Tiny add command line args for prefill delayer and unify names (#16830)

python/sglang/srt/environ.py

@@ -489,6 +489,18 @@ _warn_deprecated_env_to_cli_flag(
     "SGLANG_SUPPORT_CUTLASS_BLOCK_FP8",
     "It will be completely removed in 0.5.7. Please use '--fp8-gemm-backend=cutlass' instead.",
 )
+_warn_deprecated_env_to_cli_flag(
+    "SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE",
+    "Please use '--enable-prefill-delayer' instead.",
+)
+_warn_deprecated_env_to_cli_flag(
+    "SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES",
+    "Please use '--prefill-delayer-max-delay-passes' instead.",
+)
+_warn_deprecated_env_to_cli_flag(
+    "SGLANG_PREFILL_DELAYER_TOKEN_USAGE_LOW_WATERMARK",
+    "Please use '--prefill-delayer-token-usage-low-watermark' instead.",
+)
 
 
 def example_with_exit_stack():

python/sglang/srt/managers/scheduler.py

@@ -765,7 +765,7 @@ class Scheduler(
             self.schedule_low_priority_values_first,
         )
         self.prefill_delayer: Optional[PrefillDelayer] = None
-        if envs.SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE.get():
+        if self.server_args.enable_prefill_delayer:
             self.prefill_delayer = PrefillDelayer(
                 dp_size=self.dp_size,
                 attn_tp_size=self.attn_tp_size,
@@ -774,10 +774,8 @@ class Scheduler(
                 metrics_collector=(
                     self.metrics_collector if self.enable_metrics else None
                 ),
-                max_delay_passes=envs.SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES.get(),
-                token_usage_low_watermark=(
-                    envs.SGLANG_PREFILL_DELAYER_TOKEN_USAGE_LOW_WATERMARK.get()
-                ),
+                max_delay_passes=self.server_args.prefill_delayer_max_delay_passes,
+                token_usage_low_watermark=self.server_args.prefill_delayer_token_usage_low_watermark,
             )
         # Enable preemption for priority scheduling.
         self.try_preemption = self.enable_priority_scheduling

python/sglang/srt/managers/scheduler_metrics_mixin.py

@@ -95,7 +95,9 @@ class SchedulerMetricsMixin:
             if dp_rank is not None:
                 labels["dp_rank"] = dp_rank
             self.metrics_collector = SchedulerMetricsCollector(
-                labels=labels, enable_lora=self.enable_lora
+                labels=labels,
+                enable_lora=self.enable_lora,
+                prefill_delayer_max_delay_passes=self.server_args.prefill_delayer_max_delay_passes,
             )
 
             if ENABLE_METRICS_DEVICE_TIMER:

python/sglang/srt/metrics/collector.py

@@ -271,6 +271,7 @@ class SchedulerMetricsCollector:
         self,
         labels: Dict[str, str],
         enable_lora: bool = False,
+        prefill_delayer_max_delay_passes: int = 30,
     ) -> None:
         # We need to import prometheus_client after setting the env variable `PROMETHEUS_MULTIPROC_DIR`
         from prometheus_client import Counter, Gauge, Histogram, Summary
@@ -761,13 +762,12 @@ class SchedulerMetricsCollector:
             labelnames=list(labels.keys()) + ["category", "num_prefill_ranks"],
         )
 
-        max_delay_passes = envs.SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES.get()
         self.prefill_delayer_wait_forward_passes = Histogram(
             name="sglang:prefill_delayer_wait_forward_passes",
             documentation="Histogram of forward passes waited by prefill delayer.",
             labelnames=labels.keys(),
             # Need bucket "<=0" for zero-delay cases
-            buckets=[0, 5, 20, max_delay_passes - 1],
+            buckets=[0, 5, 20, prefill_delayer_max_delay_passes - 1],
         )
         self.prefill_delayer_wait_seconds = Histogram(
             name="sglang:prefill_delayer_wait_seconds",

python/sglang/srt/server_args.py

@@ -311,6 +311,9 @@ class ServerArgs:
     swa_full_tokens_ratio: float = 0.8
     disable_hybrid_swa_memory: bool = False
     radix_eviction_policy: str = "lru"
+    enable_prefill_delayer: bool = False
+    prefill_delayer_max_delay_passes: int = 30
+    prefill_delayer_token_usage_low_watermark: Optional[float] = None
 
     # Runtime options
     device: Optional[str] = None
@@ -665,6 +668,9 @@ class ServerArgs:
         # Handle deprecated arguments.
         self._handle_deprecated_args()
 
+        # Handle deprecated environment variables for prefill delayer.
+        self._handle_prefill_delayer_env_compat()
+
         # Set missing default values.
         self._handle_missing_default_values()
 
@@ -778,6 +784,14 @@ class ServerArgs:
             )
             self.tool_call_parser = deprecated_tool_call_parsers[self.tool_call_parser]
 
+    def _handle_prefill_delayer_env_compat(self):
+        if envs.SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE.get():
+            self.enable_prefill_delayer = True
+        if x := envs.SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES.get():
+            self.prefill_delayer_max_delay_passes = x
+        if x := envs.SGLANG_PREFILL_DELAYER_TOKEN_USAGE_LOW_WATERMARK.get():
+            self.prefill_delayer_token_usage_low_watermark = x
+
     def _handle_missing_default_values(self):
         if self.tokenizer_path is None:
             self.tokenizer_path = self.model_path
@@ -2884,6 +2898,23 @@ class ServerArgs:
             default=ServerArgs.radix_eviction_policy,
             help="The eviction policy of radix trees. 'lru' stands for Least Recently Used, 'lfu' stands for Least Frequently Used.",
         )
+        parser.add_argument(
+            "--enable-prefill-delayer",
+            action="store_true",
+            help="Enable prefill delayer for DP attention to reduce idle time.",
+        )
+        parser.add_argument(
+            "--prefill-delayer-max-delay-passes",
+            type=int,
+            default=ServerArgs.prefill_delayer_max_delay_passes,
+            help="Maximum forward passes to delay prefill.",
+        )
+        parser.add_argument(
+            "--prefill-delayer-token-usage-low-watermark",
+            type=float,
+            default=None,
+            help="Token usage low watermark for prefill delayer.",
+        )
 
         # Runtime options
         parser.add_argument(