This message was deleted.

There was an update to jGit in 2021.7.3 which is why it might be new. Thanks for the report.

ah good to know. let me know if you need more infos. but it should be easy to reproduce

Thanks.

In 2021.7.x we also added more agressive GCing to counter some scaling issues in HA. You can try to turn off that more aggressive GC and see if that helps (would be good to know). It's configurable with:

profile::master::cert_data_discard_history

it defaults to

true

(which is the more agressive behavior), you can disable it by setting it to

false

.

cert_data

isn't a name I would have expected in that context

I guess I should mention that it should only be an issue if you have HA enabled

ah that's a single primary, no compilers and no replicas

k, then probably not that issue

I also saw that in 3 other setups with additional compilers. I'm not sure if I saw it on the compilers itself or only on the primaries

thanks for the info, I got excited because I saw essentially the same error on primaries when replicating the CA data and the above setting helped, but it may just be how we're using JGit is causing this and the toggle for CA data just lessened our usage enough to stop those warning from appearing.

ah the CA data is exchanged via filesync and internal git repos, like normal puppet code?

yep, but file-sync is honestly a really bad fit for the cadata usage, so now we discard its history on every update which causes a lot of git activity and causes similar errors to what you saw. JGit seems to do no synchronization of its shared objects, but will ignore most of those errors and end up healing itself, for the most part, now.

ahhhh okay