threads / discuss / 61135

Failures in GitHub Actions linux-leaks and linux-asan-ubsan

Subject: Failures in GitHub Actions linux-leaks and linux-asan-ubsan

## tl;dr

4 messages between Mar 16, 2024 and Mar 18, 2024.

replies: 3people: 4as markdown or json

Philippe Blain· Mar 16, 2024, 19:20 UTC · lore

Hi everyone,

You might have noticed that the linux-leaks and linux-asan-ubsan jobs in GitHub Actions started failing last week. I investigated this so I'm sharing my findinds in case it helps.

The failures are due to the new ubuntu-22.04 GitHub Actions image (release 20240310.1.0, [1]) which uses a kernel where ASLR is configured in a way that is incompatible with ASan and LSan as used in the GCC and Clang versions in that image. More info can be found in [2] and [3] and pages linked there.

A workaround was already implemented in the image generation process [4], so the next version of the image should work. I think the images are released weekly. We could maybe add the same sysctl command to reduce the entropy to our YAML file, or we could live with it for the next week or so while waiting for the next image to roll out.

Cheers,

Philippe.

[1] https://github.com/actions/runner-images/commit/d67fa31aeeec3cf0d666d0eb2976b683471c6b90 [2] https://github.com/actions/runner-images/issues/9491#issuecomment-1989718917 [3] https://github.com/actions/runner-images/issues/9524#issuecomment-2002065399 [4] https://github.com/actions/runner-images/pull/9513

Junio C Hamano· Mar 16, 2024, 23:31 UTC · re: Philippe Blain · lore

Re: Failures in GitHub Actions linux-leaks and linux-asan-ubsan

Philippe Blain <levraiphilippeblain@gmail.com> writes:

Show 17 quoted lines

> Hi everyone,
>
> You might have noticed that the linux-leaks and linux-asan-ubsan
> jobs in GitHub Actions started failing last week. I investigated 
> this so I'm sharing my findinds in case it helps.
>
> The failures are due to the new ubuntu-22.04 GitHub Actions image
> (release 20240310.1.0, [1]) which uses a kernel where ASLR is configured
> in a way that is incompatible with ASan and LSan as used in 
> the GCC and Clang versions in that image. More info can be found
> in [2] and [3] and pages linked there.
>
> A workaround was already implemented in the image generation process 
> [4], so the next version of the image should work. I think the images
> are released weekly. We could maybe add the same sysctl command to reduce
> the entropy to our YAML file, or we could live with it for the next week
> or so while waiting for the next image to roll out.

Ah, just what I needed, as I was looking at the CI results and noticed these failed runs. Thanks.

Linus Arver· Mar 18, 2024, 06:52 UTC · re: Philippe Blain · lore

Re: Failures in GitHub Actions linux-leaks and linux-asan-ubsan

Philippe Blain <levraiphilippeblain@gmail.com> writes:

Show 5 quoted lines

> Hi everyone,
>
> You might have noticed that the linux-leaks and linux-asan-ubsan
> jobs in GitHub Actions started failing last week. I investigated 
> this so I'm sharing my findinds in case it helps.

Huge thanks for looking into this, Philippe!

Show 20 quoted lines

> The failures are due to the new ubuntu-22.04 GitHub Actions image
> (release 20240310.1.0, [1]) which uses a kernel where ASLR is configured
> in a way that is incompatible with ASan and LSan as used in 
> the GCC and Clang versions in that image. More info can be found
> in [2] and [3] and pages linked there.
>
> A workaround was already implemented in the image generation process 
> [4], so the next version of the image should work. I think the images
> are released weekly. We could maybe add the same sysctl command to reduce
> the entropy to our YAML file, or we could live with it for the next week
> or so while waiting for the next image to roll out.
>
> Cheers,
>
> Philippe.
>
> [1] https://github.com/actions/runner-images/commit/d67fa31aeeec3cf0d666d0eb2976b683471c6b90
> [2] https://github.com/actions/runner-images/issues/9491#issuecomment-1989718917
> [3] https://github.com/actions/runner-images/issues/9524#issuecomment-2002065399
> [4] https://github.com/actions/runner-images/pull/9513

I appreciate the links you've documented here for reference (and the care you took to write up [3]). You've got good taste. :)

Jeff King· Mar 18, 2024, 09:08 UTC · re: Philippe Blain · lore

Re: Failures in GitHub Actions linux-leaks and linux-asan-ubsan

On Sat, Mar 16, 2024 at 03:20:44PM -0400, Philippe Blain wrote:

Show 11 quoted lines

> The failures are due to the new ubuntu-22.04 GitHub Actions image
> (release 20240310.1.0, [1]) which uses a kernel where ASLR is configured
> in a way that is incompatible with ASan and LSan as used in 
> the GCC and Clang versions in that image. More info can be found
> in [2] and [3] and pages linked there.
> 
> A workaround was already implemented in the image generation process 
> [4], so the next version of the image should work. I think the images
> are released weekly. We could maybe add the same sysctl command to reduce
> the entropy to our YAML file, or we could live with it for the next week
> or so while waiting for the next image to roll out.

Thanks for digging into this! I had done a little but didn't get nearly as far. I am happy I can just ignore it and the problem will resolve itself. ;)

While I have the attention of folks who might be interested in CI failures, let me hijack the thread for a moment: has anybody figured out why macOS jobs sometimes time out after 6 hours? I assume that is an Actions limit, and something is just hanging. It sometimes hits osx-reftable[0], but sometimes osx-clang[1] and osx-gcc[2]. I've seen it on my builds and some on git/git (the last one is from git/git).

It's hard to tell which test is hanging, because the output only shows the finished tests. I tried running them without "prove" and doing it one-by-one, but then the hang doesn't seem to occur (so presumably it's a race under load). I tried comparing the list of tests reported as finishing versus the total list, but there was nothing enlightening. It's all stuff in t9xxx, which you'd expect to be running near the end anyway (and it feels like if there was _one_ test hanging, we should finish everything else, since we run with some parallelism). So it's almost like a bug in "prove" or something. But AFAIK it just started happening in the past month or two.

-Peff

[0]: example run: https://github.com/peff/git/actions/runs/8107092556/job/22158038562 [1]: example run: https://github.com/peff/git/actions/runs/8091601551/job/22110929146 [2]: example run: https://github.com/git/git/actions/runs/8273234891/job/22636626511