Comment 35 for bug 1971093

Thanks Rick and Jussi for testing as well.

@Jussi:

For a start, at least the times are not totally off, the stamps convert to:
 - Tuesday, January 24, 2023 7:22:06 AM
 - Tuesday, January 24, 2023 7:21:59 AM

Also NTP enabled and syncing, and it would (hopefully) drift slowly if you would be off.
But the values indicate you jump back in time by ~7 seconds in the example.
Since you say this happens about every ~10 seconds you seem to go back in time rather quick.

The default conf has
  # CacheTimeout 120

which obviously does not show that behavior

But I've changed a test system to use what you have

<Plugin rrdtool>
 DataDir "/var/lib/collectd/rrd"
 CacheTimeout 10
 CacheFlush 900
 RandomTimeout 5

And that also does not expose the problem.

I found in other places [1] that one can also - after the config change - delete the files in the rrd path and then restart the service.

$ rm -rf /var/lib/collectd/rrd/*
$ systemctl restart collectd

That makes the issue show up for me as well.
But only on service start, not in a regular pattern.

I'm unsure - it is an issue, but it is no regression to having no collectd at all - so I'm unsure if we should make this hold up the SRU. I'd prefer to spawn a new bug to continue on this for a potential follow on fix. Especially as we have to wait and follow upstream issue which fell silent so far.
=> Please chime in on bug 2003839 for that aspect

But as I said IMHO this should not hold up the update here. Maybe we extend the time in -proposed to give us a chance to generate insights on this other case?