Continuous Queries

Warning! This page documents an old version of InfluxDB, which is no longer actively developed. InfluxDB v1.2 is the most recent stable version of InfluxDB.

When writing large amounts of data to InfluxDB, you may often want to downsample the raw data, that is, use GROUP BY time() with an InfluxQL function to change the high frequency data into lower frequency data. Repeatedly running the same queries by hand can be tedious. InfluxDB’s continuous queries (CQ) simplify the downsampling process; CQs run automatically and write the query results to another measurement.

CQ definition

A CQ is an InfluxQL query that the system runs automatically and periodically within a database. InfluxDB stores the results of the CQ in a specified measurement. CQs require a function in the SELECT clause and must include a GROUP BY time() clause.

CQs do not maintain any state. Each execution of a CQ is a standalone query that resamples all points in the database matching the conditions of the query.

The time ranges of the CQ results have round-number boundaries that are set internally by the database. There is currently no way for users to alter the start or end times of the intervals.

Only admin users are allowed to work with continuous queries. For more on user privileges, see Authentication and Authorization.

Note: CQs only execute on data received after the CQ’s creation. If you’d like to downsample data written to InfluxDB before the CQ was created, see the examples in Data Exploration.

InfluxQL for creating a CQ

The CREATE CONTINUOUS QUERY statement

CREATE CONTINUOUS QUERY <cq_name> ON <database_name> [RESAMPLE [EVERY <interval>] [FOR <interval>]] BEGIN SELECT <function>(<stuff>)[,<function>(<stuff>)] INTO <different_measurement> FROM <current_measurement> [WHERE <stuff>] GROUP BY time(<interval>)[,<stuff>] END

The CREATE CONTINUOUS QUERY statement is essentially an InfluxQL query surrounded by CREATE CONTINUOUS QUERY [...] BEGIN and END. The following discussion breaks the CQ statement into its meta portion (everything between CREATE and BEGIN) and query portion (everything between BEGIN and END).

Meta syntax:

CREATE CONTINUOUS QUERY ON <database_name> [RESAMPLE [EVERY <interval>] [FOR <interval>]]

A CQ belongs to a database. Specify the database where you want the CQ to live with ON <database_name>.

The optional RESAMPLE clause determines how often InfluxDB runs the CQ (EVERY <interval>) and the time range over which InfluxDB runs the CQ (FOR <interval>). If included, the RESAMPLE clause must specify either EVERY, or FOR, or both. Without the RESAMPLE clause, InfluxDB runs the CQ at the same interval as the GROUP BY time() interval and it calculates the query for the most recent GROUP BY time() interval (that is, where time is between now() and now() minus the GROUP BY time() interval).

Query syntax:

BEGIN SELECT <function>(<stuff>)[,<function>(<stuff>)] INTO <different_measurement> FROM <current_measurement> [WHERE <stuff>] GROUP BY time(<interval>)[,<stuff>] END

The query portion of the statement differs from a typical SELECT [...] GROUP BY (time) statement in two ways:

  1. The INTO clause: This is where you specify the destination measurement for the query results.

  2. The optional WHERE clause: Because CQs run on regularly incremented time intervals you don’t need to (and shouldn’t!) specify a time range in the WHERE clause. When included, the CQ’s WHERE clause should filter information about tags.

Note: If you include a tag in the CQ’s SELECT clause, InfluxDB changes the tag in <current_measurement> to a field in <different_measurement>. To preserve a tag in <different_measurement>, only include the tag key in the CQ’s GROUP BY clause.

If you specify a tag in the CQ’s SELECT clause and in the CQ’s GROUP BY clause, you will not be able to query the data in <different_measurement>. See GitHub Issue #4630 for more information.

CQ examples:

  • Create a CQ with one function:

    > CREATE CONTINUOUS QUERY minnie ON world BEGIN SELECT min(mouse) INTO min_mouse FROM zoo GROUP BY time(30m) END
    

    Once executed, InfluxDB automatically calculates the 30 minute minimum of the field mouse in the measurement zoo, and it writes the results to the measurement min_mouse. Note that the CQ minnie only exists in the database world.

