pgtrackslow_queries

PostgreSQL (9.5+) extension for slow queries tracking. This extension logs SQL queries and related information into dedicated file, only if query execution duration exceeds a certain amount of time. Logged query list can be retrieve using SQL function pg_track_slow_queries(). Log file can be truncated with pg_track_slow_queries_reset().

Status

Still under development.

Installation

$ sudo PG_CONFIG=/path/to/pgsql/bin/pg_config make install

Configuration

The extension library should be loaded with Postgres parameter shared_preload_libraries and Postgres instance restarted:

shared_preload_libraries='pg_track_slow_queries'

Then, the extension could be created on postgres database with:

CREATE EXTENSION pg_track_slow_queries;

Parameters / GUCs

Parameter	unit	default	description
pg_track_slow_queries.log_min_duration	ms	-1	This parameter sets the minimum execution time (in ms) above which queries will be logged. -1 (default value) means the feature is disabled.
pg_track_slow_queries.compression	bool	on	Enable or disable row compression. Compression could have impacts on performances but will save disk space.
pg_track_slow_queries.max_file_size	kB	-1	Sets the maximum size of storage file. -1 means no limitation.
pg_track_slow_queries.log_plan	bool	on	Enable execution plan logging.
pg_track_slow_queries.cost_analyze	int	-1	Cost threshold beyond which query execution analysis is performed, like EXPLAIN ANALYZE does. Could have high impacts on performances. -1 means the feature is disabled.

Usage

Access to logged queries:

SELECT * FROM pg_track_slow_queries();

-[ RECORD 1 ]-----+--------------------------------------------------
datetime          | 2019-03-12 11:48:03.783453+01
duration          | 1001.24
username          | julien
appname           | psql
dbname            | postgres
temp_blks_written | 0
hitratio          | 100
ntuples           | 1
query             | SELECT pg_sleep(1);
plan              | {                                                   +
                  |   "Plan": {                                         +
                  |     "Node Type": "Result",                          +
                  |     "Parallel Aware": false,                        +
                  |     "Startup Cost": 0.00,                           +
                  |     "Total Cost": 0.01,                             +
                  |     "Plan Rows": 1,                                 +
                  |     "Plan Width": 4,                                +
                  |     "Output": ["pg_sleep('1'::double precision)"]   +
                  |   }                                                 +
                  | }

Reset log file:

SELECT * FROM pg_track_slow_queries_reset();

Columns

1. datetime: statement's end of execution datetime (timestamptz)
2. duration: execution duration, in seconds
3. username: username that issued the statement
4. appname: application name
5. dbname: database name
6. temp_blks_written: number of blocks written for temporary files usage
7. hitratio: statement cache hit-ratio
8. ntuples: number of tuples affected by the statement
9. query: the statement
10. plan: statement execution plan (JSON)

Caveats

• Do not support utility statements (VACUUM, REINDEX, etc).
• Do not tracks parameters values of prepared statements.

Benchmarks

Here are the results of a worse case scenario benchmark: tracking of all statements, read only queries, small database. WARNING: take it with caution, they've been made on a small machine with a low number of CPU. Better benchmarks coming soon.

TPS without any statement tracking: 42230

Method	Plan	ANALYZE	TPS	Loss
pg_stat_statements track=top,tack_utility=off,save=on	No	No	40577	4%
pg_track_slow_queries log_min_duration=0,compression=off,log_plan=off	No	No	29378	30%
core log_min_duration_statements=0,logging_collector=on	No	No	26475	38%
pg_track_slow_queries log_min_duration=0,compression=off	Yes	No	26221	38%
pg_track_slow_queries log_min_duration=0,cost_analyze=0,compression=off	Yes	Yes	24874	41%
pg_track_slow_queries log_min_duration=0	Yes	No	24597	42%
auto_explain logging_collector=on,log_min_duration=0,log_format=json	Yes	No	22654	46%
auto_explain [..],log_analyze=on,log_timing=on,log_buffers=on	Yes	Yes	21745	48%
pg_track_slow_queries log_min_duration=0,max_file_size=10GB	Yes	No	19940	53%