Segmentation fault on GitHub's macos-14 runner

[rami3l]

Describe the bug

Hello there! Rustup maintainer here.

We use AWS S3 to manage our releases, and since recently when running aws --debug s3 cp --recursive using GitHub Actions' preinstalled aws CLI, the command might fail with a segfault.

The original Zulip post reporting this issue: https://rust-lang.zulipchat.com/#narrow/stream/242791-t-infra/topic/Strange.20error.20with.20Rustup's.20release.20process.20on.20macOS.20ARM64.

Expected Behavior

The command runs without issue and exit successfully.

Current Behavior

aws --debug s3 cp --recursive deploy/ s3://rustup-builds/5b5ec92726932b280e2f6e60f17df50644803a68
2024-08-06 13:56:38,328 - MainThread - awscli.clidriver - DEBUG - CLI version: aws-cli/2.17.18 Python/3.11.9 Darwin/23.5.0 exe/x86_64
2024-08-06 13:56:38,329 - MainThread - awscli.clidriver - DEBUG - Arguments entered to CLI: ['--debug', 's3', 'cp', '--recursive', 'deploy/', 's3://rustup-builds/5b5ec92726932b280e2f6e60f17df50644803a68']
2024-08-06 13:56:38,351 - MainThread - botocore.hooks - DEBUG - Event building-command-table.main: calling handler <function add_s3 at 0x112c41120>
[..]
2024-08-06 13:56:38,524 - ThreadPoolExecutor-1_0 - s3transfer.utils - DEBUG - Acquiring 0
2024-08-06 13:56:38,524 - ThreadPoolExecutor-0_9 - s3transfer.tasks - DEBUG - CompleteMultipartUploadTask(transfer_id=0, {'bucket': 'rustup-builds', 'key': '5b5ec92726932b280e2f6e60f17df50644803a68/dist/aarch64-apple-darwin/rustup-init', 'extra_args': {}}) about to wait for the following futures [<s3transfer.futures.ExecutorFuture object at 0x11370b890>, <s3transfer.futures.ExecutorFuture object at 0x11369cd10>, <s3transfer.futures.ExecutorFuture object at 0x1137156d0>]
2024-08-06 13:56:38,524 - ThreadPoolExecutor-0_9 - s3transfer.tasks - DEBUG - CompleteMultipartUploadTask(transfer_id=0, {'bucket': 'rustup-builds', 'key': '5b5ec92726932b280e2f6e60f17df50644803a68/dist/aarch64-apple-darwin/rustup-init', 'extra_args': {}}) about to wait for <s3transfer.futures.ExecutorFuture object at 0x11370b890>
2024-08-06 13:56:38,524 - ThreadPoolExecutor-1_0 - s3transfer.utils - DEBUG - Releasing acquire 0/None
/Users/runner/work/_temp/5fd0079c-2c9a-42bc-9d2f-99ce5c32b24d.sh: line 1:  3767 Segmentation fault: 11  aws --debug s3 cp --recursive deploy/ s3://rustup-builds/5b5ec92726932b280e2f6e60f17df50644803a68
Error: Process completed with exit code 139.

https://github.com/rust-lang/rustup/actions/runs/10267911140/job/28409617969

Reproduction Steps

Run the command under the macos-14 environment, and it will fail quite often.

Possible Solution

This could also be a problem with https://github.com/actions/runner-images updating to a faulty version of AWS CLI.

Please note that the repo above claims to have installed AWS CLI like so:

echo "Installing aws..."
awscliv2_pkg_path=$(download_with_retry "https://awscli.amazonaws.com/AWSCLIV2.pkg")
sudo installer -pkg "$awscliv2_pkg_path" -target /

https://github.com/actions/runner-images/blob/22143c7c6811f8936d42f49d964007df427788cb/images/macos/scripts/build/install-aws-tools.sh#L9-L11

Additional Information/Context

This doesn't seem to reproduce with other runners in our CI, so it could be ARM64 macOS-specific:

Current runner version: '2.317.0'
Operating System
  macOS
  14.5
  23F79
Runner Image
  Image: macos-14-arm64
  Version: 20240728.1
  Included Software: https://github.com/actions/runner-images/blob/macos-14-arm64/20240728.1/images/macos/macos-14-arm64-Readme.md
  Image Release: https://github.com/actions/runner-images/releases/tag/macos-14-arm64%2F20240728.1
Runner Image Provisioner
  2.0.374.1+4097a9592d27ce71de414581a65bffbda888dd1b

CLI version used

aws-cli/2.17.18 Python/3.11.9 Darwin/23.5.0 exe/x86_64 [sic]

Environment details (OS name and version, etc.)

macOS ARM64 14.5

[tim-finnigan]

Thanks for reaching out. I was not able to reproduce this issue. You mentioned that "it will fail quite often", could you provide more details on those failures? Can you provide steps for reproducing this issue consistently?

[rami3l]

Thanks for reaching out. I was not able to reproduce this issue. You mentioned that "it will fail quite often", could you provide more details on those failures? Can you provide steps for reproducing this issue consistently?

@tim-finnigan Sorry for my late update!

As far as I know it's really just a regular aws s3 cp call that involves multipart uploads, that's why I found this bug extremely confusing.

After a few days of investigation on my side, I've noticed that this problem is only reproducible when running a x64 version of AWS CLI under Rosetta 2, and after switching to the native version the problem seems gone: rust-lang/rustup#3989

I'm not sure why GitHub Actions is shipping the x64 version of AWS CLI though. Is that the expected behavior with https://awscli.amazonaws.com/AWSCLIV2.pkg?

Anyway, since this might not be a supported scenario, my best bet would be asking GitHub instead 🤔

Have a nice day :)

[github-actions]

This issue is now closed. Comments on closed issues are hard for our team to see.
If you need more assistance, please open a new issue that references this one.