Merge ~sergiodj/ubuntu/+source/squid:memory-leak-connect-tunnel-closed into ubuntu/+source/squid:ubuntu/jammy-devel
Status: | Merged | ||||
---|---|---|---|---|---|
Approved by: | git-ubuntu bot | ||||
Approved revision: | not available | ||||
Merged at revision: | d1f45a9a3e451979516fafc71c6bf1dbe4949217 | ||||
Proposed branch: | ~sergiodj/ubuntu/+source/squid:memory-leak-connect-tunnel-closed | ||||
Merge into: | ubuntu/+source/squid:ubuntu/jammy-devel | ||||
Diff against target: |
76 lines (+54/-0) 3 files modified
debian/changelog (+8/-0) debian/patches/close-tunnel-if-to-server-conn-closes-after-client.patch (+45/-0) debian/patches/series (+1/-0) |
||||
Related bugs: |
|
Reviewer | Review Type | Date Requested | Status |
---|---|---|---|
git-ubuntu bot | Approve | ||
Bryce Harrington (community) | Approve | ||
Canonical Server Reporter | Pending | ||
Review via email: mp+436158@code.launchpad.net |
Description of the change
This MP fixes a memory leak that's affecting squid on Jammy.
Unfortunately, it's yet another one of those bugs that's hard to reproduce locally due to its non-determinism. I couldn't reproduce it, but fortunately the reporter has been very helpful and was able not only to make the memory leak happen consistently but also to confirm that this fix I'm proposing indeed solves the problem.
You can also check upstream's discussion and lengthy investigation here: https:/
As you can see, the patch is very simple and functionally just adds a call to retryOrBail, which acts like a destructor freeing up unneeded resources.
There's a PPA with the proposed changes here:
https:/
autopkgtest results:
Results: (from http://
squid @ amd64:
19.01.23 23:21:01 Log 🗒️ ✅ Triggers: squid/5.
squid @ arm64:
19.01.23 23:19:07 Log 🗒️ ✅ Triggers: squid/5.
squid @ ppc64el:
19.01.23 23:07:15 Log 🗒️ ✅ Triggers: squid/5.
squid @ s390x:
19.01.23 23:03:24 Log 🗒️ ✅ Triggers: squid/5.
I retriggered the test on armhf because it failed there due to flakiness.
The fix in question is indeed simple - just a one-liner - but it does change behavior on server closure so side-effects could be worth looking for, as is noted in LP: #1989380.
The upstream bug report itemizes a very detailed analysis and identification of the flaw, and as mentioned in the SRU text it sounds like we can rely on them if there is indeed a regression. The LP bug reporter also appears actively involved so we can depend on them for helping with the verification.
I'm going to re-read the analysis just to understand it better, but otherwise everything looks good. Packaging is solid, as is the testing and SRU text. +1 for upload.