  • Create a CQ with one function and write the results to another retention policy:

    > CREATE CONTINUOUS QUERY minnie_jr ON world BEGIN SELECT min(mouse) INTO world."7days".min_mouse FROM world."1day".zoo GROUP BY time(30m) END
    

    The CQ minnie_jr acts in the same way as the CQ minnie, however, InfluxDB calculates the 30 minute minimum of the field mouse in the measurement zoo and under the retention policy 1day, and it automatically writes the results of the query to the measurement min_mouse under the retention policy 7days.

    Combining CQs and retention policies provides a useful way to automatically downsample data and expire the unnecessary raw data. For a complete discussion on this topic, see Downsampling and Data Retention.

  • Create a CQ with two functions:

    > CREATE CONTINUOUS QUERY minnie_maximus ON world BEGIN SELECT min(mouse),max(imus) INTO min_max_mouse FROM zoo GROUP BY time(30m) END
    

    The CQ minnie_maximus automatically calculates the 30 minute minimum of the field mouse and the 30 minute maximum of the field imus (both fields are in the measurement zoo), and it writes the results to the measurement min_max_mouse.

  • Create a CQ with two functions and personalize the field keys in the results:

    > CREATE CONTINUOUS QUERY minnie_maximus_1 ON world BEGIN SELECT min(mouse) AS minuscule,max(imus) AS monstrous INTO min_max_mouse FROM zoo GROUP BY time(30m) END
    

    The CQ minnie_maximus_1 acts in the same way as minnie_maximus, however, InfluxDB names field keys miniscule and monstrous in the destination measurement instead of min and max. For more on AS, see Functions.

  • Create a CQ with a 30 minute GROUP BY time() interval that runs every 15 minutes:

    > CREATE CONTINUOUS QUERY vampires ON transylvania RESAMPLE EVERY 15m BEGIN SELECT count(dracula) INTO vampire_populations FROM raw_vampires GROUP BY time(30m) END
    

    Without RESAMPLE EVERY 15m, vampires would run every 30 minutes - the same interval as the GROUP BY time() interval.

  • Create a CQ with a 30 minute GROUP BY time() interval that runs every 30 minutes and computes the query for all GROUP BY time() intervals within the last hour:

    > CREATE CONTINUOUS QUERY vampires_1 ON transylvania RESAMPLE FOR 60m BEGIN SELECT count(dracula) INTO vampire_populations_1 FROM raw_vampires GROUP BY time(30m) END
    

    InfluxDB runs vampires_1 every 30 minutes (the same interval as the GROUP BY time() interval) and it computes two queries per run: one where time is between now() and now() - 30m and one where time is between now() - 30m and now() - 60m. Without the RESAMPLE clause, InfluxDB would compute the query for only one 30 minute interval, that is, where time is between now() and now() - 30m.

  • Create a CQ with a 30 minute GROUP BY time() interval that runs every 15 minutes and computes the query for all GROUP BY time() intervals within the last hour:

    > CREATE CONTINUOUS QUERY vampires_2 ON transylvania RESAMPLE EVERY 15m FOR 60m BEGIN SELECT count(dracula) INTO vampire_populations_2 FROM raw_vampires GROUP BY time(30m) END
    

    vampires_2 runs every 15 minutes and computes two queries per run: one where time is between now() and now() - 30m and one where time is between now() - 30m and now() - 60m

CQs with backreferencing

Use :MEASUREMENT in the INTO statement to backreference measurement names:

CREATE CONTINUOUS QUERY <cq_name> ON <database_name> BEGIN SELECT <function>(<stuff>)[,<function>(<stuff>)] INTO <database_name>.<retention_policy>.:MEASUREMENT FROM </relevant_measurement(s)/> [WHERE <stuff>] GROUP BY time(<interval>)[,<stuff>] END

CQ backreferencing example:

> CREATE CONTINUOUS QUERY elsewhere ON fantasy BEGIN SELECT mean(value) INTO reality."default".:MEASUREMENT FROM /elf/ GROUP BY time(10m) END

The CQ elsewhere automatically calculates the 10 minute average of the field value in each elf measurement in the database fantasy. It writes the results to the already-existing database reality, preserving all of the measurement names in fantasy.

A sample of the data in fantasy:

> SHOW MEASUREMENTS
name: measurements
------------------
name
elf1
elf2
wizard
>
> SELECT * FROM elf1
name: cpu_usage_idle
--------------------
time			               value
2015-12-19T01:15:30Z	 97.76333874796951
2015-12-19T01:15:40Z	 98.3129217695576
[...]
2015-12-19T01:36:00Z	 94.71778221778222
2015-12-19T01:35:50Z	 87.8

A sample of the data in reality after elsewhere runs for a bit:

> SHOW MEASUREMENTS
name: measurements
------------------
name
elf1
elf2
>
> SELECT * FROM elf1
name: elf1
--------------------
time			               mean
2015-12-19T01:10:00Z	 97.11668879244841
2015-12-19T01:20:00Z	 94.50035091670394
2015-12-19T01:30:00Z	 95.99739053789172

List CQs with SHOW

List every CQ by database with:

SHOW CONTINUOUS QUERIES

Example:

> SHOW CONTINUOUS QUERIES
name: reality
-------------
name	query

name: fantasy
-------------
name		     query
elsewhere	 CREATE CONTINUOUS QUERY elsewhere ON fantasy BEGIN SELECT mean(value) INTO reality."default".:MEASUREMENT FROM fantasy."default"./cpu/ WHERE cpu = 'cpu-total' GROUP BY time(10m) END

The output shows that the database reality has no CQs and the database fantasy has one CQ called elsewhere.

Delete CQs with DROP

Delete a CQ from a specific database with:

DROP CONTINUOUS QUERY <cq_name> ON <database_name>

Example:

> DROP CONTINUOUS QUERY elsewhere ON fantasy
>

A successful DROP CONTINUOUS QUERY returns an empty response.

Backfilling

CQs do not backfill data, that is, they do not compute results for data written to the database before the CQ existed. Instead, users can backfill data with the INTO clause. Unlike CQs, backfill queries require a WHERE clause with a time restriction.

Examples

Here is a basic backfill example:

> SELECT min(temp) as min_temp, max(temp) as max_temp INTO "reading.minmax.5m" FROM reading
WHERE time >= '2015-12-14 00:05:20' AND time < '2015-12-15 00:05:20'
GROUP BY time(5m)

Tags (sensor_id in the example below) can be used optionally in the same way as in CQs:

> SELECT min(temp) as min_temp, max(temp) as max_temp INTO "reading.minmax.5m" FROM reading
WHERE time >= '2015-12-14 00:05:20' AND time < '2015-12-15 00:05:20'
GROUP BY time(5m), sensor_id

To prevent the backfill from creating a huge number of “empty” points containing only null values, fill() can be used at the end of the query:

> SELECT min(temp) as min_temp, max(temp) as max_temp INTO "reading.minmax.5m" FROM reading
WHERE time >= '2015-12-14 00:05:20' AND time < '2015-12-15 00:05:20'
GROUP BY time(5m), fill(none)

If you would like to further break down the queries and run them with even more control, you can add additional WHERE clauses:

> SELECT min(temp) as min_temp, max(temp) as max_temp INTO "reading.minmax.5m" FROM reading
WHERE sensor_id="EG-21442" AND time >= '2015-12-14 00:05:20' AND time < '2015-12-15 00:05:20'
GROUP BY time(5m)

Further reading

Now that you know how to create CQs with InfluxDB, check out Downsampling and Data Retention for how to combine CQs with retention policies to automatically downsample data and expire unnecessary data.