Postgres-plugin

Feedback
1

I am using the postgres-plugin to monitor a postgres-database. Since the plugin is activated however i have SLOW-Queires in my database-log.

2025-05-05 08:49:29.511 UTC [3881207] LOG: duration: 7249.375 ms statement: SELECT current_database() AS db, schemaname, tablename, reltuples::bigint AS tup

This is the bloat-query which takes more than 7 seconds to execute. Either this query should be optimised or there should be the possibility to exclude it.
(Generally it would be a nice feature to include or exclude some db-monitoring-sections)

1 Like
2

Hi Frank, I also have this problem sometimes, currently maximum ist 3.5 seconds. I also had runtimes from ofter 40 seconds then the agent runs in a timeout. Then it helps if you make an ANALYZE on the database.

3

Hi @frank-wettstein and @ChrisZIT ,

after looking at the mk_postgres.py source code, I can now give a more complete picture.

Root cause: stale system catalog statistics

The bloat query reads reltuples and relpages from pg_class. When table statistics are stale (autovacuum not keeping up), the query planner chooses a poor execution plan, leading to 7–40s runtimes.

Additional finding: no statement_timeout on the bloat query

Looking at the source: get_stats() explicitly sets statement_timeout=30000 (30s) before running. get_bloat() does not. This means a slow bloat query can hang indefinitely and block the entire agent output — which explains Chris’s agent timeouts.

Also confirmed: no section exclusion is possible

The execute_all_queries() method calls all sections unconditionally in sequence. There is no EXCLUDED_SECTIONS config option. Frank’s feature request is fully legitimate — worth submitting to https://ideas.checkmk.com.

Fixes:

1. Immediate: run ANALYZE — refreshes statistics, brings bloat query back to normal speed (Chris’s workaround, correct but temporary).

2. Recommended: async execution with cache interval

Bloat data does not need to be checked every minute. Move the plugin to an async subfolder:

mkdir -p /usr/lib/check_mk_agent/plugins/3600
mv /usr/lib/check_mk_agent/plugins/mk_postgres /usr/lib/check_mk_agent/plugins/3600/

This runs the plugin (including the bloat query) only once per hour instead of every minute, eliminating the DB load issue entirely.

3. Long-term: tune autovacuum so statistics stay fresh and the bloat query stays fast.

Try and give response …

Happy monitoring!