Skip to content

Two correct alerts are being reported as dead rules #783

Description

@ecv

Summary

Two alerts shipped in this bundle are reported by production as having selected no series for over a week, which reads identically to a rule whose metric was renamed or never shipped.

Both are correct as written. The controller pod is running and not crash-looping, and the membership readiness rule reads 545 real series that are all currently ready, so counting the unready ones yields no series at all rather than a zero.

Nothing on either rule records that the emptiness is intended, so they sit in the same list as the genuinely dead ones and the reader cannot tell which is which.

What success looks like

  • Both rules declare themselves empty by design and stop being reported as dead
  • A crash-looping controller and an unready membership each still raise their alert

Background

Found while clearing the alerts that production reports as matching no data, tracked in https://github.com/datum-cloud/infra/issues/4909. The mechanism is a label the compliance exporter reads, described in https://github.com/datum-cloud/infra/blob/main/docs/runbooks/alerting/rule-matches-no-data.md.

https://claude.ai/code/session_01DirbefA5FTdhqHDUwneqYp

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions