fix: ignore SIGPIPE to prevent server crash on S3 idle connection timeout

[Its-Tanay]

What does the PR do?

This PR globally sets the SIGPIPE disposition to SIG_IGN inside the tritonserver executable.

This prevents an unhandled SIGPIPE from crashing the entire server with Exit Code 141 when the AWS C++ SDK attempts to write to an S3 keep-alive connection that has timed out during a slow model initialization. By ignoring the signal, the write() system call correctly returns EPIPE, allowing the AWS SDK to gracefully catch the error and reconnect to S3.

Checklist

• I have read the Contribution guidelines and signed the Contributor License Agreement
• PR title reflects the change and is of format <commit_type>: <Title>
• Changes are described in the pull request.
• Related issues are referenced.
• Populated github labels field
• Added test plan and verified test passes.
• Verified that the PR passes existing CI.
• I ran pre-commit locally (pre-commit install, pre-commit run --all)
• Verified copyright is correct on all changed files.
• Added succinct git squash message before merging.
• All template sections are filled out.
• Optional: Additional screenshots for behavior/output changes with before/after.

Commit Type:

• fix

Related PRs:

N/A

Where should the reviewer start?

src/triton_signal.cc -> RegisterSignalHandler()

Test plan:

Because this is an OS-level signal handler fix for a network idle timeout, standard unit tests cannot reliably cover it. I have verified this manually, and the exact deterministic reproduction steps using a dummy Python model (which forces the required 300s S3 idle timeout) are thoroughly documented in the attached GitHub issue.

Caveats:

None. The fix is explicitly placed in the standalone tritonserver executable (triton_signal.cc) rather than triton-core to ensure there are no global signal side-effects for users embedding Triton as a C++ library.

Background

This was discovered while loading a sequence of models from an S3 model repository, where one model took > 5 minutes to initialize on the GPU, causing the AWS connection to sit idle and be killed by the network layer.

Related Issues:

• Fixes Bug: tritonserver crashes with Exit Code 141 (SIGPIPE) when loading models from S3 due to idle connection timeout #8767

[Its-Tanay]

Hey folks, were you able to review this ? @pskiran1 @whoisj @yinggeh (sorry to bother)

[whoisj]

The change looks reasonable to me. Need to go through our internal lab's validation. I'll try to get that done today. Thank you for your patience.

[Its-Tanay]

Hey @whoisj, any updates on this ?

[whoisj]

CI Pipeline ID: 51849071

[Its-Tanay]

Thanks for running it @whoisj! Let me know if anything comes up. 🙇

[whoisj]

This change will be part of the Triton Inference Server 26.06 release.