Bug#1003619: skimage: autopkgtest usually fails on armhf - out of memory
Paul Gevers
elbrus at debian.org
Wed Jan 12 20:43:16 GMT 2022
Control: reassign -1 src:skimage
Control: severity -1 important
Control: found -1 0.18.1-1
Hi,
On 12-01-2022 19:25, Rebecca N. Palmer wrote:
> The skimage autopkgtest usually, but not always, fails on armhf with
> several out-of-memory errors:
> https://ci.debian.net/packages/s/skimage/testing/armhf/
Flaky test are normally RC (but see below). Please fix the test (or
disable it on armhf via the "Architecture" field in the d/t/control file
[1]).
> It currently happens that the last migration-reference passed and the
> last few non-references failed, so this counts as an autopkgtest
> regression. I consider it more likely that the server just happened to
> have more RAM free when it ran the reference, but I don't have proof of
> that.
Our host that runs armhf tests has 255 GB RAM, so I doubt it's free RAM
(we never come close to full use:
https://ci.debian.net/munin/ci-worker-armel-01/ci-worker-armel-01/memory.html)
This looks more like it's reaching the limits of the *reachable* memory
on 32 bits architectures. See e.g. the discussion we had in
https://github.com/pytroll/satpy/issues/1883 (near the end)
That said, it seems that version 0.18.1-1 was most of the time passing
until 2021-06-13. I think that's about the time that we switched hosts
for our arm64 tests. The new host is *more* powerful and has *more*
memory; 140 cores and as said 255 GB. As the maintainer of ci.d.n I
offer my help to debug the issue. I can run the test on our host, is
there something I can run or collect in the testbed that would help you
debug the issue?
Paul
[1]
https://salsa.debian.org/ci-team/autopkgtest/-/raw/master/doc/README.package-tests.rst
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 495 bytes
Desc: OpenPGP digital signature
URL: <http://alioth-lists.debian.net/pipermail/debian-science-maintainers/attachments/20220112/b3d5f7c0/attachment.sig>
More information about the debian-science-maintainers
mailing list