[Nightly] Optimize nightly CI (#4509)

### What this PR does / why we need it?
1. Optimize multi-node waiting logic
2. Remove the `tee` pipeline for logs, which will lead to hang issue

### How was this patch tested?


- vLLM version: v0.12.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.12.0

Signed-off-by: wangli <wangli858794774@gmail.com>

.github/workflows/_e2e_nightly_multi_node.yaml

@@ -84,8 +84,6 @@ jobs:
            pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
            pip install jinja2-cli
 
-           #apt-get update -y && apt-get install -y git curl
-
         - name: Install kubectl
           run: |
             # Install kubectl
@@ -112,6 +110,8 @@ jobs:
           run: |
             # prepare for lws entrypoint scripts
             install -D tests/e2e/nightly/multi_node/scripts/run.sh /root/.cache/tests/run.sh
+            # clear log directory
+            rm -fr $RESULT_FILE
 
         - name: Clear resources
           run: |
@@ -263,5 +263,5 @@ jobs:
         - name: Post process
           if: always()
           run: |
-            kubectl get pods -n $NAMESPACE
+            kubectl get pods -n $NAMESPACE --ignore-not-found=true
-            kubectl delete -f ./lws.yaml
+            kubectl delete -f ./lws.yaml --ignore-not-found=true || true

.github/workflows/vllm_ascend_test_nightly_a3.yaml

@@ -61,9 +61,6 @@ jobs:
           - name: multi-node-qwenw8a8-2node
             config_file_path: Qwen3-235B-W8A8.yaml
             size: 2
-          - name: multi-node-glm-2node
-            config_file_path: GLM-4_5.yaml
-            size: 2
           - name: multi-node-dpsk3.2-exp-2node
             config_file_path: DeepSeek-V3_2-Exp-bf16.yaml
             size: 2
@@ -134,9 +131,6 @@ jobs:
           - name: deepseek3_2-exp-w8a8
             os: linux-aarch64-a3-16
             tests: tests/e2e/nightly/models/test_deepseek_v3_2_exp_w8a8.py
-          - name: glm-4-5
-            os: linux-aarch64-a3-16
-            tests: tests/e2e/nightly/models/test_glm4_5.py
     uses: ./.github/workflows/_e2e_nightly_single_node.yaml
     with:
       vllm: v0.12.0

examples/disaggregated_prefill_v1/gen_ranktable.sh

@@ -78,6 +78,7 @@ if [ -n "$LOCAL_DEVICE_IDS" ]; then
 fi
 
 if [[ -n "${GEN_RANKTABLE}" || ! -e ${PWD}/ranktable.json ]]; then
+    timeout 180s \
     GLOO_SOCKET_IFNAME=$NETWORK_CARD_NAME torchrun \
         --nproc_per_node 1 \
         --nnodes ${NNODES} \

tests/e2e/conftest.py

@@ -20,6 +20,7 @@
 import contextlib
 import gc
 import json
+import logging
 import os
 import shlex
 import subprocess
@@ -35,6 +36,7 @@ import requests
 import torch
 from modelscope import snapshot_download  # type: ignore[import-untyped]
 from PIL import Image
+from requests.exceptions import RequestException
 from torch import nn
 from transformers import (AutoConfig, AutoModelForCausalLM, AutoTokenizer,
                           BatchEncoding, BatchFeature)
@@ -70,6 +72,7 @@ _PromptMultiModalInput = Union[List[_M], List[List[_M]]]
 PromptImageInput = _PromptMultiModalInput[Image.Image]
 PromptAudioInput = _PromptMultiModalInput[Tuple[np.ndarray, int]]
 PromptVideoInput = _PromptMultiModalInput[np.ndarray]
+logger = logging.getLogger(__name__)
 
 _TEST_DIR = os.path.dirname(__file__)
 
