Transient, one-off labels?

[atombender]

I’d like to generate alerts for PostgreSQL statements that have been running for more than N minutes and are probably stuck due to locking or some I/O issue. The only way to report specific queries would be if the entire SQL statement was a label (e.g. postgres_query_time_total{sql="select * from foo"} 9204881), otherwise an alert would be completely opaque until the developer or person on duty can consult the table of running statements.

Does Prometheus handle this use case well, or is it a bad idea to generate so many transient labels?

[matthiasr]

The MySQL exporter kind of does this based on MySQLs summary tables. It uses a sanitized query as label value, so the set of queries remains bounded.

If you generate a new label for every query with every parameter, Prometheus will blow up very fast.

[brian-brazil]

Prometheus could alert that a query has been running too long, but it's not going to be able to tell you which one. Combining with a logging tool is the best approach.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.