From bd4b0c6d07a7ce784562907e1c8e57c0991ae9bc Mon Sep 17 00:00:00 2001
From: Jim Schaff <schaff.jim@gmail.com>
Date: Wed, 24 Jun 2026 08:03:09 -0400
Subject: [PATCH] Stop release jobs hanging 6h on failure (gate tmate, add
 timeouts)

The release-main workflow ran a `Setup tmate` step on any failure
(`if: failure()`), which opens an interactive debug session that blocks the job
until the 6h runner limit. During the 0.0.16 release the PyPI publish step
failed (project size > 10GB), and tmate then held 7 jobs open for 6h each until
the whole run was cancelled.

- Gate the tmate step behind a `workflow_dispatch` boolean input `debug_tmate`
  (default false), so release events never open a session and failures fail
  fast; the session is still available for manual debug runs.
- Bound the tmate step with `timeout-minutes: 30`.
- Add a job-level `timeout-minutes: 180` as defense-in-depth (successful builds
  take ~35-60 min).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .github/workflows/on-release-main.yml | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/on-release-main.yml b/.github/workflows/on-release-main.yml
index 9c7ecf0..6be48a2 100644
--- a/.github/workflows/on-release-main.yml
+++ b/.github/workflows/on-release-main.yml
@@ -2,6 +2,11 @@ name: release-main
 
 on:
   workflow_dispatch:
+    inputs:
+      debug_tmate:
+        description: "Open an interactive tmate debug session if a job fails"
+        type: boolean
+        default: false
   release:
     types: [published]
     branches: [main]
@@ -22,6 +27,9 @@ jobs:
 
       fail-fast: false
     runs-on: ${{ matrix.os }}
+    # Successful builds take ~35-60 min; cap well under the 6h default so a stuck
+    # job (e.g. a hung publish) fails fast instead of burning a runner for 6 hours.
+    timeout-minutes: 180
     defaults:
       run:
         shell: bash
@@ -109,8 +117,13 @@ jobs:
           poetry publish --skip-existing
         if: github.event_name == 'release'
 
+      # Only open a tmate session when explicitly requested via workflow_dispatch.
+      # On a release event `inputs.debug_tmate` is false, so a failed job fails fast
+      # instead of holding the runner open until the 6h job timeout. The step is also
+      # bounded by timeout-minutes as a safety net.
       - name: Setup tmate
-        if: failure()
+        if: ${{ failure() && inputs.debug_tmate }}
+        timeout-minutes: 30
         uses: mxschmitt/action-tmate@v3
         with:
           limit-access-to-actor: true