deprecate prefill-round-robin-balance (#16195)

Signed-off-by: Chang Huaixin (OpenAnolis) <changhuaixin@linux.alibaba.com>
Co-authored-by: Liangsheng Yin <lsyincs@gmail.com>

docs/advanced_features/server_arguments.md

@@ -207,7 +207,6 @@ Please consult the documentation below and [server_args.py](https://github.com/s
 | `--data-parallel-size`<br>`--dp-size` | The data parallelism size. | `1` | Type: int |
 | `--load-balance-method` | The load balancing strategy for data parallelism. The Minimum Token algorithm can only be used when DP attention is applied. This algorithm performs load balancing based on the real-time token load of the DP workers. | `auto` | `auto`, `round_robin`, `follow_bootstrap_room`, `shortest_queue`, `minimum_tokens` |
 | `--load-watch-interval` | The interval of load watching in seconds. | `0.1` | Type: float |
-| `--prefill-round-robin-balance` | Prefill is round robin balanced. This is used to promise decode server can get the correct dp rank. | `False` | bool flag (set to enable) |
 
 ## Multi-node distributed serving
 | Argument | Description | Defaults | Options |
@@ -461,6 +460,7 @@ Please consult the documentation below and [server_args.py](https://github.com/s
 | --- | --- | --- | --- |
 | `--enable-ep-moe` | NOTE: --enable-ep-moe is deprecated. Please set `--ep-size` to the same value as `--tp-size` instead. | `None` | N/A |
 | `--enable-deepep-moe` | NOTE: --enable-deepep-moe is deprecated. Please set `--moe-a2a-backend` to 'deepep' instead. | `None` | N/A |
+| `--prefill-round-robin-balance` | Note: Note: --prefill-round-robin-balance is deprecated now. | `None` | N/A |
 | `--enable-flashinfer-cutlass-moe` | NOTE: --enable-flashinfer-cutlass-moe is deprecated. Please set `--moe-runner-backend` to 'flashinfer_cutlass' instead. | `None` | N/A |
 | `--enable-flashinfer-cutedsl-moe` | NOTE: --enable-flashinfer-cutedsl-moe is deprecated. Please set `--moe-runner-backend` to 'flashinfer_cutedsl' instead. | `None` | N/A |
 | `--enable-flashinfer-trtllm-moe` | NOTE: --enable-flashinfer-trtllm-moe is deprecated. Please set `--moe-runner-backend` to 'flashinfer_trtllm' instead. | `None` | N/A |

docs/platforms/ascend_npu_deepseek_example.md

@@ -157,7 +157,6 @@ python -m sglang.launch_server \
     --speculative-num-steps 3 \
     --speculative-eagle-topk 1 \
     --speculative-num-draft-tokens 4 \
-    --prefill-round-robin-balance \
     --disable-shared-experts-fusion \
     --dtype bfloat16 \
     --tokenizer-worker-num 4
@@ -293,7 +292,6 @@ do
       --speculative-eagle-topk 1 \
       --speculative-num-draft-tokens 3  \
       --tokenizer-worker-num 4 \
-      --prefill-round-robin-balance \
       --disable-shared-experts-fusion \
       --dtype bfloat16
 done

docs/platforms/ascend_npu_support_features.md

@@ -132,7 +132,6 @@ If you want to know the meaning and usage of each parameter, click [Service Argu
 |----------------------------------------|---------------|-------------------------------------------------------------|:----------------------------------------:|:----------------------------------------:|
 | `--data-parallel-size`<br/>`--dp-size` | `1`           | Type: int                                                   | **<span style="color: green;">√</span>** | **<span style="color: green;">√</span>** |
 | `--load-balance-method`                | `round_robin` | `round_robin`,<br/> `shortest_queue`,<br/> `minimum_tokens` | **<span style="color: green;">√</span>** | **<span style="color: green;">√</span>** |
-| `--prefill-round-robin-balance`        | `False`       | bool flag<br/> (set to enable)                              | **<span style="color: green;">√</span>** | **<span style="color: green;">√</span>** |
 
