ci: OPS-3802: Increase trtllm timeout, skip unsupported tests (#7104)

Signed-off-by: Dillon Cullinan <dcullinan@nvidia.com>
Signed-off-by: Harrison King Saturley-Hall <hsaturleyhal@nvidia.com>
Co-authored-by: Harrison Saturley-Hall <hsaturleyhal@nvidia.com>

.github/workflows/release.yml

@@ -150,9 +150,9 @@ jobs:
       builder_name: b-${{ github.run_id }}-${{ github.run_attempt }}
       build_timeout_minutes: 120
       cpu_only_test_markers: '(pre_merge or post_merge) and trtllm and gpu_0'
-      cpu_only_test_timeout_minutes: 60
+      cpu_only_test_timeout_minutes: 10
       single_gpu_test_markers: '(pre_merge or post_merge) and trtllm and gpu_1'
-      single_gpu_test_timeout_minutes: 60
+      single_gpu_test_timeout_minutes: 90
       multi_gpu_test_markers: '(pre_merge or post_merge) and trtllm and (gpu_2 or gpu_4)'
       multi_gpu_test_timeout_minutes: 60
     secrets: inherit

tests/fault_tolerance/migration/test_trtllm.py

@@ -237,7 +237,9 @@ def test_request_migration_trtllm_aggregated(
                 )
 
 
-@pytest.mark.xfail(strict=False, reason="Prefill migration not yet supported")
+@pytest.mark.skip(
+    reason="Prefill migration not yet supported, XFail eats up CI time due to timeout"
+)
 @pytest.mark.timeout(350)  # 3x average
 @pytest.mark.nightly
 def test_request_migration_trtllm_prefill(