chore: fix post merge (#6704)

Signed-off-by: Dmitry Tokarev <dtokarev@nvidia.com>

.github/workflows/pre-merge.yml

@@ -61,6 +61,9 @@ jobs:
     - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065  # v5.6.0
     - uses: pre-commit/action@2c7b3805fd2a0fd8c1884dcaebf91fc102a13ecd # v3.0.1
       timeout-minutes: 3
+      # TODO: enable this once OPS-3006 is done
+      env:
+        SKIP: pytest-marker-report
 
   rust-tests:
     needs: changed-files

tests/fault_tolerance/cancellation/test_sglang.py

@@ -186,7 +186,7 @@ class DynamoWorkerProcess(ManagedProcess):
 
 @pytest.mark.timeout(160)  # 3x average
 @pytest.mark.gpu_1
-@pytest.mark.xfail(strict=False)
+@pytest.mark.skip(reason="DYN-2265")
 def test_request_cancellation_sglang_aggregated(
     request, runtime_services_dynamic_ports, predownload_models
 ):

tests/fault_tolerance/migration/test_trtllm.py

@@ -197,8 +197,8 @@ class DynamoWorkerProcess(ManagedProcess):
         return False
 
 
-@pytest.mark.timeout(290)
-@pytest.mark.post_merge  # 3x average
+@pytest.mark.timeout(290)  # 3x average
+@pytest.mark.post_merge
 def test_request_migration_trtllm_aggregated(
     request,
     runtime_services_dynamic_ports,

tests/frontend/test_vllm.py

@@ -93,6 +93,8 @@ class VllmWorkerProcess(ManagedProcess):
             "harmony",
             "--dyn-reasoning-parser",
             "gpt_oss",
+            "--max-model-len",  # this reduced max context window and amount of GPU memory allocated for context
+            "32768",
         ]
 
         env = os.environ.copy()
@@ -220,7 +222,7 @@ def _validate_chat_response(response: requests.Response) -> Dict[str, Any]:
     return response_json
 
 
-@pytest.mark.timeout(240)  # ~3x measured total (~70s/test), rounded up
+@pytest.mark.timeout(300)  # ~3x measured total (~70s/test), rounded up
 @pytest.mark.post_merge
 def test_reasoning_effort(
     request, start_services: ServicePorts, predownload_models

tests/router/test_router_e2e_with_sglang.py

@@ -426,6 +426,7 @@ def test_router_decisions_sglang_multiple_workers(
 @pytest.mark.post_merge
 @pytest.mark.parametrize("request_plane", ["tcp"], indirect=True)
 @pytest.mark.timeout(600)  # 10 min max (multi-GPU + DP startup variance)
+@pytest.mark.skip(reason="DYN-2265")
 def test_router_decisions_sglang_dp(
     request,
     runtime_services_dynamic_ports,

tests/serve/test_sglang.py

@@ -73,7 +73,11 @@ sglang_configs = {
         name="disaggregated",
         directory=sglang_dir,
         script_name="disagg.sh",
-        marks=[pytest.mark.gpu_2, pytest.mark.post_merge],
+        marks=[
+            pytest.mark.gpu_2,
+            pytest.mark.post_merge,
+            pytest.mark.skip(reason="DYN-2265"),
+        ],
         model="Qwen/Qwen3-0.6B",
         env={},
         frontend_port=DefaultPort.FRONTEND.value,
@@ -120,7 +124,11 @@ sglang_configs = {
         name="kv_events",
         directory=sglang_dir,
         script_name="agg_router.sh",
-        marks=[pytest.mark.gpu_2, pytest.mark.post_merge],
+        marks=[
+            pytest.mark.gpu_2,
+            pytest.mark.post_merge,
+            pytest.mark.skip(reason="DYN-2265"),
+        ],
         model="Qwen/Qwen3-0.6B",
         env={
             "DYN_LOG": "dynamo_llm::kv_router::publisher=trace,dynamo_kv_router::scheduling::selector=info",
@@ -298,6 +306,7 @@ sglang_configs = {
             pytest.mark.gpu_1,
             pytest.mark.post_merge,
             pytest.mark.timeout(240),
+            pytest.mark.skip(reason="DYN-2261"),
         ],
         model="Qwen/Qwen3-0.6B",
         env={"DYN_ENABLE_ANTHROPIC_API": "1"},

