "9" ) { digit } .

Floats

InfluxQL supports floating-point literals. Exponents are not currently supported.

float_lit           = int_lit "." int_lit .

Strings

String literals must be surrounded by single quotes. Strings may contain ' characters as long as they are escaped (i.e., \').

string_lit          = `'` { unicode_char } `'` .

Durations

Duration literals specify a length of time. An integer literal followed immediately (with no spaces) by a duration unit listed below is interpreted as a duration literal.

Duration units

Units Meaning
u or µ microseconds (1 millionth of a second)
ms milliseconds (1 thousandth of a second)
s second
m minute
h hour
d day
w week
duration_lit        = int_lit duration_unit .
duration_unit       = "u" | "µ" | "ms" | "s" | "m" | "h" | "d" | "w" .

Dates & Times

The date and time literal format is not specified in EBNF like the rest of this document. It is specified using Go's date / time parsing format, which is a reference date written in the format required by InfluxQL. The reference date time is:

InfluxQL reference date time: January 2nd, 2006 at 3:04:05 PM

time_lit            = "2006-01-02 15:04:05.999999" | "2006-01-02" .

Booleans

bool_lit            = TRUE | FALSE .

Regular Expressions

regex_lit           = "/" { unicode_char } "/" .

Comparators: =~ matches against !~ doesn't match against

Note: Use regular expressions to match measurements and tags. You cannot use regular expressions to match databases, retention policies, or fields.

Queries

A query is composed of one or more statements separated by a semicolon.

query               = statement { ";" statement } .

statement           = alter_retention_policy_stmt |
                      create_continuous_query_stmt |
                      create_database_stmt |
                      create_retention_policy_stmt |
                      create_subscription_stmt |
                      create_user_stmt |
                      delete_stmt |
                      drop_continuous_query_stmt |
                      drop_database_stmt |
                      drop_measurement_stmt |
                      drop_retention_policy_stmt |
                      drop_series_stmt |
                      drop_shard_stmt |
                      drop_subscription_stmt |
                      drop_user_stmt |
                      grant_stmt |
                      kill_query_statement |
                      show_continuous_queries_stmt |
                      show_databases_stmt |
                      show_field_keys_stmt |
                      show_grants_stmt |
                      show_measurements_stmt |
                      show_queries_stmt |
                      show_retention_policies |
                      show_series_stmt |
                      show_shard_groups_stmt |
                      show_shards_stmt |
                      show_subscriptions_stmt|
                      show_tag_keys_stmt |
                      show_tag_values_stmt |
                      show_users_stmt |
                      revoke_stmt |
                      select_stmt .

Statements

ALTER RETENTION POLICY

alter_retention_policy_stmt  = "ALTER RETENTION POLICY" policy_name on_clause
                               retention_policy_option
                               [ retention_policy_option ]
                               [ retention_policy_option ]
                               [ retention_policy_option ] .

Replication factors do not serve a purpose with single node instances.

Examples:

-- Set default retention policy for mydb to 1h.cpu.
ALTER RETENTION POLICY "1h.cpu" ON "mydb" DEFAULT

-- Change duration and replication factor.
ALTER RETENTION POLICY "policy1" ON "somedb" DURATION 1h REPLICATION 4

CREATE CONTINUOUS QUERY

create_continuous_query_stmt = "CREATE CONTINUOUS QUERY" query_name on_clause
                               [ "RESAMPLE" resample_opts ]
                               "BEGIN" select_stmt "END" .

query_name                   = identifier .

resample_opts                = (every_stmt for_stmt | every_stmt | for_stmt) .
every_stmt                   = "EVERY" duration_lit
for_stmt                     = "FOR" duration_lit

Examples:

-- selects from DEFAULT retention policy and writes into 6_months retention policy
CREATE CONTINUOUS QUERY "10m_event_count"
ON "db_name"
BEGIN
  SELECT count("value")
  INTO "6_months"."events"
  FROM "events"
  GROUP BY time(10m)
END;

-- this selects from the output of one continuous query in one retention policy and outputs to another series in another retention policy
CREATE CONTINUOUS QUERY "1h_event_count"
ON "db_name"
BEGIN
  SELECT sum("count") as "count"
  INTO "2_years"."events"
  FROM "6_months"."events"
  GROUP BY time(1h)
END;

