refactor(router): move default worker selection into builtin plugin

.github/workflows/pre-merge.yml

@@ -541,7 +541,7 @@ jobs:
       working-directory: ${{ matrix.dir }}
       run: |
         # custom-policy is on by default; also lint it alongside the standalone selection
-        # service, and the build that links no worker-selection policies at all.
+        # service, and the build without the optional policy catalog.
         cargo clippy --locked --no-deps --features custom-policy,select-service --all-targets -- -D warnings
         cargo clippy --locked --no-deps --no-default-features --all-targets -- -D warnings
     - name: Lint feature-gated benchmark entrypoints

Cargo.toml

@@ -55,6 +55,7 @@ repository = "https://github.com/ai-dynamo/dynamo.git"
 keywords = ["llm", "genai", "inference", "nvidia", "distributed"]
 
 [workspace.dependencies]
+dynamo-custom-policy-builtin = { path = "lib/router-plugins/builtin", version = "1.6.0" }
 # Published separately by https://github.com/ai-dynamo/aisimulate.
 aisimulate-core = { version = "=0.12.0" }
 

components/src/dynamo/common/configuration/groups/kv_router_args.py

@@ -295,11 +295,12 @@ def add_arguments(self, parser) -> None:
             default=False,
             dest="load_aware",
             help=(
-                "KV Router: Enable load-aware routing without cache-reuse signals. "
+                "KV Router: Enable the load-aware routing preset. "
                 "On the frontend, this implies --router-mode kv. "
                 "This preset sets overlap_score_credit=0, disables KV events and "
                 "KV-reuse assumptions, enables active-block "
-                "and prefill-token load tracking, and disables remote/shared cache indexers."
+                "and prefill-token load tracking, and disables remote/shared cache indexers. "
+                "The builtin policy does not request cache inputs in this mode."
             ),
         )
         add_argument(

deploy/inference-gateway/ext-proc/Cargo.toml

@@ -16,6 +16,7 @@ name = "dynamo-ext-proc"
 path = "src/bin/dynamo-ext-proc.rs"
 
 [dependencies]
+dynamo-custom-policy-builtin = { workspace = true }
 async-trait = { workspace = true }
 dynamo-kv-router = { workspace = true, features = [
     "metrics",

deploy/inference-gateway/ext-proc/src/peer_discovery.rs

@@ -268,9 +268,7 @@ mod tests {
         use axum::{Json, Router};
         use dynamo_kv_router::WorkerType;
         use dynamo_kv_router::config::kv_router_config_from_dynamo_env;
-        use dynamo_kv_router::services::selection::{
-            SelectionServiceBuilder, WorkerSelectionPolicyRegistry,
-        };
+        use dynamo_kv_router::services::selection::SelectionServiceBuilder;
         use kube::Config;
         use tokio::net::TcpListener;
         use tokio::sync::Notify;
@@ -285,7 +283,7 @@ mod tests {
             SelectionServiceBuilder::new(
                 kv_router_config_from_dynamo_env(),
                 WorkerType::Aggregated,
-                WorkerSelectionPolicyRegistry::default(),
+                dynamo_custom_policy_builtin::default_registry(),
             )
             .indexer_threads(1)
             .replica_sync(listener_port, Vec::new())

deploy/inference-gateway/ext-proc/src/selector.rs

@@ -110,9 +110,12 @@ impl Selector {
             .await?;
         }
 
-        let mut builder =
-            SelectionServiceBuilder::new(kv_router_config, WorkerType::Aggregated, policy_registry)
-                .indexer_threads(cfg.selector_threads);
+        let mut builder = SelectionServiceBuilder::new(
+            kv_router_config,
+            WorkerType::Aggregated,
+            policy_registry.with_default_factory(dynamo_custom_policy_builtin::default_factory()),
+        )
+        .indexer_threads(cfg.selector_threads);
         if let Some(peer_replication) = peer_replication {
             builder = builder.replica_sync(peer_replication.sync_port, Vec::new());
         }

docs/fern/pages/developer-guide/knowledge-base/modular-components/router/configuration-and-tuning.md

@@ -70,13 +70,13 @@ its configuration even if the endpoint is unchanged.
 
