checkpoints_timed not found when monitoring Postgres 17beta3

[HansK-p]

What did you do? Upgraded from Postgresql 16 to Postgresql 17beta3

What did you expect to see? Monitoring working as before.

What did you see instead? Under which circumstances?
The error message below is logged by postgres_exporter:

caller=collector.go:202 level=error msg="collector failed" name=stat_bgwriter duration_seconds=0.018634044 err="pq: column \"checkpoints_timed\" does not exist"

This must be as checkpoint related columns in PG 17 have been moved from pg_stat_bgwriter to pg_stat_checkpointer according to https://www.dbi-services.com/blog/postgresql-17-new-catalog-view-pg_stat_checkpointer/.

Environment Running in Kubernetes, any environment.

• System information:

image: postgres:17beta3
prometheuscommunity/postgres-exporter:v0.15.0

• postgres_exporter version: v0.15.0

• postgres_exporter flags: none

• PostgreSQL version: postgres 17beta3

• Logs:

caller=collector.go:202 level=error msg="collector failed" name=stat_bgwriter duration_seconds=0.018634044 err="pq: column \"checkpoints_timed\" does not exist"

[n-rodriguez]

Same here :

PostgreSQL version

postgresql-17               17.0-1.pgdg120+1

postgres_exporter version :

root@32132694b4b9:/# postgres_exporter --version
postgres_exporter, version 0.15.0 (branch: HEAD, revision: 68c176b8833b7580bf847cecf60f8e0ad5923f9a)
  build user:       root@7c5a76bc737a
  build date:       20231027-14:38:04
  go version:       go1.21.3
  platform:         linux/arm64
  tags:             unknown

postgres-exporter-1   | {"caller":"collector.go:202","duration_seconds":0.01493675,"err":"pq: column \"checkpoints_timed\" does not exist","level":"error","msg":"collector failed","name":"stat_bgwriter","ts":"2024-09-28T16:02:37.125Z"}
postgres-1            | 2024-09-28 16:02:37 UTC [158-1] concerto@concerto ERROR:  column "checkpoints_timed" does not exist at character 10
postgres-1            | 2024-09-28 16:02:37 UTC [158-2] concerto@concerto STATEMENT:  SELECT
postgres-1            |       checkpoints_timed
postgres-1            |       ,checkpoints_req
postgres-1            |       ,checkpoint_write_time
postgres-1            |       ,checkpoint_sync_time
postgres-1            |       ,buffers_checkpoint
postgres-1            |       ,buffers_clean
postgres-1            |       ,maxwritten_clean
postgres-1            |       ,buffers_backend
postgres-1            |       ,buffers_backend_fsync
postgres-1            |       ,buffers_alloc
postgres-1            |       ,stats_reset
postgres-1            |     FROM pg_stat_bgwriter;

[vincejv]

is there a way to silence these logs while #1072 is being reviewed?

[hiteshnayak305]

+1

[High5Apps]

@vincejv I was able to silence the error logs by running the exporter with the --no-collector.stat_bgwriter flag.

For more info see the Flags section of the README

[vincejv]

@High5Apps thanks a lot, saved my day, also based from the linked commit
add this to docker compose command: '--no-collector.stat_bgwriter'

[garry-t]

any movements here? :)

[Enterprize-1701]

Same thing
2024-12-03 17:37:56.053 UTC [1599924] postgres_exporter@postgres ERROR: column "checkpoints_timed" does not exist at character 10 2024-12-03 17:37:56.053 UTC [1599924] postgres_exporter@postgres STATEMENT: SELECT checkpoints_timed ,checkpoints_req ,checkpoint_write_time ,checkpoint_sync_time ,buffers_checkpoint ,buffers_clean ,maxwritten_clean ,buffers_backend ,buffers_backend_fsync ,buffers_alloc ,stats_reset FROM pg_stat_bgwriter;

Using in ansible role postgres_exporter for now

postgres_exporter_disabled_collectors:
  - stat_bgwriter

[agjs]

--no-collector.stat_bgwriter

Did it for me as well.

[Nikolay-Ch]

--no-collector.stat_bgwriter

Did it for me as well.

This is a workaround, not a solution...

[Enterprize-1701]

Solution for now is to use SQL_exporter and make your own queries:

collector_name: pg_writer
metrics:
  - metric_name: pg_stat_bgwriter_buffers_clean_total
    type: counter
    help: "Total buffers cleaned by the background writer"
    values: [buffers_clean]
    query_ref: pg_stat_bgwriter

  - metric_name: pg_stat_bgwriter_maxwritten_clean
    type: gauge
    help: "Number of times the background writer stopped due to max written clean limit"
    values: [maxwritten_clean]
    query_ref: pg_stat_bgwriter

  - metric_name: pg_stat_bgwriter_buffers_alloc_total
    type: counter
    help: "Total buffers allocated"
    values: [buffers_alloc]
    query_ref: pg_stat_bgwriter

queries:
  - query_name: pg_stat_bgwriter
    query: |
      SELECT
        buffers_clean,
        maxwritten_clean,
        buffers_alloc
      FROM
        pg_stat_bgwriter;

[n-rodriguez]

Solution for now is to use SQL_exporter and make your own queries:

A better solution #1072 😄

[Enterprize-1701]

Solution for now is to use SQL_exporter and make your own queries:

A better solution #1072 😄

It was't merge.

[n-rodriguez]

It was't merge.

I know, I'm still waiting

[mustdiechik]

https://www.dbi-services.com/blog/postgresql-17-new-catalog-view-pg_stat_checkpointer/

[jfeldstein-cw]

You can now use the --no-collector.stat_bgwriter flag (docs)

Or set the corresponding helm chart value:

prometheus-postgres-exporter:
  config:
    disableCollectorBgwriter: true