#8464 Increase replication changelog trimming interval
Closed: fixed 8 months ago by abbra. Opened 8 months ago by mreynolds.

Issue

A long time ago the DS team recommended that the changelog trimming interval be set to 7 days. However, more recently we tend to see more time skews on certain platforms, and issues where it appears changes were trimmed too early (which can break replication).

It would be better to set the trimming interval to 30 days. This still prevents the changelog from getting too large, and it should help with some of the other issues we are now seeing.


I will be filing a PR shortly...

I agree that 30 days delay is probably safer with minimal size impact.
However I have a doubt how it can break replication with established replica. IIRC trimming is based on oldest csn in replication agreements and should not be impacted by csn jumps. We may have a bug in the way replication computes the starting csn of the trimming.

I agree that 30 days delay is probably safer with minimal size impact.
However I have a doubt how it can break replication with established replica. IIRC trimming is based on oldest csn in replication agreements and should not be impacted by csn jumps. We may have a bug in the way replication computes the starting csn of the trimming.

My main concern is that we see replication outages in FreeIPA because changes are not in the changelog anymore. We also see customers who want to temporarily remove replicas and eventually bring them back. This would make that process more robust.

I am also just trying to be somewhat proactive in the time skew scenarios, but like I said my main concern is the number of cases I see where changes are no longer present in the changelog.

master:

  • c08c7e1 Increase replication changelog trimming to 30 days

ipa-4-8:

  • 14f27d2 Increase replication changelog trimming to 30 days

Metadata Update from @abbra:
- Issue close_status updated to: fixed
- Issue status updated to: Closed (was: Open)

8 months ago

Login to comment on this ticket.

Metadata