Repository navigation
ci: test unsupported AOT runtimes in the build legs - #5811
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #5811 +/- ##
==========================================
- Coverage 76.13% 75.93% -0.20%
==========================================
Files 528 528
Lines 19579 19579
Branches 3829 3829
==========================================
- Hits 14906 14868 -38
- Misses 3791 3824 +33
- Partials 882 887 +5 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
1f30f05 to
a807d74
Compare
2a0f2f4 to
1634a0f
Compare
a807d74 to
44331a8
Compare
1634a0f to
3db1268
Compare
| path: integration-test | ||
|
|
||
| # Sentry turns SentryNative off on runtimes that have no native SDK. aot.Tests.ps1 checks that against this | ||
| # leg's own packages: linux-arm on the linux-arm64 runner, and win-x86 on the win-x64 runner. |
There was a problem hiding this comment.
aot.Tests.ps1 checks that against this leg's own packages
Similar issue to https://github.com/getsentry/sentry-dotnet/pull/5809/changes#r4383085676 but probably not as much of a concern here as I think AOT builds/tests for non-mobile code are less sensitive to changes to the runners. So OK to do it in this case - just worth noting we'd be breaking an otherwise long held and inflexible rule that we only/always test the same thing that we ship.
There was a problem hiding this comment.
Agreed. With the #5809 change, release runs test the shipped MAUI packages again. The unsupported AOT cases stay on the leg's own packages: whether SentryNative is on comes from a fixed RID list in the package's targets, not from what's bundled.
3db1268 to
4bf3776
Compare
44331a8 to
0552c5b
Compare
bitsandfoxes
left a comment
bitsandfoxes
left a comment
There was a problem hiding this comment.
I get the intention of the test and the benefit of guaranteeing that the fallback/backoff work. It just seems a bit odd to install packages and dependencies to run something somewhere it is not really supported, to then check that the thing that is not supported.. falls back to disabling something?
If you're happy, LGTM? 🤷♂️
|
Fair question! It's a regression test added in #4347 for a previous crash, so I'm currently only moving it away from the critical path. |
4bf3776 to
050b256
Compare
The Unsupported AOT jobs waited for the whole build matrix to use the macOS packages, but any leg's packages work. Every leg already packs and runs aot.Tests.ps1 against its own. linux-arm now runs in the linux-arm64 leg and win-x86 in the win-x64 leg, which removes two jobs and the tail after the matrix. Some arm64 hosts can't run 32-bit ARM binaries and failed with "Exec format error", so QEMU runs them on every host. Its binfmt_misc handler comes before the kernel's own loader, so it runs them even on hosts that could. On the first runs a small uname program confirmed that: it reported armv7l, which only QEMU does, on a host that runs 32-bit ARM natively. The leg now only checks that the handler is enabled. The action runs the tonistiigi/binfmt image privileged, so that image is pinned by digest here and in alpine.yml, like the action itself. On its first run, the AOT and container cases passed under QEMU in 38 seconds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
0552c5b to
cd1480d
Compare
Part 5 of 6 for #5806, based on #5810. The two Unsupported AOT jobs check that Sentry turns SentryNative off on runtimes without a native SDK, linux-arm and win-x86. They waited for the whole build matrix to test the macOS packages, and then added their own minutes after it. Any leg's packages work, though, and every leg already packs its own. So linux-arm now runs in the linux-arm64 leg and win-x86 in the win-x64 leg, and the two jobs are gone.
The linux-arm job was also flaky. 4 of 57 attempts failed, and the one log left showed "Exec format error", because some arm64 hosts can't run 32-bit ARM binaries.
docker/setup-qemu-action, pinned as inalpine.yml.unameprogram confirmed that, reportingarmv7lon such a host.This adds one dependency:
setup-qemu-actionpullstonistiigi/binfmtfrom Docker Hub on every linux-arm64 run, asalpine.ymlalready does. If Docker Hub is down, that leg fails at that step. The action runs that image privileged, so it's pinned by digest, here and inalpine.yml, the way the action itself is pinned.In #5791 the tail after the matrix was gone. On linux-arm64 the new steps took about a minute, and the AOT and container cases passed under QEMU.
On this PR's own CI
3 runs. On linux-arm64 the four new steps took 1.0 minute each time, 0.6 of it the linux-arm AOT case, and the handler check passed every time. On win-x64 the win-x86 case took 0.5 to 0.8. No job waits for every leg anymore. On
main, the two AOT jobs ran 1.7 minutes past the last leg. The whole stack's numbers are in #5806.#skip-changelog