-- this customizes the resample interval so the interval is queried every 10s and intervals are resampled until 2m after their start time
-- when resample is used, at least one of "EVERY" or "FOR" must be used
CREATE CONTINUOUS QUERY "cpu_mean"
ON "db_name"
RESAMPLE EVERY 10s FOR 2m
BEGIN
  SELECT mean("value")
  INTO "cpu_mean"
  FROM "cpu"
  GROUP BY time(1m)
END;

CREATE DATABASE

create_database_stmt = "CREATE DATABASE" db_name
                       [ WITH
                           [ retention_policy_duration ]
                           [ retention_policy_replication ]
                           [ retention_policy_shard_group_duration ]
                           [ retention_policy_name ]
                       ] .

Replication factors do not serve a purpose with single node instances.

Examples:

-- Create a database called foo
CREATE DATABASE "foo"

-- Create a database called bar with a new DEFAULT retention policy and specify the duration, replication, shard group duration, and name of that retention policy
CREATE DATABASE "bar" WITH DURATION 1d REPLICATION 1 SHARD DURATION 30m NAME "myrp"

-- Create a database called mydb with a new DEFAULT retention policy and specify the name of that retention policy
CREATE DATABASE "mydb" WITH NAME "myrp"

CREATE RETENTION POLICY

create_retention_policy_stmt = "CREATE RETENTION POLICY" policy_name on_clause
                               retention_policy_duration
                               retention_policy_replication
                               [ retention_policy_shard_group_duration ]
                               [ "DEFAULT" ] .

Replication factors do not serve a purpose with single node instances.

Examples

-- Create a retention policy.
CREATE RETENTION POLICY "10m.events" ON "somedb" DURATION 60m REPLICATION 2

-- Create a retention policy and set it as the DEFAULT.
CREATE RETENTION POLICY "10m.events" ON "somedb" DURATION 60m REPLICATION 2 DEFAULT

-- Create a retention policy and specify the shard group duration.
CREATE RETENTION POLICY "10m.events" ON "somedb" DURATION 60m REPLICATION 2 SHARD DURATION 30m

CREATE SUBSCRIPTION

Subscriptions tell InfluxDB to send all the data it receives to Kapacitor or other third parties.

create_subscription_stmt = "CREATE SUBSCRIPTION" subscription_name "ON" db_name "." retention_policy "DESTINATIONS" ("ANY"|"ALL") host { "," host} .

Examples:

-- Create a SUBSCRIPTION on database 'mydb' and retention policy 'autogen' that send data to 'example.com:9090' via UDP.
CREATE SUBSCRIPTION "sub0" ON "mydb"."autogen" DESTINATIONS ALL 'udp://example.com:9090'

-- Create a SUBSCRIPTION on database 'mydb' and retention policy 'autogen' that round robins the data to 'h1.example.com:9090' and 'h2.example.com:9090'.
CREATE SUBSCRIPTION "sub0" ON "mydb"."autogen" DESTINATIONS ANY 'udp://h1.example.com:9090', 'udp://h2.example.com:9090'

CREATE USER

create_user_stmt = "CREATE USER" user_name "WITH PASSWORD" password
                   [ "WITH ALL PRIVILEGES" ] .

Examples:

-- Create a normal database user.
CREATE USER "jdoe" WITH PASSWORD '1337password'

-- Create an admin user.
-- Note: Unlike the GRANT statement, the "PRIVILEGES" keyword is required here.
CREATE USER "jdoe" WITH PASSWORD '1337password' WITH ALL PRIVILEGES

Note: The password string must be wrapped in single quotes.

DELETE

delete_stmt = "DELETE" ( from_clause | where_clause | from_clause where_clause ) .

Examples:

DELETE FROM "cpu"
DELETE FROM "cpu" WHERE time < '2000-01-01T00:00:00Z'
DELETE WHERE time < '2000-01-01T00:00:00Z'

DROP CONTINUOUS QUERY

drop_continuous_query_stmt = "DROP CONTINUOUS QUERY" query_name on_clause .

Example:

DROP CONTINUOUS QUERY "myquery" ON "mydb"

DROP DATABASE

drop_database_stmt = "DROP DATABASE" db_name .

