Add manual routing policy for router (#15586)

sgl-model-gateway/bindings/python/sglang_router/router.py

@@ -21,6 +21,7 @@ def policy_from_str(policy_str: Optional[str]) -> PolicyType:
         "cache_aware": PolicyType.CacheAware,
         "power_of_two": PolicyType.PowerOfTwo,
         "bucket": PolicyType.Bucket,
+        "manual": PolicyType.Manual,
     }
     return policy_map[policy_str]
 

sgl-model-gateway/bindings/python/sglang_router/router_args.py

@@ -171,21 +171,28 @@ class RouterArgs:
             f"--{prefix}policy",
             type=str,
             default=RouterArgs.policy,
-            choices=["random", "round_robin", "cache_aware", "power_of_two"],
+            choices=["random", "round_robin", "cache_aware", "power_of_two", "manual"],
             help="Load balancing policy to use. In PD mode, this is used for both prefill and decode unless overridden",
         )
         parser.add_argument(
             f"--{prefix}prefill-policy",
             type=str,
             default=None,
-            choices=["random", "round_robin", "cache_aware", "power_of_two", "bucket"],
+            choices=[
+                "random",
+                "round_robin",
+                "cache_aware",
+                "power_of_two",
+                "manual",
+                "bucket",
+            ],
             help="Specific policy for prefill nodes in PD mode. If not specified, uses the main policy",
         )
         parser.add_argument(
             f"--{prefix}decode-policy",
             type=str,
             default=None,
-            choices=["random", "round_robin", "cache_aware", "power_of_two"],
+            choices=["random", "round_robin", "cache_aware", "power_of_two", "manual"],
             help="Specific policy for decode nodes in PD mode. If not specified, uses the main policy",
         )
 

sgl-model-gateway/bindings/python/src/lib.rs

@@ -12,6 +12,7 @@ pub enum PolicyType {
     CacheAware,
     PowerOfTwo,
     Bucket,
+    Manual,
 }
 
 #[pyclass(eq)]
@@ -267,6 +268,7 @@ impl Router {
                     balance_rel_threshold: self.balance_rel_threshold,
                     bucket_adjust_interval_secs: self.bucket_adjust_interval_secs,
                 },
+                PolicyType::Manual => ConfigPolicyConfig::Manual,
             }
         };