ci: run s390x job

fanquake commented at 1:40 pm on September 19, 2025: member

How fast does this run? (-md).

DrahtBot added the label Tests on Sep 19, 2025

DrahtBot commented at 1:40 pm on September 19, 2025: contributor

The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.

Code Coverage & Benchmarks

For details see: https://corecheck.dev/bitcoin/bitcoin/pulls/33436.

Reviews

See the guideline for information on the review process. A summary of reviews will appear here.

Conflicts

Reviewers, this pull request conflicts with the following ones:

#33302 (ci: disable cirrus cache in 32bit arm job by willcl-ark)

If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.

willcl-ark commented at 2:54 pm on September 19, 2025: member

I think this needs qemu:

 0commit 766b0b32361ae2eca3ee0b85caf1e27701fc0275
 1Author: will <will@256k1.dev>
 2Date:   Fri Sep 19 15:47:01 2025 +0100
 3
 4    use qemu
 5
 6diff --git a/.github/actions/configure-docker/action.yml b/.github/actions/configure-docker/action.yml
 7index c78df86b6cf..db5001c140d 100644
 8--- a/.github/actions/configure-docker/action.yml
 9+++ b/.github/actions/configure-docker/action.yml
10@@ -4,9 +4,16 @@ inputs:
11   use-cirrus:
12     description: 'Use cirrus cache'
13     required: true
14+  use-qemu:
15+    description: 'Use qemu'
16+    default: false
17 runs:
18   using: 'composite'
19   steps:
20+    - name: Set up QEMU
21+      if: ${{ inputs.use-qemu }}
22+      uses: docker/setup-qemu-action@v3
23+
24     - name: Set up Docker Buildx
25       uses: docker/setup-buildx-action@v3
26       with:
27diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
28index 6869c5cf25d..cf5ae7c717e 100644
29--- a/.github/workflows/ci.yml
30+++ b/.github/workflows/ci.yml
31@@ -481,6 +481,7 @@ jobs:
32             fallback-runner: 'ubuntu-24.04'
33             timeout-minutes: 120
34             file-env: './ci/test/00_setup_env_s390x.sh'
35+            use-qemu: true
36 
37           - name: 'CentOS, depends, gui'
38             cirrus-runner: 'ghcr.io/cirruslabs/ubuntu-runner-amd64:24.04-lg'
39@@ -520,6 +521,7 @@ jobs:
40         uses: ./.github/actions/configure-docker
41         with:
42           use-cirrus: ${{ needs.runners.outputs.use-cirrus-runners }}
43+          use-qemu: ${{ matrix.use-qemu }}
44 
45       - name: Enable bpfcc script
46         if: ${{ env.CONTAINER_NAME == 'ci_native_asan' }}

use qemu 6f69cc77f7

ci: run s390x job 69474fd616

fanquake force-pushed on Sep 19, 2025

willcl-ark commented at 3:00 pm on September 19, 2025: member

Or we just enable qemu on all jobs (no downsides that I know of) with:

 0commit 6e7fba0b93d1dec0deafe7b7f87054c9a629134f
 1Author: will <will@256k1.dev>
 2Date:   Fri Sep 19 15:47:01 2025 +0100
 3
 4    use qemu
 5
 6diff --git a/.github/actions/configure-docker/action.yml b/.github/actions/configure-docker/action.yml
 7index c78df86b6cf..b5f3601a24b 100644
 8--- a/.github/actions/configure-docker/action.yml
 9+++ b/.github/actions/configure-docker/action.yml
10@@ -7,6 +7,9 @@ inputs:
11 runs:
12   using: 'composite'
13   steps:
14+    - name: Set up QEMU
15+      uses: docker/setup-qemu-action@v3
16+
17     - name: Set up Docker Buildx
18       uses: docker/setup-buildx-action@v3
19       with:

which avoids passing options to the configure-docker action and things

[wip] ci: exclude qt from s390x job f379d597a7

fanquake commented at 4:40 pm on September 19, 2025: member

Thanks @willcl-ark. Pulled in (some version) of your change, and added a NO_QT=1 for now.

fanquake commented at 5:04 pm on September 19, 2025: member

Looks like this passed in 98m (without Qt and nothing cached).

maflcko commented at 8:32 am on September 23, 2025: member

Will the be useful overall? I think it is clear that no developer is using this platform right now (maybe not even a user), so the benefit of the CI config is mostly a sanity check. The issues it finds seems to happen less than once per year:

#26820
#30587
Possibly #27529, albeit the CI task was only later adjusted to find it

All the issues were trivial to fixup individually as a follow-up. However, the task here (even with a full depends cache) will take 60 minutes with an empty ccache (40 minutes compile + 20 minutes tests).

No strong opinion, but I think some tasks are fine as nightly tasks. It should be trivial to take the actions from this repo and call them from one of the nightly CI repos. C.f. https://github.com/bitcoin-core/qa-assets/blob/db30a6290f4070a31572801e2fd62c8876282ecc/.github/workflows/ci.yml#L56-L75

in .github/actions/configure-docker/action.yml:15 in f379d597a7

10 runs:
11   using: 'composite'
12   steps:
13+    - name: Set up QEMU
14+      if: ${{ inputs.use-qemu }}
15+      uses: docker/setup-qemu-action@v3

maflcko commented at 8:38 am on September 23, 2025:

Will this cache anything?

According to https://github.com/bitcoin/bitcoin/actions/caches

I see:

docker.io–tonistiigi–binfmt-latest-linux-x64 30 MB cached 4 days ago refs/pull/33436/merge Last used 4 days ago

Which seems a bit odd, but I am not familiar with GHA, nor the docker/ actions.

willcl-ark commented at 9:49 am on September 23, 2025: member

Will the be useful overall? I think it is clear that no developer is using this platform right now (maybe not even a user), so the benefit of the CI config is mostly a sanity check.

My understanding of this job is that it represents all Big Endian systems, and therefore, as you say just exists as a sanity check (that we don’t have any endianness bugs). If nobody uses any BE systems (IBM Z, older SPARC, some powerpc) perhaps it’s not even useful in that role though…

I’d agree with having this run as a nightly (or even less frequent) job. Would also note that secp256k1 are already running an s390x task:

https://github.com/bitcoin-core/secp256k1/blob/de6af6ae3568461edc9967500d7a70beeb66a642/.github/workflows/ci.yml#L161-L164

…so critical cryptography code is already being tested for endianness bugs that way. Perhaps that’s enough testing for us?

DrahtBot commented at 1:35 pm on September 23, 2025: contributor

🐙 This pull request conflicts with the target branch and needs rebase.

DrahtBot added the label Needs rebase on Sep 23, 2025

l0rinc commented at 5:20 pm on September 23, 2025: contributor

The issues it finds seems to happen less than once per year

For the record a big-endian CI machine would have helped with #31144

maflcko commented at 10:46 am on September 24, 2025: member

For the record a big-endian CI machine would have helped with #31144

It should be trivial to run the CI task locally, on demand. For development/debugging that should even be easier than to rely on a remove short-lived server. It should also be easy to temporarily add the config to the CI to trigger it on demand, if needed, similar to *BSD stuff, like #33435 (review).

Again, I am not against adding this. My only concern is that this would likely be one of the slowest tasks. With the given timeout it doesn’t even finish on GHA at all right now.

powerpc

I am sure there are some ppc users, so the s390x CI config must be kept and maintained and run before every release at least once. My only concern is that the overhead to run this on every pull request is more than the benefit it provides. (Similar to how there are no *BSD CI tasks on pull requests)

ci: run s390x job #33436

Code Coverage & Benchmarks

Reviews

Conflicts