Under some workloads, it is beneficial to avoid log-based recovery and use backfill instead. This lets I/O proceed on objects being backfilled, since it occurs in the background. This is implemented by setting the min/max pg log size to 1 and 2. This has a side effect of eliminating our duplicate op detection. To make this configuration robust, we need to keep a separate structure to detect duplicate ops.
This PR should likely go into this BZ as it fixes a bug in the dup ops code: https://github.com/ceph/ceph/pull/16744 At the time of this writing it hasn't been merged upstream yet.
Josh, my understanding is that this won't make it in 2.4 unless it's labeled a blocker. I can do more analysis to see it could create problems other than putting items in the dup index when it should not, but do you have a sense?
(In reply to Eric Ivancich from comment #5) > Josh, my understanding is that this won't make it in 2.4 unless it's labeled > a blocker. I can do more analysis to see it could create problems other than > putting items in the dup index when it should not, but do you have a sense? It just saves a tiny amount of cpu/memory in the rare case when we don't index anything. It's not necessary to backport the fix.
Marking this bug as verified as planned regression testing and bug verfication is done. Verified on 10.2.7-32.el7cp.x86_64
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2017:2640