perf(policies): drop per-request locks and string probes from routing scoring

crates/kv_index/benches/throughput_bench.rs

@@ -667,7 +667,9 @@ async fn run_benchmark(args: &Args, traces: Vec<Vec<TimedEntry>>) -> BenchmarkRe
 
     let num_total_workers = args.num_workers * args.duplication_factor;
     for w in 0..num_total_workers {
-        indexer.intern_worker(&format!("worker-{w}"));
+        indexer
+            .intern_worker(&format!("worker-{w}"))
+            .expect("worker id space exhausted");
     }
 
     let traces: Vec<Arc<Vec<TimedEntry>>> = traces.into_iter().map(Arc::new).collect();

crates/kv_index/src/lib.rs

@@ -21,7 +21,7 @@ mod token_tree;
 pub use common::{MatchResult, TenantId};
 pub use event_tree::{
     compute_content_hash, compute_request_content_hashes, ApplyError, ContentHash, OverlapScores,
-    PositionalIndexer, SequenceHash, StoredBlock, WorkerBlockMap, WorkerId,
+    PositionalIndexer, SequenceHash, StoredBlock, WorkerBlockMap, WorkerId, WorkerIdExhausted,
 };
 pub use path_hash::{hash_node_path, hash_token_path, GLOBAL_EVICTION_HASH};
 // Re-export under names matching old tree.rs API for easier migration

model_gateway/benches/radix_tree_benchmark.rs

@@ -445,7 +445,7 @@ fn build_populated_indexer(
     let mut all_worker_chunks = Vec::with_capacity(workers.len());
 
     for worker in workers {
-        let worker_id = indexer.intern_worker(worker);
+        let worker_id = indexer.intern_worker(worker).unwrap();
         let mut wb = WorkerBlockMap::default();
         indexer
             .apply_stored(worker_id, &shared_blocks, None, &mut wb)
@@ -500,7 +500,7 @@ macro_rules! bench_indexer_store {
                 for _ in 0..iters {
                     let indexer = PositionalIndexer::new(32);
                     for worker in &workers {
-                        let worker_id = indexer.intern_worker(worker);
+                        let worker_id = indexer.intern_worker(worker).unwrap();
                         let mut wb = WorkerBlockMap::default();
                         let chunks = generate_token_chunks($blocks_per_worker, $block_size);
                         let blocks = chunks_to_stored_blocks(&chunks);
@@ -596,7 +596,7 @@ macro_rules! bench_indexer_concurrent {
             (0..$num_threads)
                 .map(|t| {
                     let chunks = &worker_chunks[t % workers.len()];
-                    let worker_id = indexer.intern_worker(&workers[t % workers.len()]);
+                    let worker_id = indexer.intern_worker(&workers[t % workers.len()]).unwrap();
 
                     // Read data: pre-computed content hashes
                     let query_tokens = flatten_tokens(chunks);

model_gateway/src/observability/metrics.rs

@@ -230,6 +230,11 @@ pub(crate) fn init_metrics() {
         "smg_worker_errors_total",
         "Worker-level errors by worker_type, connection_mode, error_type"
     );
+    describe_counter!(
+        "smg_kv_event_subscription_failures_total",
+        "KV event subscription task failures by worker and reason \
+         (panic, join_error, intern_failed)"
+    );
     describe_gauge!(
         "smg_manual_policy_cache_entries",
         "Number of routing entries in manual policy cache"
@@ -944,6 +949,18 @@ impl Metrics {
         .set(if healthy { 1.0 } else { 0.0 });
     }
 
+    /// Record a KV event subscription task failure (panic, join error, or
+    /// worker-id intern failure)
+    pub fn record_kv_event_subscription_failure(worker_url: &str, reason: &'static str) {
+        let worker_interned = intern_string(worker_url);
+        counter!(
+            "smg_kv_event_subscription_failures_total",
+            "worker" => worker_interned,
+            "reason" => reason
+        )
+        .increment(1);
+    }
+
     // ========================================================================
     // Layer 3: Worker resilience metrics (circuit breaker)
     // ========================================================================