Not presented in acting set.

[Nivekiba]

Hello @ljflores,
I try your solution (because it's simple to use) and I sometimes got this error:

like the system try to set an OSD not presented in the acting set of a pg as primary.
don't know if it's normal or if I used it badly.

[ljflores]

Hey @Nivekiba, this error message is meant to come up if you try to set an OSD as primary in a PG that's not in the acting set.

For instance, if you run: ceph osd pg-upmap-primary 2.19 7, but the acting set for pg 2.19 is [1,4,5], you will get this error message since osd.7 is not part of the acting set. (It needs to already be part of the PG's acting set to become primary, as this guarantees no data movement).

If this happens again, try to capture these pieces of information:

1. The error output, as you attached above
2. The output of ceph pg dump pgs_brief (this will give us the latest acting sets for each PG in your cluster)
3. Attach the osdmap you used to apply the read balancer commands (the one which resulted in error messages)

My guess is that the osdmap may have rebalanced capacity-wise after you initially grabbed the osdmap, resulting in the read balancer referring to acting sets that no longer exist.

[ljflores]

By the way, thanks for your feedback and for trying out the read balancer!

[ljflores]

Hey @Nivekiba can you attach the osdmap file you used to run the command so I can have a closer look?

[ljflores]

@JoshSalomon

[Nivekiba]

Hello @ljflores, actually i moved on another part of my work so I don't have the osdmap with me. Since it's a bug not appearing all the time, I will try to reproduce it as soon as possible and get back to you with the osdmap.

I've another sneaky question, I'm working actually on the ceph source code, especially the objecter source code but it is difficult since there's no dev doc about this component and in the slack people are busy I guess, do you know someone in particular to whom I can write ?

[ljflores]

Hello @ljflores, actually i moved on another part of my work so I don't have the osdmap with me. Since it's a bug not appearing all the time, I will try to reproduce it as soon as possible and get back to you with the osdmap.

Sounds good.

I've another sneaky question, I'm working actually on the ceph source code, especially the objecter source code but it is difficult since there's no dev doc about this component and in the slack people are busy I guess, do you know someone in particular to whom I can write ?

I would check the commit history on the files you're looking into. Whoever has contributed most recently would be good to reach out to.