From 50993554d8ec62111ae3dc32bc075801a1d726dc Mon Sep 17 00:00:00 2001
From: ybyang <10629930+whybeyoung@users.noreply.github.com>
Date: Sun, 14 Jun 2026 15:01:52 +0800
Subject: [PATCH] fix(health): make health-check rid unique across tokenizer
 workers (#28143)

---
 python/sglang/srt/disaggregation/encode_server.py | 7 +++++--
 python/sglang/srt/entrypoints/http_server.py      | 5 ++++-
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/python/sglang/srt/disaggregation/encode_server.py b/python/sglang/srt/disaggregation/encode_server.py
index 6ea60c397..89c181c12 100644
--- a/python/sglang/srt/disaggregation/encode_server.py
+++ b/python/sglang/srt/disaggregation/encode_server.py
@@ -11,6 +11,7 @@ import pickle
 import threading
 import time
 import traceback
+import uuid
 from collections import defaultdict
 from http import HTTPStatus
 from typing import Dict, List, Optional, Set, Tuple, Union
@@ -2550,7 +2551,8 @@ async def _dp_worker_health_encode(enc: MMEncoder) -> None:
         # No processor → can't functionally probe; liveness alone is healthy.
         return None
 
-    req_id = f"{HEALTH_CHECK_RID_PREFIX}_{time.time()}"
+    # uuid keeps rids unique across workers; a bare time.time() can collide.
+    req_id = f"{HEALTH_CHECK_RID_PREFIX}_{uuid.uuid4().hex}"
     try:
         _, _, _, error_msg, error_code = await enc.encode(
             mm_items=mm_items,
@@ -3757,7 +3759,8 @@ async def health_generate():
         return Response(status_code=200)
 
     try:
-        req_id = f"{HEALTH_CHECK_RID_PREFIX}_{time.time()}"
+        # uuid keeps rids unique across workers; a bare time.time() can collide.
+        req_id = f"{HEALTH_CHECK_RID_PREFIX}_{uuid.uuid4().hex}"
 
         dummy_request = {
             "mm_items": mm_items,
diff --git a/python/sglang/srt/entrypoints/http_server.py b/python/sglang/srt/entrypoints/http_server.py
index 86efbd085..1108ea4ff 100644
--- a/python/sglang/srt/entrypoints/http_server.py
+++ b/python/sglang/srt/entrypoints/http_server.py
@@ -24,6 +24,7 @@ import os
 import tempfile
 import threading
 import time
+import uuid
 from contextlib import asynccontextmanager
 from http import HTTPStatus
 from typing import (
@@ -607,7 +608,9 @@ async def health_generate(request: Request) -> Response:
         return Response(status_code=200)
 
     sampling_params = {"max_new_tokens": 1, "temperature": 0.0}
-    rid = f"{HEALTH_CHECK_RID_PREFIX}_{time.time()}"
+    # uuid keeps rids unique across tokenizer workers (a bare time.time() can
+    # collide and crash the shared DetokenizerManager decode_status).
+    rid = f"{HEALTH_CHECK_RID_PREFIX}_{uuid.uuid4().hex}"
 
     if _global_state.tokenizer_manager.is_generation:
         gri = GenerateReqInput(