Cluster topology change recursion

[thejj]

Cluster topology change recursion

[TheJJ]

Hi!
This error occurs when the osdmap version id increases while the script gathers data from the cluster.
It's a safety measure because in ceph there's unfortunately no transaction for queries: i have to call ceph osd dump, ceph pg dump etc one after the other, and hope the cluster state didn't change in between. And i validate this by getting the osdmap version before and after these calls, and then abort if the epochs don't match.
You can check this with ceph osd dump | grep epoch.

The osdmap epoch is increased whenever the osdmap changes, e.g. a scrub is finished, a pg is moved, ...
To see what the differences are, you can try this:

• ceph pg dump > /tmp/pgdump-0
• wait for a bit, 15s or something
• ceph pg dump > /tmp/pgdump-1
• diff -u /tmp/pgdump-0 /tmp/pgdump-1

My guess is if you pause your scrubs, it could be enough.
Or, if you want, I can add an override flag to ignore this check :)

(v15.2.16 is octopus, not pacific, right?)

[TheJJ]

Due to inactivity I'm gonna close this for now. If you have further questions, please just answer here or come by in the Matrix chat!