Example:

DROP DATABASE "mydb"

DROP MEASUREMENT

drop_measurement_stmt = "DROP MEASUREMENT" measurement .

Examples:

-- drop the cpu measurement
DROP MEASUREMENT "cpu"

DROP RETENTION POLICY

drop_retention_policy_stmt = "DROP RETENTION POLICY" policy_name on_clause .

Example:

-- drop the retention policy named 1h.cpu from mydb
DROP RETENTION POLICY "1h.cpu" ON "mydb"

DROP SERIES

drop_series_stmt = "DROP SERIES" ( from_clause | where_clause | from_clause where_clause ) .

Example:

DROP SERIES FROM "telegraf"."autogen"."cpu" WHERE cpu = 'cpu8'

DROP SHARD

drop_shard_stmt = "DROP SHARD" ( shard_id ) .

Example:

DROP SHARD 1

DROP SUBSCRIPTION

drop_subscription_stmt = "DROP SUBSCRIPTION" subscription_name "ON" db_name "." retention_policy .

Example:

DROP SUBSCRIPTION "sub0" ON "mydb"."autogen"

DROP USER

drop_user_stmt = "DROP USER" user_name .

Example:

DROP USER "jdoe"

GRANT

NOTE: Users can be granted privileges on databases that do not exist.

grant_stmt = "GRANT" privilege [ on_clause ] to_clause .

Examples:

-- grant admin privileges
GRANT ALL TO "jdoe"

-- grant read access to a database
GRANT READ ON "mydb" TO "jdoe"

KILL QUERY

kill_query_statement = "KILL QUERY" query_id .

Examples:

--- kill a query with the query_id 36
KILL QUERY 36

NOTE: Identify the query_id from the SHOW QUERIES output.

SHOW CONTINUOUS QUERIES

show_continuous_queries_stmt = "SHOW CONTINUOUS QUERIES" .

Example:

-- show all continuous queries
SHOW CONTINUOUS QUERIES

SHOW DATABASES

show_databases_stmt = "SHOW DATABASES" .

Example:

-- show all databases
SHOW DATABASES

SHOW FIELD KEYS

show_field_keys_stmt = "SHOW FIELD KEYS" [ from_clause ] .

Examples:

-- show field keys and field value data types from all measurements
SHOW FIELD KEYS

-- show field keys and field value data types from specified measurement
SHOW FIELD KEYS FROM "cpu"

SHOW GRANTS

show_grants_stmt = "SHOW GRANTS FOR" user_name .

Example:

-- show grants for jdoe
SHOW GRANTS FOR "jdoe"

SHOW MEASUREMENTS

show_measurements_stmt = "SHOW MEASUREMENTS" [ with_measurement_clause ] [ where_clause ] [ limit_clause ] [ offset_clause ] .

Examples:

-- show all measurements
SHOW MEASUREMENTS

-- show measurements where region tag = 'uswest' AND host tag = 'serverA'
SHOW MEASUREMENTS WHERE "region" = 'uswest' AND "host" = 'serverA'

-- show measurements that start with 'h2o'
SHOW MEASUREMENTS WITH MEASUREMENT =~ /h2o.*/

SHOW QUERIES

show_queries_stmt = "SHOW QUERIES" .

Example:

-- show all currently-running queries
SHOW QUERIES

SHOW RETENTION POLICIES

show_retention_policies = "SHOW RETENTION POLICIES" on_clause .

Example:

-- show all retention policies on a database
SHOW RETENTION POLICIES ON "mydb"

SHOW SERIES

show_series_stmt = "SHOW SERIES" [ from_clause ] [ where_clause ] [ limit_clause ] [ offset_clause ] .

Example:

SHOW SERIES FROM "telegraf"."autogen"."cpu" WHERE cpu = 'cpu8'

SHOW SHARD GROUPS

show_shard_groups_stmt = "SHOW SHARD GROUPS" .

Example:

SHOW SHARD GROUPS

SHOW SHARDS

show_shards_stmt = "SHOW SHARDS" .

Example:

SHOW SHARDS

SHOW SUBSCRIPTIONS

show_subscriptions_stmt = "SHOW SUBSCRIPTIONS" .

Example:

SHOW SUBSCRIPTIONS

SHOW TAG KEYS

show_tag_keys_stmt = "SHOW TAG KEYS" [ from_clause ] [ where_clause ] [ group_by_clause ]
                     [ limit_clause ] [ offset_clause ] .