tests/serve/test_trtllm.py

@@ -119,7 +119,14 @@ trtllm_configs = {
         name="disaggregated_logprobs",
         directory=trtllm_dir,
         script_name="disagg.sh",
-        marks=[pytest.mark.gpu_2, pytest.mark.post_merge, pytest.mark.trtllm],
+        marks=[
+            pytest.mark.gpu_2,
+            pytest.mark.post_merge,
+            pytest.mark.trtllm,
+            pytest.mark.skip(
+                reason="DYN-2265 https://github.com/ai-dynamo/dynamo/pull/6704/changes#r2866554157 TODO enable this test when upgrading from trtllm 1.3.0rc5 to 1.3.0rc5.post1"
+            ),
+        ],
         model="Qwen/Qwen3-0.6B",
         frontend_port=DefaultPort.FRONTEND.value,
         request_payloads=[
@@ -238,6 +245,7 @@ trtllm_configs = {
             pytest.mark.gpu_1,
             pytest.mark.trtllm,
             pytest.mark.post_merge,
+            pytest.mark.skip(reason="DYN-2262"),
             pytest.mark.timeout(
                 480
             ),  # 3x measured time (83.85s) + download time (210s) for 7B model

tests/serve/test_vllm.py

@@ -192,7 +192,11 @@ vllm_configs = {
         name="agg-router",
         directory=vllm_dir,
         script_name="agg_router.sh",
-        marks=[pytest.mark.gpu_2, pytest.mark.post_merge],
+        marks=[
+            pytest.mark.gpu_2,
+            pytest.mark.post_merge,
+            pytest.mark.skip(reason="DYN-2263"),
+        ],
         model="Qwen/Qwen3-0.6B",
         request_payloads=[
             chat_payload_default(
@@ -211,7 +215,11 @@ vllm_configs = {
         name="agg-router-approx",
         directory=vllm_dir,
         script_name="agg_router_approx.sh",
-        marks=[pytest.mark.gpu_2, pytest.mark.post_merge],
+        marks=[
+            pytest.mark.gpu_2,
+            pytest.mark.post_merge,
+            pytest.mark.skip(reason="DYN-2264"),
+        ],
         model="Qwen/Qwen3-0.6B",
         request_payloads=[
             # Test approximate KV routing (--no-kv-events mode)
@@ -341,7 +349,11 @@ vllm_configs = {
         name="multimodal_disagg_qwen3vl_2b_epd",
         directory=vllm_dir,
         script_name="disagg_multimodal_epd.sh",
-        marks=[pytest.mark.gpu_1, pytest.mark.pre_merge],
+        marks=[
+            pytest.mark.gpu_1,
+            pytest.mark.pre_merge,
+            pytest.mark.skip(reason="DYN-2265"),
+        ],
         model="Qwen/Qwen3-VL-2B-Instruct",
         script_args=["--model", "Qwen/Qwen3-VL-2B-Instruct", "--single-gpu"],
         timeout=360,
@@ -834,6 +846,7 @@ def lora_chat_payload(
 @pytest.mark.model("Qwen/Qwen3-0.6B")
 @pytest.mark.timeout(600)
 @pytest.mark.post_merge
+@pytest.mark.skip(reason="DYN-2260")
 def test_lora_aggregated(
     request,
     runtime_services_dynamic_ports,
@@ -890,6 +903,7 @@ def test_lora_aggregated(
 @pytest.mark.timeout(600)
 @pytest.mark.post_merge
 @pytest.mark.parametrize("num_system_ports", [2], indirect=True)
+@pytest.mark.skip(reason="DYN-2265")
 def test_lora_aggregated_router(
     request,
     runtime_services_dynamic_ports,