If a sentinel command is configured, it runs every minute.
The configured sentinel command runs every minute, instead of once every period. The reason for this is appears to be the Prometheus metrics updater maintainRebootRequiredMetric() here:
https://github.com/kubereboot/kured/blob/9e4b69f8189cbf685e459ad1be2eb15f4dfd0cd2/cmd/kured/main.go#L566C1-L575C2
This is arguably incorrect. The metric should be updated only when the sentinel command is run in the main code (i.e. once every period interval).
A compromise would be to have the function maintainRebootRequiredMetric() sleep for period instead of the hard-coded time.Minute that is there now.
This issue was automatically considered stale due to lack of activity. Please update it and/or join our slack channels to promote it, before it automatically closes (in 7 days).
The configured sentinel command runs every minute, instead of once every
period.
I can confirm this behaviour for version 1.15.0. Although I set period to 30min the sentinel command runs every 1min.
This issue was automatically considered stale due to lack of activity. Please update it and/or join our slack channels to promote it, before it automatically closes (in 7 days).
Since no one has turned up to defend the code's position, I have to infer that we'd be open to some kind of PR to fix this.
(Un-labeled as inactive)
@haraldkoch @kingdonb is this as simple as this? https://github.com/kubereboot/kured/pull/919
Note the potential downside for folks who want to have a separate periodicity for metrics data.
Huzzah!
Note the potential downside for folks who want to have a separate periodicity for metrics data.
The Prometheus library caches the metric values passed to metric.Set(), and returns the cached value to anyone who queries the metrics endpoint, so this shouldn't affect any metrics data.
This issue was automatically considered stale due to lack of activity. Please update it and/or join our slack channels to promote it, before it automatically closes (in 7 days).
It would be lovely to merge the linked PR instead of letting this issue decay again?
This issue was automatically considered stale due to lack of activity. Please update it and/or join our slack channels to promote it, before it automatically closes (in 7 days).