test: fix KV Router and KVBM tests in CI (#3848)

tests/kvbm/test_determinism_disagg.py

@@ -223,8 +223,10 @@ class LLMServerManager:
         )
         print(f"Decoder process started with PID: {self.process_decoder.pid}")
 
-        # Give decoder time to start up
-        time.sleep(5)
+        # The prefiller and decoder cannot download the model simultaneously,
+        # because the Hugging Face rust library (invoked by fetch_llm) needs to hold an exclusive lock on the model files.
+        print("Sleeping for 60 seconds to allow the decoder to download the model. ")
+        time.sleep(60)
 
         # Launch prefiller
         self.process_prefiller = subprocess.Popen(

tests/serve/test_sglang.py

@@ -71,7 +71,7 @@ sglang_configs = {
         request_payloads=[
             chat_payload_default(
                 expected_log=[
-                    r"ZMQ listener .* received batch with \d+ events \(seq=\d+\)",
+                    r"ZMQ listener .* received batch with \d+ events \(seq=\d+(?:, [^)]*)?\)",
                     r"Event processor for worker_id \d+ processing event: Stored\(",
                     r"Selected worker: worker_id=\d+ dp_rank=.*?, logit: ",
                 ]

tests/serve/test_vllm.py

@@ -51,7 +51,7 @@ vllm_configs = {
         request_payloads=[
             chat_payload_default(
                 expected_log=[
-                    r"ZMQ listener .* received batch with \d+ events \(seq=\d+\)",
+                    r"ZMQ listener .* received batch with \d+ events \(seq=\d+(?:, [^)]*)?\)",
                     r"Event processor for worker_id \d+ processing event: Stored\(",
                     r"Selected worker: worker_id=\d+ dp_rank=.*?, logit: ",
                 ]