Created attachment 1326062 [details]
PUT success

I have a potential fix in the works. Just making sure I don't destroy caching or do other potential things. Should have a PR up sometime early next week I think.

I haven't been able to successfully reproduce the bug but from looking at the logs/code and talking to people I think we just have a place where we aren't refreshing the cache. Being careful since I'm unfamiliar here.

This bug is targeted for RHCEPH 2.5 and this fix is not in RHCEPH 3.

Would you please cherry-pick the change to ceph-3.0-rhel-patches (with the RHCEPH 3 clone ID number, "Resolves: rhbz#1530801") so customers do not experience a regression?

In accordance with our current theory the setup is:

A cluster with at least three RGWs
Under heavy enough load that we get occaisonal failed to transmit cache errors

And to reproduce:

Make metadata changes to the same buckets through multiple RGWs (such as the POST mentioned in the original filing)
The behavior to look for is a 500 error that /persists/ until one of the RGWs is restarted
The log should mention receiving -ECANCELED

If the fix works:

You should not get the 500 error.
(It's still in principle possible if you end up with enough races that it exceeds the retry count, but even in that case, the failure shouldn't be /durable/)

Thanks Adam!

I clicked on verified and selected 'Any' but it still shows the warning about the doc text not being verified?

Oh. Well.

Ack!

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-2018:0340