Den tors 6 feb. 2020 kl 15:06 skrev Mario Giammarco <mgiammarco@gmail.com>:
Hello, if I have a pool with replica 3 what happens when one replica is corrupted?
The PG on which this happens will turn from active+clean to active+inconsistent.
I suppose ceph detects bad replica using checksums and replace it with good one
There is a "osd fix on error = true/false" setting (whose name I can't remember right off the bat now) which controls this. If false, you need to "ceph pg repair" it, then it happens as you describe.
If I have a pool with replica 2 what happens?
Same. Except with repl=2, you run a higher chance of surprises* on the remaining replica while the first one is bad until it gets repaired. *) ie, data loss, tears and less sleep for ceph admins -- May the most significant bit of your life be positive.