Possibility to track all containerized OSDs on host

[Dyadya-Sasha]

Hello Taodd!
Is it possible somehow to track with osdtrace all contrainers with OSD running inside?
Now I need to specify exact PID to track only one OSD and this makes the whole picture unseen.

[taodd]

I think it's possible, either by feeding multiple OSD IDs or automatically detecting all OSD process IDs, but you have to make sure all processes are running the same version

[Dyadya-Sasha]

but how to specify multiple PIDs ? If I set it as ./osdtrace -p 2474017 -p 2478082 -i my_dwarf.json -x --skip-version-check it tacks only first PID, which is here 2474017

[taodd]

Sorry for the confusion, the functionality to specify multiple IDs is not implemented yet. We'll add it to our to-do list. cc to @pponnuvel

[pponnuvel]

Currently, osdtrace doesn't take multiple PIDs. We can enhance it so that it takes multiple PIDs (e.g, comma-separated list).

@Dyadya-Sasha how do you deploy the "containerised OSDs"? Multiple lxc containers/qemu on the same host? Just asking so that we can test it in the same way when supporting multiple PIDs.

[Dyadya-Sasha]

Hello guys!

Actually this is rook deployment. So ceph is in kubernetes. Another cluster is deployed with cephadm. THis is what I mean with "containerized OSDs"

[pponnuvel]

Thanks @Dyadya-Sasha I'll use Rook to test this when implementing this.