How to collect server level metrics for PostgreSQL

[XiangyuZhang-G]

I am developing a collector for PostgreSQL, and this is a part of the content.

collector_name: pgsql_connection

metrics:
  - metric_name: pg_setting_max_connections
    type: gauge
    help: 'Maximum number of connections in settings'
    values:
      - max
    query_ref: _ref_connection_setting

  - metric_name: pg_setting_superuser_reserved_connections
    type: gauge
    help: 'The number of reserved connections for super users in settings'
    values:
      - superuser_reserved
    query_ref: _ref_connection_setting

queries:
  - query_name: _ref_connection_setting
    query: |
      select
        current_setting('max_connections')::int as max,
        current_setting('superuser_reserved_connections')::int as superuser_reserved
      from
        pg_stat_activity;

but there are some error info:

time=2025-10-24T01:00:37.682-05:00 level=ERROR source=promhttp.go:51 msg="Error gathering metrics" error="collected metric \"pg_setting_superuser_reserved_connections\" { gauge:{value:15}} was collected before with the same name and label values"

I used postgres exporter before. In that tool, I could avoid collecting server-level metrics redundantly by specifying "master: true". Does SQL exporter have a similar method?

[burningalchemist]

@XiangyuZhang-G SQL Exporter is not compatible with Postgres Exporter one-to-one, so some changes might be needed.

I'd say in your case, your query returns multiple rows with the same value. For sql_exporter it's important to avoid collecting duplicate data points. So we have the following options here:

• use SELECT DISTINCT to filter results on the database end (this should address redundant values);
• add an extra column to the query that would make the result set distinct (if possible), and add this column name to key_labels parameter;
• limit the result set with LIMIT 1.

sql_exporter is a database-agnostic exporter, so there is no master: true setting. It doesn't do any data processing to stay fast and scalable. Instead, we may use a database engine to prepare the result set to be exported beforehand.

I'm happy to assist, so let me know 😃👍