 ## Routing Behavior
 
-- `--router-kv-overlap-score-credit`: Device-local prefix-overlap credit multiplier in the prefill cost calculation. It must be finite and nonnegative. Values greater than `1.0` give overlap extra credit, but the adjusted prefill contribution is clamped at zero. When set to `0`, the router ignores prefix caches and skips creating a local indexer. Defaults to `1.0`.
+- `--router-kv-overlap-score-credit`: Device-local prefix-overlap credit multiplier in the prefill cost calculation. It must be finite and nonnegative. Values greater than `1.0` give overlap extra credit, but the adjusted prefill contribution is clamped at zero. When set to `0`, device-local overlap receives no score credit. Cache setup still follows the policy's `WorkerInputs::CACHE` declaration. Defaults to `1.0`.
 - `--router-kv-overlap-score-credit-decay`: Decays device-local overlap credit for workers whose active prefill load exceeds the least-loaded eligible worker. `0` disables decay. Defaults to 0.
 - `--router-prefill-load-scale`: Scale applied to adjusted prompt-side prefill load after device, lower-tier, and shared-cache credits are subtracted. Defaults to 1.
 - `--router-decode-active-request-weight`: Experimental finite, nonnegative block-equivalent decode cost added for each active request on a candidate worker. Defaults to 0.
 - `--router-host-cache-hit-weight`: Credit multiplier for host-pinned (CPU offload) prefix overlap, from 0.0 to 1.0. Symmetric to `--router-kv-overlap-score-credit` but applied to the host-pinned tier when a backend exposes CPU offload via a KV connector. Defaults to 0.75.
 - `--router-disk-cache-hit-weight`: Credit multiplier for disk/lower-tier (e.g. NVMe-backed) prefix overlap, from 0.0 to 1.0. Defaults to 0.25.
-- `--load-aware`: Preset for load-aware KV routing without cache-reuse signals. On the frontend, it implies `--router-mode kv`. It sets `overlap_score_credit=0`, disables KV events and KV reuse assumptions, enables active-block and prefill-token load tracking, disables remote/shared cache indexers, and preserves `--router-prefill-load-scale`, `--router-host-cache-hit-weight`, and `--router-disk-cache-hit-weight`.
+- `--load-aware`: Preset for load-aware KV routing. On the frontend, it implies `--router-mode kv`. It sets `overlap_score_credit=0`, disables KV events and KV reuse assumptions, enables active-block and prefill-token load tracking, disables remote/shared cache indexers, and preserves `--router-prefill-load-scale`, `--router-host-cache-hit-weight`, and `--router-disk-cache-hit-weight`. The policy's `WorkerInputs::CACHE` declaration still controls local approximate indexing; the builtin default declares `CACHE`.
 - `--router-temperature`: Controls worker selection randomness through softmax sampling of normalized router cost logits. A value of 0 (default) ensures deterministic selection of the lowest-cost worker, while higher values introduce more randomness.
 - `--router-conditional-disagg`: **Experimental.** Enables conditional disaggregation in frontend-embedded disaggregated serving. Requires `--router-mode kv`, `--router-kv-events`, separate prefill/decode worker pools, and decode-worker KV event publishing. Use `--router-conditional-disagg-config` for policy settings. See [Conditional Disaggregation](../../../advanced-customizations/conditional-disaggregation.md) for backend requirements and policy tuning.
 - `--router-track-prefill-tokens`: Enables prompt-side load accounting in the worker cost model. This should stay enabled if you want queue thresholds, `active_prefill_tokens`, and AIC prefill load decay to reflect prompt work.
@@ -445,7 +445,7 @@ Use `--router-kv-overlap-score-credit-decay` to reduce that device-local credit
 
 Use `--load-aware` when you want the KV scheduler's active load model without prefix/cache reuse. This is equivalent to using KV mode with overlap credit set to 0, KV events disabled, KV reuse assumptions disabled, active load tracking enabled, and shared-cache routing disabled. `--router-prefill-load-scale` remains available to tune prompt-side load relative to decode blocks.
 
-Deprecated: `--router-kv-overlap-score-weight`, `--kv-overlap-score-weight`, `DYN_ROUTER_KV_OVERLAP_SCORE_WEIGHT`, and `DYN_OVERLAP_SCORE_WEIGHT` are still accepted, but emit deprecation warnings. Nonzero legacy values map to `prefill_load_scale` to preserve existing behavior without changing overlap credit. A legacy value of 0 maps to both `prefill_load_scale=0` and `overlap_score_credit=0`, which preserves the old no-overlap/no-indexer behavior. If a deprecated overlap score weight is still present, it takes precedence over the newer prefill load scale field; a legacy value of 0 also takes precedence over the newer overlap credit field. When migrating to `--router-prefill-load-scale` or `DYN_ROUTER_PREFILL_LOAD_SCALE`, remove the deprecated flag, env var, or JSON field from the deployment config. Use `--router-kv-overlap-score-credit` or `DYN_ROUTER_KV_OVERLAP_SCORE_CREDIT` only when you mean to tune the cache-overlap credit itself.
+Deprecated: `--router-kv-overlap-score-weight`, `--kv-overlap-score-weight`, `DYN_ROUTER_KV_OVERLAP_SCORE_WEIGHT`, and `DYN_OVERLAP_SCORE_WEIGHT` are still accepted, but emit deprecation warnings. Nonzero legacy values map to `prefill_load_scale` to preserve existing behavior without changing overlap credit. A legacy value of 0 maps to both `prefill_load_scale=0` and `overlap_score_credit=0`, which zeros those scoring weights. Indexer setup follows the policy's `WorkerInputs::CACHE` declaration independently of these values. If a deprecated overlap score weight is still present, it takes precedence over the newer prefill load scale field; a legacy value of 0 also takes precedence over the newer overlap credit field. When migrating to `--router-prefill-load-scale` or `DYN_ROUTER_PREFILL_LOAD_SCALE`, remove the deprecated flag, env var, or JSON field from the deployment config. Use `--router-kv-overlap-score-credit` or `DYN_ROUTER_KV_OVERLAP_SCORE_CREDIT` only when you mean to tune the cache-overlap credit itself.
 
 When migrating the deprecated overlap score weight, use `--router-prefill-load-scale` to preserve its scaling role. Tune `--router-kv-overlap-score-credit` separately only when you intend to change device-local cache credit; values above `1.0` are supported, with adjusted prefill cost clamped at zero.