Ceph

09/14/2020

03:31 PM bluestore Bug #46124: Potential race condition regression around new OSD flock()s

I believe this bug was not fixed properly; the added commit "just retries 3 times".
This makes success more likely...

07/06/2020

10:30 PM bluestore Bug #46124: Potential race condition regression around new OSD flock()s

@Neha Ojha:
I think the bug remains as real as it gets; I did not retract that this is a bug. With my last comment...

06/21/2020

09:09 PM bluestore Bug #46124: Potential race condition regression around new OSD flock()s

> I suspect that Ceph starts other threads (using clone() on Linux) while the lock is held
Sorry, this should be t...

05:27 PM bluestore Bug #46124: Potential race condition regression around new OSD flock()s

From the strace above, we can see that there's always a `close()` after a matching `flock()` within the same PID, so ...

02:17 PM RADOS Feature #24099: osd: Improve workflow when creating OSD on raw block device if there was bluestor...

John Spray wrote:
> This seems like an odd idea -- if someone is doing OSD creation by hand, why would they want to ...

01:53 PM bluestore Bug #46124: Potential race condition regression around new OSD flock()s

Another question:
Would it not be better to use OFD locks (Open File Description locks), that is via ...

03:18 AM bluestore Bug #46124: Potential race condition regression around new OSD flock()s

In case it helps, here are `strace` invocations, each showing slightly different behaviour and error messages, that i...

03:08 AM bluestore Bug #46124: Potential race condition regression around new OSD flock()s

I did not experience that in Mimic.

03:08 AM bluestore Bug #38150: KernelDevice exclusive lock broken

I suspect this may have introduced a regression: #46124

03:07 AM bluestore Bug #46124 (Resolved): Potential race condition regression around new OSD flock()s

In #38150 and PR https://github.com/ceph/ceph/pull/26245, a new `flock()` approach was introuduced.
When I use `ce...