Examples:

-- show all tag keys
SHOW TAG KEYS

-- show all tag keys from the cpu measurement
SHOW TAG KEYS FROM "cpu"

-- show all tag keys from the cpu measurement where the region key = 'uswest'
SHOW TAG KEYS FROM "cpu" WHERE "region" = 'uswest'

-- show all tag keys where the host key = 'serverA'
SHOW TAG KEYS WHERE "host" = 'serverA'

SHOW TAG VALUES

show_tag_values_stmt = "SHOW TAG VALUES" [ from_clause ] with_tag_clause [ where_clause ]
                       [ group_by_clause ] [ limit_clause ] [ offset_clause ] .

Examples:

-- show all tag values across all measurements for the region tag
SHOW TAG VALUES WITH KEY = "region"

-- show tag values from the cpu measurement for the region tag
SHOW TAG VALUES FROM "cpu" WITH KEY = "region"

-- show tag values across all measurements for all tag keys that do not include the letter c
SHOW TAG VALUES WITH KEY !~ /.*c.*/

-- show tag values from the cpu measurement for region & host tag keys where service = 'redis'
SHOW TAG VALUES FROM "cpu" WITH KEY IN ("region", "host") WHERE "service" = 'redis'

SHOW USERS

show_users_stmt = "SHOW USERS" .

Example:

-- show all users
SHOW USERS

REVOKE

revoke_stmt = "REVOKE" privilege [ on_clause ] "FROM" user_name .

Examples:

-- revoke admin privileges from jdoe
REVOKE ALL PRIVILEGES FROM "jdoe"

-- revoke read privileges from jdoe on mydb
REVOKE READ ON "mydb" FROM "jdoe"

SELECT

select_stmt = "SELECT" fields from_clause [ into_clause ] [ where_clause ]
              [ group_by_clause ] [ order_by_clause ] [ limit_clause ]
              [ offset_clause ] [ slimit_clause ] [ soffset_clause ] .

Examples:

-- select mean value from the cpu measurement where region = 'uswest' grouped by 10 minute intervals
SELECT mean("value") FROM "cpu" WHERE "region" = 'uswest' GROUP BY time(10m) fill(0)

-- select from all measurements beginning with cpu into the same measurement name in the cpu_1h retention policy
SELECT mean("value") INTO "cpu_1h".:MEASUREMENT FROM /cpu.*/

Clauses

from_clause     = "FROM" measurements .

group_by_clause = "GROUP BY" dimensions fill(fill_option).

into_clause     = "INTO" ( measurement | back_ref ).

limit_clause    = "LIMIT" int_lit .

offset_clause   = "OFFSET" int_lit .

slimit_clause   = "SLIMIT" int_lit .

soffset_clause  = "SOFFSET" int_lit .

on_clause       = "ON" db_name .

order_by_clause = "ORDER BY" sort_fields .

to_clause       = "TO" user_name .

where_clause    = "WHERE" expr .

with_measurement_clause = "WITH MEASUREMENT" ( "=" measurement | "=~" regex_lit ) .

with_tag_clause = "WITH KEY" ( "=" tag_key | "!=" tag_key | "=~" regex_lit | "IN (" tag_keys ")"  ) .

Expressions

binary_op        = "+" | "-" | "*" | "/" | "AND" | "OR" | "=" | "!=" | "<>" | "<" |
                   "<=" | ">" | ">=" .

expr             = unary_expr { binary_op unary_expr } .

unary_expr       = "(" expr ")" | var_ref | time_lit | string_lit | int_lit |
                   float_lit | bool_lit | duration_lit | regex_lit .