 ## Multi-node distributed serving
 

docs/references/multi_node_deployment/rbg_pd/deepseekv32_pd.md

@@ -260,7 +260,6 @@ spec:
                   -  mlx5_0,mlx5_1,mlx5_2,mlx5_3,mlx5_4,mlx5_5,mlx5_6,mlx5_7
                   - --chunked-prefill-size
                   - "131072"
-                  - --prefill-round-robin-balance
                   - --eplb-rebalance-layers-per-chunk
                   - "29"
                   - --page-size
@@ -332,7 +331,6 @@ spec:
                 -  /log
                 - --chunked-prefill-size
                 - "262144"
-                - --prefill-round-robin-balance
                 - --eplb-rebalance-layers-per-chunk
                 - "29"
                 - --page-size

python/sglang/srt/server_args.py

@@ -377,8 +377,6 @@ class ServerArgs:
     # Data parallelism
     dp_size: int = 1
     load_balance_method: str = "auto"
-    # FIXME: remove this after dp rank scheduling is fully supported with PD-Disaggregation
-    prefill_round_robin_balance: bool = False
 
     # Multi-node distributed serving
     dist_init_addr: Optional[str] = None
@@ -2221,13 +2219,6 @@ class ServerArgs:
             self.disable_radix_cache = True
             logger.warning("KV cache is forced as chunk cache for decode server")
 
-            if self.dp_size > 1 and not is_in_ci():
-                assert self.prefill_round_robin_balance, (
-                    "Prefill round robin balance is required when dp size > 1. "
-                    "Please make sure that the prefill instance is launched with `--load-balance-method auto` "
-                    "or `--load-balance-method follow_bootstrap_room` "
-                    "and `--prefill-round-robin-balance` is set for decode server."
-                )
         elif self.disaggregation_mode == "prefill":
             if self.disaggregation_decode_tp is None:
                 self.disaggregation_decode_tp = self.tp_size
@@ -3224,9 +3215,8 @@ class ServerArgs:
         )
         parser.add_argument(
             "--prefill-round-robin-balance",
-            default=ServerArgs.prefill_round_robin_balance,
-            action="store_true",
-            help="Prefill is round robin balanced. This is used to promise decode server can get the correct dp rank.",
+            action=DeprecatedAction,
+            help="Note: --prefill-round-robin-balance is deprecated now.",
         )
 
         # Multi-node distributed serving
@@ -5173,6 +5163,10 @@ class LoRAPathAction(argparse.Action):
         setattr(namespace, self.dest, lora_paths)
 
 
+def print_deprecated_warning(message: str):
+    logger.warning(f"\033[1;33m{message}\033[0m")
+
+
 class DeprecatedAction(argparse.Action):
     def __init__(self, option_strings, dest, nargs=0, **kwargs):
         super(DeprecatedAction, self).__init__(
@@ -5180,11 +5174,9 @@ class DeprecatedAction(argparse.Action):
         )
 
     def __call__(self, parser, namespace, values, option_string=None):
-        raise ValueError(self.help)
-
-
-def print_deprecated_warning(message: str):
-    logger.warning(f"\033[33m{message}\033[0m")
+        print_deprecated_warning(
+            f"The command line argument '{option_string}' is deprecated and will be removed in future versions."
+        )
 
 
 def auto_choose_speculative_params(self: ServerArgs):

test/srt/test_disaggregation_dp_attention.py

@@ -69,7 +69,6 @@ class TestDisaggregationDPAttention(PDDisaggregationServerBase):
             "--enable-dp-attention",
             "--base-gpu-id",
             str(cls.PREFILL_DP_SIZE),
-            "--prefill-round-robin-balance",
         ]
         decode_args += cls.transfer_backend + cls.rdma_devices
         cls.process_decode = popen_launch_pd_server(