I'm a part of the team that manages the ASF ATS CI images. I did a routine update of the ci.trafficserver.apache.org/ats/rockylinux:8 image to make it slimmer by cleaning up build artifacts. We noticed, however, that with that update, the certifier.test.py test started failing due to what looked like a shutdown crash from ATS. Through some investigation, I observed:
Start with the previous ci.trafficserver.apache.org/ats/rockylinux:8 docker image and run the certifier autest, it passes fine. The original image is available on docker1 in the CI system, which I have access to.
Now run sudo yum update and update the system packages.
Then rebuild ATS after a full git clean.
Then re-run the certifier autest. The test now fails due to the ATS crash.
@maskit and @bryancall both reproduced this crash on later Fedora versions, so this isn't a peculiarity of the docker image itself. It's just occasioned by some package update.
We agreed to merge that in temporarily to keep CI green so that PRs can be merged. However, we need to prioritize figuring out why that log results in a crash with that certifier test.
This issue has been automatically marked as stale because it has not had recent activity. Marking it stale to flag it for further consideration by the community.
I'm a part of the team that manages the ASF ATS CI images. I did a routine update of the ci.trafficserver.apache.org/ats/rockylinux:8 image to make it slimmer by cleaning up build artifacts. We noticed, however, that with that update, the certifier.test.py test started failing due to what looked like a shutdown crash from ATS. Through some investigation, I observed:
sudo yum update
and update the system packages.@maskit and @bryancall both reproduced this crash on later Fedora versions, so this isn't a peculiarity of the docker image itself. It's just occasioned by some package update.
@maskit found that the crash could be avoided by commenting out the debug log from a destructor in the certifier.cc file, see: https://github.com/apache/trafficserver/pull/9792
We agreed to merge that in temporarily to keep CI green so that PRs can be merged. However, we need to prioritize figuring out why that log results in a crash with that certifier test.