On Tue, 26 Aug 2025 at 00:19:26 +0100, Luca Boccassi wrote:
The gcr4 autopkgtests seem to get stuck for hours until timing out in
the installed-tests test.
This was really at least two bugs:
1. test-ssh-agent-service (aka gcr:gcr-ssh-agent / service when run as a
build-time test) intermittently gets stuck and does not terminate
until killed by a timeout. I sometimes see this when building gcr4
locally, especially if built in parallel with other packages (a race
condition triggered by system load?), but it's intermittent.
2. debian/tests/installed-tests didn't put any timeout on individual
test-cases. Fix pending.
As a stopgap answer to the autopkgtests often failing (#1112076) and the
build intermittently failing (no bug report), I'm going to patch the
failing test-case to not run by default. As with similar workarounds in
GLib, it can be re-enabled by exporting DEB_ALLOW_FLAKY_TESTS=1.
(1.) is still a valid bug, so I've cloned #1112659 to represent it. I
don't have the domain knowledge or the spoons to dig further into it
right now - help welcome. A first step would be to reproduce it, get a
backtrace and report it upstream.
It seems that #1112659 is more reproducible on riscv64 than on other
architectures. If it's a race condition (as I suspect), then it might be
correlated with having a slower CPU or high system load, so that would
be a good place for someone to start investigating.
It looks as though test-ssh-agent-process might also be flaky. I'll open
a separate bug for that.
smcv
