Missing prometheus metrics

[slimsag]

This is a list of metrics that do not exist and would be helpful in debugging customer issues:

• # of repositories that are indexed, unindexed, indexing
• repo cloning
  • # of repositories that are cloning -> src_gitserver_clone_queue
  • # of repositories that have cloned since gitserver started -> src_gitserver_clone_queue
  • # of repositories that are not cloned on the instance overall
  • # of repositories that are cloned on the instance overall
• Total memory & CPU available to container/host-machine in sourcegraph/server deployments
• Network / disk IO in sourcegraph/server deployments
• # of HTTP requests made to Sourcegraph by status code (src_http_request_duration_seconds is a histogram and has src_http_request_duration_seconds_count)
• # of GraphQL requests made by (type, field) -- current histogram doesn't cover this
• # of searches hitting zoekt vs. searcher
• # of symbol searches hitting zoekt vs. symbols
• # of diff/commit searches
• # of searches hitting indexed vs. unindexed branches

[uwedeportivo]

number of repositories that are cloned, not cloned, cloning:

src_gitserver_repo_cloned, src_gitserver_clone_queue

[slimsag]

src_gitserver_repo_cloned appears to be the number of repos that have been cloned since gitserver started, not the # that are cloned on the instance period

[slimsag]

I broke out that section into four bullets to indicate the two missing metrics

[uwedeportivo]

Cool, thanks

[uwedeportivo]

each histogram has a counter: src_http_request_duration_seconds_count should do it.

[slimsag]

Nice, we should perhaps declare rules to make that more obvious, so that src_http_request_count maps to that for each relevant histogram?

[slimsag]

Closing this issue.

of repositories that are indexed, unindexed, indexing

Superseded by:

• If configuration becomes invalid for any reason, no alert fires for all repositories being removed #9790
• Missing monitoring to notify admins of mass recloning events #9789
• Alert when search indexing is failing #9783
• Alert when repository updates are failing #9782

of repositories that are not cloned on the instance overall

Superseded by #9837

of repositories that are cloned on the instance overall

Not useful / can't be alerted on.

• Total memory & CPU available to container/host-machine in sourcegraph/server deployments
• Network / disk IO in sourcegraph/server deployments

No longer a goal with Docker Compose deployments existing.

of GraphQL requests made by (type, field)

Superseded by #9797

• of searches hitting zoekt vs. searcher

• of symbol searches hitting zoekt vs. symbols

Superseded by #9838

of diff/commit searches

Superseded by #9744

of searches hitting indexed vs. unindexed branches

Superseded by #9838

that are cloned on the instance period

Already present in new monitoring via clone queue size (which effectively communicates the same thing).