Skip to content

Conversation

@bernardbeckerman
Copy link
Contributor

Summary:

Adds a robust wrapper method for poll_trial_status that:

  1. If poll_trial_status fails, falls back to polling trials one-at-a-time.
  2. Each trial for which individual status polling fails is marked as abandoned.

This is important for runners whose status-polling logic can undergo sporadic exceptions, in order to avoid failing the entire orchestrator process.

Alternatives considered:

  • Institute a PollStatusResult analogous to MetricFetchResult. Decided against because I don't think this logic belongs in runner.poll_trial_status, since it's good to have this centralized instead of replicated across all runners' poll_trial_status methods. Also this would require migration across all runner.poll_trial_status methods.
  • Add this to Orchestrator.poll_trial_status (V1 of this diff). This could make sense because this logic is mainly for use in the Orchestrator anyway, but I prefer keeping retry logic closer to where polling happens.

Also considered making this configurable, but opted for an opinionated solution that avoids config bloat.

Differential Revision: D90532278

@meta-codesync
Copy link

meta-codesync bot commented Jan 12, 2026

@bernardbeckerman has exported this pull request. If you are a Meta employee, you can view the originating Diff in D90532278.

@meta-cla meta-cla bot added the CLA Signed Do not delete this pull request or issue due to inactivity. label Jan 12, 2026
Summary:
Adds a robust poll_trial_status method to Orchestrator that:
1. If runner.poll_trial_status fails, falls back to polling trials one-at-a-time.
2. Each trial for which individual status polling fails is marked as abandoned.

This is important for runners whose status-polling logic can undergo sporadic exceptions, in order to avoid failing the entire orchestrator process.

Alternatives considered:
* **Institute a PollStatusResult analogous to [MetricFetchResult](https://fburl.com/code/yyp9j66m).** Decided against because I don't think this logic belongs in runner.poll_trial_status, since it's good to have this centralized instead of replicated across all runners' poll_trial_status methods. Also this would require migration across all runner.poll_trial_status methods.
* **Add a new Runner method robust_poll_trial_status (D90390842)**. It could make sense to keep retry logic closer to where polling happens, but realistically it's only used here.

Also considered making this configurable, e.g., via a boolean flag `OrchestratorOptions.should_poll_status_robustly`, but opted for an opinionated solution that avoids config bloat.

Differential Revision: D90532278
@codecov-commenter
Copy link

Codecov Report

❌ Patch coverage is 98.18182% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 96.72%. Comparing base (7bce65b) to head (8ae3ef3).

Files with missing lines Patch % Lines
ax/orchestration/orchestrator.py 94.44% 1 Missing ⚠️
Additional details and impacted files
@@           Coverage Diff           @@
##             main    #4756   +/-   ##
=======================================
  Coverage   96.72%   96.72%           
=======================================
  Files         583      583           
  Lines       60860    60914   +54     
=======================================
+ Hits        58866    58921   +55     
+ Misses       1994     1993    -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

@meta-codesync
Copy link

meta-codesync bot commented Jan 15, 2026

This pull request has been merged in f575efe.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed Do not delete this pull request or issue due to inactivity. fb-exported Merged meta-exported

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants