Incorrect values for millisSinceLastRepairForSchedule in prometheusMetrics

[rtib]

Project board link

I'm running Reaper in sidecar mode and tried to monitor it via Prometheus metrics. I've created a dashboard showing repair progress along with the time since the last scheduled repair. While repair progress is working fine, time since last repair is showing correct value for only one schedule, others are arrested to epoch.

Reaper shows on the webui that all schedules have run within 7 days

and the schedules are still active

When looking at the prometheusMetrics endpoint, however, the values exported are wrong

However, this is not a prometheus-exporter issue, the Dropwizard report contains the same problem

Looking into it I've found that millis since epoch is the fallback value for a repair schedule if no repairs from this schedule were completed.

cassandra-reaper/src/server/src/main/java/io/cassandrareaper/service/RepairScheduleService.java

Line 190 in 0661688

.orElse(DateTime.now().getMillis()); // Return epoch if no repairs from this schedule were completed

[rtib]

Taking a look into the database, the according repair_run entry does have a valid end_time.

[rtib]

Digging a bit deeper unveiled that the last_run field of repair_schedule_v1 contains null for all but one entries. That makes millisSinceLastRepairForSchedule to fall back to return epoch.

[Nassz]

This metric is also not working for multiple hosts (for 3 nodes cluster), which can answer why U have only 1 KS metric. Alternatively, U can use 7days - io_cassandrareaper_service_RepairRunner_millisSinceLastRepair :)

[rtib]

That would also contain manually started repairs, which is okay, but then it is hard to distinguish between schedules and manual runs.