Fluent Bit stream processor uses common SQL to perform record queries. The following section describe the features available and examples of it.
You can find the detailed query language syntax in BNF form here. The following section will be a brief introduction on how to write SQL queries for Fluent Bit stream processing.
SELECT results_statementFROM STREAM:stream_name | TAG:match_rule[WINDOW TUMBLING (integer SECOND)][WHERE condition][GROUP BY groupby]
Select keys from records coming from a stream or records matching a specific Tag pattern. Note that a simple
SELECT statement not associated from a stream creation will send the results to the standard output interface (stdout), useful for debugging purposes.
The query allows filtering the results by applying a condition using
WHERE statement. We will explain
GROUP BY statements later in aggregation functions section.
Select all keys from records coming from a stream called apache:
SELECT * FROM STREAM:apache;
Select code key from records which Tag starts with apache.:
SELECT code AS http_status FROM TAG:'apache.*';
Since the TAG selector allows the use of wildcards, we put the value between single quotes.
CREATE STREAM stream_name[WITH (property_name=value, [...])]AS select_statement
Create a new stream of data using the results from the
SELECT statement. New stream created can be optionally re-ingested back into Fluent Bit pipeline if the property Tag is set in the WITH statement.
Create a new stream called hello from stream called apache:
CREATE STREAM hello AS SELECT * FROM STREAM:apache;
Create a new stream called hello for all records which original Tag starts with apache:
CREATE STREAM hello AS SELECT * FROM TAG:'apache.*';
Aggregation functions are used in
results_statement on the keys, allowing to perform data calculation on groups of records. Group of records that aggregation functions apply on are determined by
WINDOW keyword. When
WINDOW is not specified, aggregation functions apply on the current buffer of records received, which may have non-deterministic number of elements. Aggregation functions can be applied on records in a window of a specific time interval (see the syntax of
WINDOW in select statement).
Fluent Bit streaming currently supports tumbling window, which is non-overlapping window type. That means, a window of size 5 seconds performs aggregation computations on records over a 5-second interval, and then starts new calculations for the next interval.
In addition, the syntax support
GROUP BY statement, which groups the results by the one or more keys, when they have the same values.
SELECT AVG(size) FROM STREAM:apache WHERE method = 'POST' ;
Calculates the average of request sizes in POST requests.
SELECT host, COUNT(*) FROM STREAM:apache WINDOW TUMBLING (5 SECOND) GROUP BY host;
Count the number of records in 5 second windows group by host IP addresses.
SELECT MIN(key) FROM STREAM:apache;
Gets the minimum value of a key in a set of records.
SELECT MIN(key) FROM STREAM:apache;
Gets the maximum value of a key in a set of records.
SELECT SUM(key) FROM STREAM:apache;
Calculates the sum of all values of key in a set of records.
Time functions adds a new key into the record with timing data
SELECT NOW() FROM STREAM:apache;
Add system time using format: %Y-%m-%d %H:%M:%S. Output example: 2019-03-09 21:36:05.
SELECT UNIX_TIMESTAMP() FROM STREAM:apache;
Add current Unix timestamp to the record. Output example: 1552196165 .
Record functions append new keys to the record using values from the record context.
SELECT RECORD_TAG() FROM STREAM:apache;
Append Tag string associated to the record as a new key.
SELECT RECORD_TIME() FROM STREAM:apache;
Similar to conventional SQL statements,
WHERE condition is supported in Fluent Bit query language. The language supports conditions over keys and subkeys, for instance:
SELECT AVG(size) FROM STREAM:apache WHERE method = 'POST' AND status = 200;
It is possible to check the existence of a key in the record using record-specific function
SELECT MAX(key) FROM STREAM:apache WHERE @record.contains(key);
And to check if the value of a key is/is not
SELECT MAX(key) FROM STREAM:apache WHERE key IS NULL;
SELECT * FROM STREAM:apache WHERE user IS NOT NULL;
Append a new key with the record Timestamp in double format: seconds.nanoseconds. Output example: 1552196165.705683 .