On 2024-01-07 21:59, Paul Gevers wrote: > Hi, > > On 07-01-2024 18:21, Aurelien Jarno wrote: > > > Timeout while building: > > > https://ci.debian.net/packages/g/glibc/unstable/arm64/41516611/ > > > > There are indeed many failures with cc1 getting killed, it seems that it > > started around 2024-01-02. I haven't spotted any change to the toolchain > > nor kernel version. > > > > I am not able to reproduce the issue, so it is very likely linked to > > debci. Would it be possible to now why cc1 get killed? OOM? timeout? > > I extracted the attached journal piece indicating OOM. > Thanks that's very useful. It appears that the tgmath3-fma test got killed after using 2611964kB. In practice it needs a few MB more, and that's not something that has changed recently, it needs about the same amount of memory with gcc-12 from bookworm. This test is known to take a lot of memory to compile, and on amd64, it needs around 3.2GB. I am still not sure why it got killed on arm64 and not on other debci workers, there as still swap available. Actually looking at the munin plot, it seems that the arm64 debci workers stopped using swap around September 2023 contrary to the other architectures. The good news is that it seems that gcc-13 needs a bit less RAM to compile that file. We can switch to it once we get glibc 2.38 into testing. Regards Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurelien@aurel32.net http://aurel32.net
Attachment:
signature.asc
Description: PGP signature