@@ -161,22 +164,17 @@ class RemoteOpenAIServer:
         max_wait_seconds = max_wait_seconds or 1800
         if self.disaggregated_prefill:
             assert proxy_port is not None, "for disaggregated_prefill, proxy port must be provided"
-            self._wait_for_server_pd(proxy_port=proxy_port,
+            self._wait_for_server_pd(timeout=max_wait_seconds)
-                                     timeout=max_wait_seconds)
         else:
-            self._wait_for_server(url=self.url_for("health"),
+            self._wait_for_multiple_servers(
+                [(self.host, self.url_for("health"))],
                 timeout=max_wait_seconds)
 
     def __enter__(self):
         return self
 
     def __exit__(self, exc_type, exc_value, traceback):
-        self.proc.terminate()
+        self._terminate_server()
-        try:
-            self.proc.wait(8)
-        except subprocess.TimeoutExpired:
-            # force kill if needed
-            self.proc.kill()
 
     def _poll(self) -> Optional[int]:
         """Subclasses override this method to customize process polling"""
@@ -201,48 +199,95 @@ class RemoteOpenAIServer:
         finally:
             if isinstance(client, httpx.Client):
                 client.close()
+            self._terminate_server()
 
-    def _wait_for_server_pd(self, proxy_port: int, timeout: float):
+    def _wait_for_server_pd(self, timeout: float):
         # Wait for all api_server nodes ready
         assert self.nodes_info is not None, "cluster info must be provided"
-        for node_info in self.nodes_info:
+        proxy_port = self.proxy_port
-            if node_info.headless:
-                continue
 
-            url_health = f"http://{node_info.ip}:{node_info.server_port}/health"
+        def url_health(ip: str, port: int) -> str:
-            self._wait_for_server(url=url_health, timeout=timeout)
+            return f"http://{ip}:{port}/health"
+
+        targets = [(node_info.ip,
+                    url_health(node_info.ip, node_info.server_port))
+                   for node_info in self.nodes_info if not node_info.headless]
 
         # Wait for proxy ready
         master_node = self.nodes_info[0]
         url_proxy = f"http://{master_node.ip}:{proxy_port}/healthcheck"
-        self._wait_for_server(url=url_proxy, timeout=timeout)
 
-    def _wait_for_server(self, *, url: str, timeout: float):
+        # Wait for master node proxy first
-        # run health check
+        self._wait_for_multiple_servers([(master_node.ip, url_proxy)],
+                                        timeout=timeout)
+
+        # Then wait for all api_server nodes
+        self._wait_for_multiple_servers(targets=targets, timeout=timeout)
+
+    def _wait_for_multiple_servers(self, targets, timeout: float):
+        """
+        targets: List[(node_ip, url)]
+        """
         start = time.time()
         client = requests
+
+        # track readiness
+        ready = {node_ip: False for node_ip, _ in targets}
+
+        # polling loop
         while True:
+            all_ready = True
+
+            for node_ip, url in targets:
+                if ready[node_ip]:
+                    continue  # already ready
+
                 try:
-                if client.get(url).status_code == 200:
+                    resp = client.get(url)
-                    break
+                    if resp.status_code == 200:
-            except Exception:
+                        ready[node_ip] = True
-                # this exception can only be raised by requests.get,
+                        logger.info(f"[READY] Node {node_ip} is ready.")
-                # which means the server is not ready yet.
+                    else:
-                # the stack trace is not useful, so we suppress it
+                        all_ready = False
-                # by using `raise from None`.
+                        logger.info(f"[WAIT] {url}: HTTP {resp.status_code}")
+                except RequestException:
+                    all_ready = False
+                    logger.info(f"[WAIT] {url}: connection failed")
+
+                    # underlying process died?
                     result = self._poll()
                     if result is not None and result != 0:
-                    raise RuntimeError("Server exited unexpectedly.") from None
+                        raise RuntimeError(
+                            f"Server at {node_ip} exited unexpectedly."
+                        ) from None
+
+            # if all nodes ready, exit
+            if all_ready:
+                break
+
+            # check timeout
+            if time.time() - start > timeout:
+                not_ready_nodes = [n for n, ok in ready.items() if not ok]
+                self._terminate_server()
+                raise RuntimeError(
+                    f"Timeout: these nodes did not become ready: {not_ready_nodes}"
+                ) from None
 
             time.sleep(5)
-                if time.time() - start > timeout:
-                    raise RuntimeError(
-                        "Server failed to start in time.") from None
 
     @property
     def url_root(self) -> str:
         return f"http://{self.host}:{self.port}"
 
+    def _terminate_server(self) -> None:
+        """Subclasses override this method to customize server process termination"""
+        self.proc.terminate()
+        try:
+            self.proc.wait(8)
+        except subprocess.TimeoutExpired:
+            # force kill if needed
+            self.proc.kill()
+
     def url_for(self, *parts: str) -> str:
         return self.url_root + "/" + "/".join(parts)
 

tests/e2e/nightly/multi_node/config/models/GLM-4_5.yaml

@@ -1,50 +0,0 @@
-test_name: "test GLM-4.5 multi-dp"
-model: "ZhipuAI/GLM-4.5"
-num_nodes: 2
-npu_per_node: 16
-env_common:
-  VLLM_USE_MODELSCOPE: true
-  OMP_PROC_BIND: false
-  OMP_NUM_THREADS: 100
-  HCCL_BUFFSIZE: 1024
-  SERVER_PORT: 8080
-  NUMEXPR_MAX_THREADS: 128
-
-deployment:
-  -
-    server_cmd: >
-        vllm serve "ZhipuAI/GLM-4.5"
-        --host 0.0.0.0
-        --port $SERVER_PORT
-        --data-parallel-size 4
-        --data-parallel-size-local 2
-        --data-parallel-address $LOCAL_IP
-        --data-parallel-rpc-port 13389
-        --tensor-parallel-size 8
-        --seed 1024
-        --enable-expert-parallel
-        --max-num-seqs 16
-        --max-model-len 8192
-        --max-num-batched-tokens 8192
-        --trust-remote-code
-        --no-enable-prefix-caching
-        --gpu-memory-utilization 0.9
-  -
-    server_cmd: >
-        vllm serve "ZhipuAI/GLM-4.5"
-        --headless
-        --data-parallel-size 4
-        --data-parallel-size-local 2
-        --data-parallel-start-rank 2
-        --data-parallel-address $MASTER_IP
-        --data-parallel-rpc-port 13389
-        --tensor-parallel-size 8
-        --seed 1024
-        --max-num-seqs 16
-        --max-model-len 8192
-        --max-num-batched-tokens 8192
-        --enable-expert-parallel
-        --trust-remote-code
-        --no-enable-prefix-caching
-        --gpu-memory-utilization 0.9
-benchmarks:

tests/e2e/nightly/multi_node/config/multi_node_config.py

@@ -31,8 +31,8 @@ class NodeInfo:
         return (f"NodeInfo:\n"
                 f"  index={self.index}\n"
                 f"  ip={self.ip}\n"
-                f"  server_port={self.server_port}\n"
+                f"  headless={self.headless}\n"
-                f"  headless={self.headless}")
+                f"  server_port={self.server_port}")
 
 
 class MultiNodeConfig:

tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2

@@ -87,7 +87,7 @@ spec:
               - name: VLLM_ASCEND_REMOTE_URL
                 value: {{ vllm_ascend_remote_url | default("https://github.com/vllm-project/vllm-ascend.git") }}
               - name: RESULT_FILE_PATH
-                value: {{ result_file_path | default("/root/.cache/tests/ret/test_result.txt") }}
+                value: {{ result_file_path | default("/root/.cache/tests/ret") }}
               - name: FAIL_TAG
                 value: {{ fail_tag | default("FAIL_TAG") }}
             command:

tests/e2e/nightly/multi_node/scripts/run.sh

@@ -127,19 +127,14 @@ kill_npu_processes() {
 run_tests_with_log() {
     set +e
     kill_npu_processes
-    BASENAME=$(basename "$CONFIG_YAML_PATH" .yaml)
+    pytest -sv tests/e2e/nightly/multi_node/test_multi_node.py
-    # each worker should have log file
+    ret=$?
-    LOG_FILE="${RESULT_FILE_PATH}/${BASENAME}_worker_${LWS_WORKER_INDEX}.log"
-    mkdir -p ${RESULT_FILE_PATH}
-    pytest -sv tests/e2e/nightly/multi_node/test_multi_node.py 2>&1 | tee $LOG_FILE
-    ret=${PIPESTATUS[0]}
     set -e
     if [ "$LWS_WORKER_INDEX" -eq 0 ]; then
         if [ $ret -eq 0 ]; then
             print_success "All tests passed!"
         else
             print_failure "Some tests failed!"
-            mv LOG_FILE error_${LOG_FILE}
         fi
     fi
 }