Hello,
I'm sure the technical details can be ironed out, but the question is
more whether someone is willing to do the maintenance work required to
keep Arrow working on big-endian platforms, and if possible enable it
for more components (most of us don't have access to such a platform).
If a CI build is red half of the time, then there is not much of a point
to it.
Regards
Antoine.
Le 12/05/2025 à 10:48, Raúl Cumplido a écrit :
Hi,
Thanks for pushing this, adding s390x to our testing matrix sounds
interesting to me. We have had discussions in the past about ephemeral
runners [1] and the potential security risks. In that issue there was a
discussion about how the pytorch project sets those "ephemeral" runners,
see their workflow here [2]. Do you know if the idea is to follow a similar
approach as described on that issue?
Regards,
Raúl
[1] https://github.com/apache/arrow/issues/43817#issuecomment-2619171798
[2]
https://github.com/pytorch/pytorch/blob/47df195065d62a649169565293b24595249ca3b1/.github/workflows/s390.yml#L25
El jue, 8 may 2025 a las 20:32, ISHIZAKI Kazuaki (<ki...@acm.org.invalid>)
escribió:
Hi,
CI for s390x on Apache Arrow has not been enabled since we previously
relied on the unstable Travis environment.
Now, IBM offers permanent virtual instances for open source projects
[1]. I’m considering using these instances for s390x CI as a
self-hosted runner for GitHub Actions. I’ve already obtained an
instance.
There are two possible approaches for setting it up:
a. Using the ursacomputing/crossbow repository [2]
b. Going through the ASF infrastructure [3]
Based on @Raul’s suggestion, option (a) would likely be the simpler path.
I’d appreciate any comments or suggestions you may have.
[1] https://community.ibm.com/zsystems/l1cc/
[2] https://github.com/apache/arrow/issues/46014
[3] https://issues.apache.org/jira/browse/INFRA-24268
Best Regards,
--
Kazuaki Ishizaki, IBM Research - Tokyo
Mail: ki...@acm.org
