Jaap Eldering wrote:
This is to warn you that the 5.3.1 and 5.2.1 releases from a few days ago are broken due to a regression bug in runguard: https://github.com/DOMjudge/domjudge/issues/310
We're still investigating the precise cause before releasing a fix. As a workaround we recommend using 5.3.0 instead of 5.3.1, since the actual changes (besides this bug) are minimal. [...]
Thank you for sending this alert. I was in the process of installing 5.3.1 for the Southern California contest this weekend. I will install 5.3.0 instead.
Do you recommend I install the work-around patch for "runguard" on the judgehosts separately?
Regards, Marc
Hi Marc,
On 08/11/17 05:37, Marc Furon wrote:
Jaap Eldering wrote:
This is to warn you that the 5.3.1 and 5.2.1 releases from a few days ago are broken due to a regression bug in runguard: https://github.com/DOMjudge/domjudge/issues/310
We're still investigating the precise cause before releasing a fix. As a workaround we recommend using 5.3.0 instead of 5.3.1, since the actual changes (besides this bug) are minimal. [...]
Thank you for sending this alert. I was in the process of installing 5.3.1 for the Southern California contest this weekend. I will install 5.3.0 instead.
Do you recommend I install the work-around patch for "runguard" on the judgehosts separately?
If you install 5.3.0, then you only need to apply
https://github.com/DOMjudge/domjudge/commit/9bad6d6dd02542ebbfbccfa0115217d2...
The revert of
https://github.com/DOMjudge/domjudge/commit/bbd05895d83cff22b374a30e749e49ca...
is just undoing the commit that introduced the bug in 5.3.1.
However, I've only seen the problem that 9bad6d6dd02542ebb fixes occur on my own machine, so there's a high likelihood that it doesn't even affect you. If you want to test that, you should rejudge a contest (e.g. the test submissions for the demo contest) a couple of times (on a single judgehost) and see if you get a few TLEs instead of previous judgements. Alternatively, you can check if after judgement on the judgehost, there are still directories of the form domjudge/dj_cgroup_$PID under the cpuacct cgroup. That would typically be mounted under /sys/fs/cgroup/cpuacct, so the full path you're looking for is /sys/fs/cgroup/cpuacct/domjudge/dj_cgroup_$PID. If these directories exist after judgings are done, then you're most likely affected by the bug that 9bad6d6dd02542ebb fixes.
Unfortunately we don't have a good understanding yet what causes this regression, and how to fix it without reintroducing the original bug I fixed.
Jaap