Other

alias            = "AS" identifier .

back_ref         = ( policy_name ".:MEASUREMENT" ) |
                   ( db_name "." [ policy_name ] ".:MEASUREMENT" ) .

db_name          = identifier .

dimension        = expr .

dimensions       = dimension { "," dimension } .

field_key        = identifier .

field            = expr [ alias ] .

fields           = field { "," field } .

fill_option      = "null" | "none" | "previous" | "linear" | int_lit | float_lit .

host             = string_lit .

measurement      = measurement_name |
                   ( policy_name "." measurement_name ) |
                   ( db_name "." [ policy_name ] "." measurement_name ) .

measurements     = measurement { "," measurement } .

measurement_name = identifier | regex_lit .

password         = string_lit .

policy_name      = identifier .

privilege        = "ALL" [ "PRIVILEGES" ] | "READ" | "WRITE" .

query_id         = int_lit .

query_name       = identifier .

retention_policy = identifier .

retention_policy_option      = retention_policy_duration |
                               retention_policy_replication |
                               retention_policy_shard_group_duration |
                               "DEFAULT" .

retention_policy_duration    = "DURATION" duration_lit .

retention_policy_replication = "REPLICATION" int_lit .

retention_policy_shard_group_duration = "SHARD DURATION" duration_lit .

retention_policy_name = "NAME" identifier .

series_id        = int_lit .

shard_id         = int_lit .

sort_field       = field_key [ ASC | DESC ] .

sort_fields      = sort_field { "," sort_field } .

subscription_name = identifier .

tag_key          = identifier .

tag_keys         = tag_key { "," tag_key } .

user_name        = identifier .

var_ref          = measurement .

Query Engine Internals

Once you understand the language itself, it's important to know how these language constructs are implemented in the query engine. This gives you an intuitive sense for how results will be processed and how to create efficient queries.

The life cycle of a query looks like this:

  1. InfluxQL query string is tokenized and then parsed into an abstract syntax tree (AST). This is the code representation of the query itself.

  2. The AST is passed to the QueryExecutor which directs queries to the appropriate handlers. For example, queries related to meta data are executed by the meta service and SELECT statements are executed by the shards themselves.

  3. The query engine then determines the shards that match the SELECT statement's time range. From these shards, iterators are created for each field in the statement.

  4. Iterators are passed to the emitter which drains them and joins the resulting points. The emitter's job is to convert simple time/value points into the more complex result objects that are returned to the client.

Understanding Iterators

Iterators are at the heart of the query engine. They provide a simple interface for looping over a set of points. For example, this is an iterator over Float points:

type FloatIterator interface {
    Next() (*FloatPoint, error)
}

These iterators are created through the IteratorCreator interface:

type IteratorCreator interface {
    CreateIterator(m *Measurement, opt IteratorOptions) (Iterator, error)
}

The IteratorOptions provide arguments about field selection, time ranges, and dimensions that the iterator creator can use when planning an iterator. The IteratorCreator interface is used at many levels such as the Shards, Shard, and Engine. This allows optimizations to be performed when applicable such as returning a precomputed COUNT().

Iterators aren't just for reading raw data from storage though. Iterators can be composed so that they provided additional functionality around an input iterator. For example, a DistinctIterator can compute the distinct values for each time window for an input iterator. Or a FillIterator can generate additional points that are missing from an input iterator.

This composition also lends itself well to aggregation. For example, a statement such as this:

SELECT MEAN(value) FROM cpu GROUP BY time(10m)

In this case, MEAN(value) is a MeanIterator wrapping an iterator from the underlying shards. However, if we can add an additional iterator to determine the derivative of the mean:

SELECT DERIVATIVE(MEAN(value), 20m) FROM cpu GROUP BY time(10m)

Understanding Auxiliary Fields

Because InfluxQL allows users to use selector functions such as FIRST(), LAST(), MIN(), and MAX(), the engine must provide a way to return related data at the same time with the selected point.

For example, in this query:

SELECT FIRST(value), host FROM cpu GROUP BY time(1h)

We are selecting the first value that occurs every hour but we also want to retrieve the host associated with that point. Since the Point types only specify a single typed Value for efficiency, we push the host into the auxiliary fields of the point. These auxiliary fields are attached to the point until it is passed to the emitter where the fields get split off to their own iterator.

Built-in Iterators

There are many helper iterators that let us build queries:

  • Merge Iterator - This iterator combines one or more iterators into a single new iterator of the same type. This iterator guarantees that all points within a window will be output before starting the next window but does not provide ordering guarantees within the window. This allows for fast access for aggregate queries which do not need stronger sorting guarantees.

  • Sorted Merge Iterator - This iterator also combines one or more iterators into a new iterator of the same type. However, this iterator guarantees time ordering of every point. This makes it slower than the MergeIterator but this ordering guarantee is required for non-aggregate queries which return the raw data points.

