Travis Build gets stalled for ARM64 job in pandas package

Hi Team,

I have been working on adding the ARM64 job is Travis-CI for the pandas package. I have added successfully and the job is getting started. But build is getting stalled for arm64 job and giving timeout error “No output has been received in the last 10m0s, this potentially indicates a stalled build or something wrong with the build itself.” in the testing phase, whereas it is running successfully for AMD64 jobs. Please have a look at travis build report: https://travis-ci.org/github/ossdev07/pandas/jobs/664711811.

I have raised a PR(https://github.com/pandas-dev/pandas/pull/30641) for adding arm64 jobs in Travis for pandas package and maintainer has suggested that issue facing is not an issue with pandas package.

Please suggest what needs to be done for resolving this issue. Please revert if any more information is required. Thanks.

Thanks for the reply.
May I know how can we resolve this issue? Thanks.

What helped in my case was

  1. breaking the ARM64 job into smaller jobs
  2. using travis_wait on each script command

Due to truncation, the last command in the log isn’t necessarily the one that timed out.

Just sharing Ruby project’s status, The arm64 is sometimes still very slow.

https://travis-ci.org/github/ruby/ruby/builds/668226170

  • amd64 (x86_64-linux): 9 minutes
  • amd64 (i686-linux): 11 minutes
  • arm64 (arm64-linux): 14 minutes
  • arm64 (arm32-linux): 1 hours 50 minutes (time out).

Thanks for the suggestion. But facing the same issue after adding travis_wait as well.

Hi @Michal,

It will be really helpful if you could help out with this issue. Thanks.

Could anyone please suggest on what can we done for resolving this issue? Thanks.

There are plenty of other posts discussing this; I tried to summarize them in the past here: Multiarch builds (lxd) always time out on failure instead of exit

The only thing that worked for me in all cases was to add || travis_terminate $? to all script lines (and to make sure there was no set -e in my .travis.yml anywhere). See here for the result: https://github.com/gap-system/gap/blob/master/.travis.yml#L184

Even without that, there are so many issues with arm64/ppc64le/s390x support in Travis, and absolutely no support (these issues have been reported many times now over several months, with ZERO reaction by the Travis team), that I am seriously considering to drop this again (as I have to constantly restart builds for these architectures which suffer from various temporary failures, e.g. fluke "quota exceeded error). Instead, I’ll look into using GitHub Actions with qemu – I just discovered https://github.com/marketplace/actions/run-on-architecture and will give a try.