@@ -19,10 +19,6 @@ name: GPU Tests (merge queue)
1919on :
2020 merge_group :
2121 workflow_dispatch :
22- # TEMP(testing): run on pushes to this branch pre-merge (no merge queue needed to test the
23- # rent -> test -> destroy path). REMOVE before merging.
24- push :
25- branches : [ci_run_tests_gpu]
2622
2723permissions :
2824 contents : read
4238 # Pin the Vast CLI to an immutable commit (a PyPI version can be re-published; a commit
4339 # hash can't) — avoids pulling untrusted code at run time.
4440 VAST_CLI_COMMIT : " 28494d92c6c03d887f8375085243c22eb68c5874"
45- # TEMP(debugging): "1" skips teardown so the box stays up for SSH debugging. The
46- # "Connection info" step prints how to connect and how to destroy it manually.
47- # SET BACK TO "0" (or remove) so the box is destroyed again.
48- KEEP_INSTANCE : " 1"
4941
5042jobs :
5143 gpu-tests :
@@ -295,31 +287,9 @@ jobs:
295287 fi
296288 } >> "$GITHUB_STEP_SUMMARY"
297289
298- # TEMP(debugging): when KEEP_INSTANCE=1, leave the box up and print how to reach it.
299- - name : Connection info (instance kept for debugging)
300- if : always() && env.KEEP_INSTANCE == '1'
301- env :
302- HOST : ${{ steps.ssh.outputs.host }}
303- PORT : ${{ steps.ssh.outputs.port }}
304- IID : ${{ steps.instance.outputs.id }}
305- run : |
306- {
307- echo "## ⚠️ Instance KEPT for debugging (KEEP_INSTANCE=1)"
308- echo "SSH in with your team key (baked into the box by the template onstart):"
309- echo '```'
310- echo "ssh -o StrictHostKeyChecking=accept-new -p ${PORT:-?} root@${HOST:-?}"
311- echo "cd /workspace/lambda_vm # the failing tests live here"
312- echo '```'
313- echo "Destroy it when done (it bills hourly):"
314- echo '```'
315- echo "vastai destroy instance ${IID:-?} --yes # label: $RUN_LABEL"
316- echo '```'
317- } | tee -a "$GITHUB_STEP_SUMMARY"
318- echo "::warning::Instance $IID kept for debugging — destroy it manually: vastai destroy instance $IID --yes"
319-
320290 # --- Teardown: ALWAYS destroy the instance (cost guardrail) ---
321291 - name : Destroy instance
322- if : always() && env.KEEP_INSTANCE != '1'
292+ if : always()
323293 run : |
324294 # Retry transient failures (network/auth) so a paid box isn't stranded.
325295 # --yes: skip the interactive [y/N] confirm (CI has no tty).
0 commit comments