In case of database lock or low performance prometheus-exporer accumulates connections to the database

[pierresouchay]

What did you do?

1. Have postgres-exporter running
2. Have a kind of lock to the database -> some requests of postgres-exporter are locked (for instance, on a WAL saturated instance, a big DELETE locking the stats)
3. Have a prometheus server fetching the data

What did you expect to see?

1. one or a few connections connection opened to the server
2. Some errors of scrapping maybe

What did you see instead? Under which circumstances?

1. On the database, a huge perf issue arise (likely due to dbt)
2. postgres-exporter accumulates postgresql queries stuck
3. Since we did not cap enough the user, it accumulates more than 50% of available connections (in an example, 53 open connections (fair enough, I should have used a user with less privileges, will do), avoiding other clients to connect

I would expect the connections to be canceled at some point (for instance when the prometheus query times out.

Example:

We have 53 queries in flight with the user...

Environment

• System information:

In the container:

uname -srm
Linux 5.15.0-161-generic x86_64

• postgres_exporter version:

Using docker image prometheuscommunity/postgres-exporter:v0.17.1:

postgres_exporter --version
postgres_exporter, version 0.17.1 (branch: HEAD, revision: 1e574cf4fd2a75a8a707d424eafcaa0b88cb7af4)
  build user:       root@b0ba132eddb0
  build date:       20250226-13:53:15
  go version:       go1.23.6
  platform:         linux/amd64
  tags:             unknown

• postgres_exporter flags:

None set, using the default ones only

• PostgreSQL version: 17.1

• Logs:

insert logs relevant to the issue here

[pierresouchay]

@sysadmind @SuperQ Fix available in #1229