  • Limit Iterator - This iterator limits the number of points per name/tag group. This is the implementation of the LIMIT & OFFSET syntax.

  • Fill Iterator - This iterator injects extra points if they are missing from the input iterator. It can provide null points, points with the previous value, or points with a specific value.

  • Buffered Iterator - This iterator provides the ability to "unread" a point back onto a buffer so it can be read again next time. This is used extensively to provide lookahead for windowing.

  • Reduce Iterator - This iterator calls a reduction function for each point in a window. When the window is complete then all points for that window are output. This is used for simple aggregate functions such as COUNT().

  • Reduce Slice Iterator - This iterator collects all points for a window first and then passes them all to a reduction function at once. The results are returned from the iterator. This is used for aggregate functions such as DERIVATIVE().

  • Transform Iterator - This iterator calls a transform function for each point from an input iterator. This is used for executing binary expressions.

  • Dedupe Iterator - This iterator only outputs unique points. It is resource intensive so it is only used for small queries such as meta query statements.

Call Iterators

Function calls in InfluxQL are implemented at two levels. Some calls can be wrapped at multiple layers to improve efficiency. For example, a COUNT() can be performed at the shard level and then multiple CountIterators can be wrapped with another CountIterator to compute the count of all shards. These iterators can be created using NewCallIterator().

Some iterators are more complex or need to be implemented at a higher level. For example, the DERIVATIVE() needs to retrieve all points for a window first before performing the calculation. This iterator is created by the engine itself and is never requested to be created by the lower levels.

Subqueries

Subqueries are built on top of iterators. Most of the work involved in supporting subqueries is in organizing how data is streamed to the iterators that will process the data.

The final ordering of the stream has to output all points from one series before moving to the next series and it also needs to ensure those points are printed in order. So there are two separate concepts we need to consider when creating an iterator: ordering and grouping.

When an inner query has a different grouping than the outermost query, we still need to group together related points into buckets, but we do not have to ensure that all points from one buckets are output before the points in another bucket. In fact, if we do that, we will be unable to perform the grouping for the outer query correctly. Instead, we group all points by the outermost query for an interval and then, within that interval, we group the points for the inner query. For example, here are series keys and times in seconds (fields are omitted since they don't matter in this example):

cpu,host=server01 0
cpu,host=server01 10
cpu,host=server01 20
cpu,host=server01 30
cpu,host=server02 0
cpu,host=server02 10
cpu,host=server02 20
cpu,host=server02 30

With the following query:

SELECT mean(max) FROM (SELECT max(value) FROM cpu GROUP BY host, time(20s)) GROUP BY time(20s)

The final grouping keeps all of the points together which means we need to group server01 with server02. That means we output the points from the underlying engine like this:

cpu,host=server01 0
cpu,host=server01 10
cpu,host=server02 0
cpu,host=server02 10
cpu,host=server01 20
cpu,host=server01 30
cpu,host=server02 20
cpu,host=server02 30

Within each one of those time buckets, we calculate the max() value for each unique host so the output stream gets transformed to look like this:

cpu,host=server01 0
cpu,host=server02 0
cpu,host=server01 20
cpu,host=server02 20

Then we can process the mean() on this stream of data instead and it will be output in the correct order. This is true of any order of grouping since grouping can only go from more specific to less specific.

When it comes to ordering, unordered data is faster to process, but we always need to produce ordered data. When processing a raw query with no aggregates, we need to ensure data coming from the engine is ordered so the output is ordered. When we have an aggregate, we know one point is being emitted for each interval and will always produce ordered output. So for aggregates, we can take unordered data as the input and get ordered output. Any ordered data as input will always result in ordered data so we just need to look at how an iterator processes unordered data.

raw query selector (without group by time) selector (with group by time) aggregator
ordered input ordered output ordered output ordered output ordered output
unordered input unordered output unordered output ordered output ordered output

Since we always need ordered output, we just need to work backwards and determine which pattern of input gives us ordered output. If both ordered and unordered input produce ordered output, we prefer unordered input since it is faster.

There are also certain aggregates that require ordered input like median() and percentile(). These functions will explicitly request ordered input. It is also important to realize that selectors that are grouped by time are the equivalent of an aggregator. It is only selectors without a group by time that are different.