[ci] flaky test: TestCheckpoint on almalinux-8

[lifubang]

=== RUN TestCheckpoint
time="2024-10-18T08:55:44Z" level=warning msg="--- Quoting "/tmp/TestCheckpoint214687474/003/criu-parent/dump.log""
time="2024-10-18T08:55:44Z" level=warning msg="118:(09.517977) freezer.state=FREEZING"
time="2024-10-18T08:55:44Z" level=warning msg="119:(09.618087) freezer.state=FREEZING"
time="2024-10-18T08:55:44Z" level=warning msg="120:(09.718192) freezer.state=FREEZING"
time="2024-10-18T08:55:44Z" level=warning msg="121:(09.818291) freezer.state=FREEZING"
time="2024-10-18T08:55:44Z" level=warning msg="122:(09.918412) freezer.state=FREEZING"
time="2024-10-18T08:55:44Z" level=warning msg="123:(10.001045) Error (criu/cr-dump.c:1779): Timeout reached. Try to interrupt: 0"
time="2024-10-18T08:55:44Z" level=warning msg="124:(10.001084) freezer.state=FREEZING"
time="2024-10-18T08:55:44Z" level=warning msg="125:(10.001125) Unfreezing tasks into 1"
time="2024-10-18T08:55:44Z" level=warning msg="126:(10.001128) \tUnseizing 45035 into 1"
time="2024-10-18T08:55:44Z" level=warning msg="127:(10.001140) Error (compel/src/lib/infect.c:418): Unable to detach from 45035: No such process"
time="2024-10-18T08:55:44Z" level=warning msg="128:(10.001144) Writing image inventory (version 1)"
time="2024-10-18T08:55:44Z" level=warning msg="129:(10.001223) Error (criu/cr-dump.c:1893): Pre-dumping FAILED."
time="2024-10-18T08:55:44Z" level=warning msg=---
checkpoint_test.go:93: criu failed: type PRE_DUMP errno 0
--- FAIL: TestCheckpoint (10.24s)

[lifubang]

I see it many times.

[rata]

It wouldn't be the first time criu becomes unreliable on old kernels. Maybe @kolyshkin has an insight on what to do (just skip it?)

[kolyshkin]

In general, yes, older kernels have issues when trying to freeze a cgroup, and as a result criu fails sometimes. We have a similar issue with runc itself (runc pause is flaky).

Both criu and runc retries (and I've changed the retry timings and attempts in runc at least twice), but sometimes it's still not enough.

The issue is probably the same as #4273.

[kolyshkin]

In general, yes, older kernels have issues when trying to freeze a cgroup

Let me correct myself: it's not older kernels, it's cgroup v1 freezer.