From b53a39c5e4e3cbe59ad0d3ef68fae5d4aa2efef2 Mon Sep 17 00:00:00 2001
From: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>
Date: Fri, 7 Aug 2026 13:41:19 -0700
Subject: [PATCH] Limit prefill delayer debug logs to rank zero (#34020)

Co-authored-by: Lianmin Zheng <lianminzheng@gmail.com>
---
 python/sglang/srt/managers/prefill_delayer.py | 7 ++++++-
 python/sglang/srt/managers/scheduler.py       | 1 +
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/python/sglang/srt/managers/prefill_delayer.py b/python/sglang/srt/managers/prefill_delayer.py
index eacae42a7..650048791 100644
--- a/python/sglang/srt/managers/prefill_delayer.py
+++ b/python/sglang/srt/managers/prefill_delayer.py
@@ -52,9 +52,11 @@ class PrefillDelayer:
         metrics_collector: Optional["SchedulerMetricsCollector"] = None,
         device: Optional["torch.device"] = "cpu",
         device_group=None,
+        debug_log_enabled: bool = True,
     ):
         self._max_delay_passes = max_delay_passes
         self._token_usage_low_watermark = token_usage_low_watermark
+        self._debug_log_enabled = _DEBUG_LOG and debug_log_enabled
         # Queue-based trigger is opt-in: activates only when queue_min_ratio
         # is explicitly set. Additive with the slot-based trigger.
         self._queue_min_ratio = server_args.prefill_delayer_queue_min_ratio
@@ -358,6 +360,7 @@ class PrefillDelayerSinglePassExecutor:
             actual_execution=actual_prefill,
             output=self._result,
             metrics_collector=self._prefill_delayer._metrics_collector,
+            debug_log_enabled=self._prefill_delayer._debug_log_enabled,
         )
 
     def negotiate_should_allow_prefill(
@@ -384,8 +387,10 @@ def _record_single_pass_result(
     actual_execution: bool,
     output: _NegotiateOutput,
     metrics_collector: Optional["SchedulerMetricsCollector"],
+    *,
+    debug_log_enabled: bool,
 ) -> None:
-    if _DEBUG_LOG:
+    if debug_log_enabled:
         if output.output_allow and (output.output_reason == "wait_timeout"):
             logger.info(
                 f"PrefillDelayer timeout thus not forbid prefill "
diff --git a/python/sglang/srt/managers/scheduler.py b/python/sglang/srt/managers/scheduler.py
index 4ec1e169e..07aaf4f3e 100644
--- a/python/sglang/srt/managers/scheduler.py
+++ b/python/sglang/srt/managers/scheduler.py
@@ -1211,6 +1211,7 @@ class Scheduler(
                     max_delay_passes=get_schedule().prefill_delayer_max_delay_passes,
                     token_usage_low_watermark=get_schedule().prefill_delayer_token_usage_low_watermark,
                     device=self.tp_group.device,
+                    debug_log_enabled=self.ps.attn_tp_rank == 0,
                 )
 
         # NOTE: preemption is enabled by default for priority scheduling.