Created attachment 15353 [details]
patch for Samba 4.10 (backported/cherry-picked from master)
13978 relies on 13959 and the narrative behind both bugs is now confusing as I made a few mistakes while learning how to backport properly.
This bug contains patches for backports for 4.9.11 and 4.10.6 with all necessary stuff from 13978 and 13959 in the right order.
Created attachment 15354 [details]
patch for Samba 4.9 (backported/cherry-picked from master)
There's an issue with the 4.9.11 patch that I'm following up on but the 4.10.6 passed on devel and catalyst gitlabs:
The 4.10 patch still applies so that can go in, as is, with the next release. @abartlet and @garming please pass review on the 4.10 patch if you agree.
Info on 4.9 to come.
We need this in Samba 4.10 (and ideally Samba 4.9 if we understand why Aaron was seeing CI failures) because otherwise users doing a downgrade without first using the script from bug 14059 will see segfaults.
Pushed to autobuild-v4-10-test.
(In reply to Karolin Seeger from comment #4)
The further info on the 4.9 patch is what while Aaron is quite sure he has seen a convincing pattern that this patch series makes the GitLab CI flap more often when the final patch lands, the CI jobs that he sees failing more often don't even run the new test seen in the last commit.
I've looked at the failures and they are
- schema failures due to races we have seen often and finally fixed in master
- the nbt.dgram flapping test
- the KCC race between the explicit test and the server-triggered kcc run. The test just returns 'busy' to the request to run it while running.
- one runtime exceeded issue
Aaron is trying to get more data over the weekend.
I initially got 5 failures in a row from a 4.9.11 branch with this bug's patch set on, and 2 passes from samba-v4.9.11 HEAD. That lead me to believe there was a problem, so I tried manually bisecting the patch set using CI but the failure pattern wasn't strong enough to give me any certainty as to the offending patch. I did see some failures on 4.9.11 HEAD. Andrew believes what I'm seeing is merely the pre-existing instability of 4.9, and further testing has made that theory more likely. I'm a little uneasy but it's probably unreasonable to insist on holding this back, so let's push it.
Created attachment 15426 [details]
patch for Samba 4.9 (backported/cherry-picked from master) v2
This updated patch addresses some whitespace issues in the part of the patch that couldn't be directly backported.
(In reply to Andrew Bartlett from comment #7)
Pushed to autobuild-v4-9-test.
Pushed to v4-9-test.
Closing out bug report.