Expose perf cache counts by bounded policy

Author: durable-workflow-ops

config/dw-bounded-growth.php

@@ -219,6 +219,17 @@
             'suppression' => 'No labels are exposed.',
         ],
 
+        'dw_perf_redis_server_keys_by_policy' => [
+            'owner' => 'scripts/perf/server_soak.py',
+            'surface' => 'Perf harness /metrics scrape; optional remote_write.',
+            'dimensions' => [
+                'policy' => 'finite_cache_policy_inventory',
+            ],
+            'cardinality' => 'policy series are fixed to the cache_keys inventory in this bounded-growth policy file.',
+            'selection' => 'latest sampled Redis keys for each declared server-owned cache policy.',
+            'suppression' => 'No suppression path is needed because the cache policy inventory is finite and reviewed.',
+        ],
+
         'dw_perf_redis_db_keys' => [
             'owner' => 'scripts/perf/server_soak.py',
             'surface' => 'Perf harness /metrics scrape; optional remote_write.',

docs/bounded-growth.md

@@ -53,6 +53,7 @@ added without a TTL, admission, or cardinality contract.
 | `dw_perf_redis_memory_bytes` | Perf harness `/metrics`; optional remote_write | No labels; single gauge series per soak run. |
 | `dw_perf_redis_polling_keys` | Perf harness `/metrics`; optional remote_write | No labels; single gauge series per soak run. |
 | `dw_perf_redis_server_keys` | Perf harness `/metrics`; optional remote_write | No labels; single gauge series per soak run. Counts all Redis keys in the server-owned `server:*` cache namespace, not only the workflow-task polling subset. |
+| `dw_perf_redis_server_keys_by_policy` | Perf harness `/metrics`; optional remote_write | The only label is `policy`, fixed to the reviewed `cache_keys` inventory in `config/dw-bounded-growth.php`. |
 | `dw_perf_redis_db_keys` | Perf harness `/metrics`; optional remote_write | No labels; single gauge series per soak run. |
 | `dw_perf_assertion_failed` | Perf harness `/metrics`; optional remote_write | No labels; single gauge series per soak run. |
 
@@ -101,7 +102,9 @@ evidence.
 `max_server_cache_keys_by_policy` / `final_server_cache_keys_by_policy`. Those
 per-policy maps mirror the `cache_keys` inventory so a long soak can show which
 bounded cache family produced growth instead of only reporting a total
-`server:*` count.
+`server:*` count. The same finite per-policy inventory is exposed as
+`dw_perf_redis_server_keys_by_policy{policy="..."}` for optional remote-write
+alerting.
 
 Per-policy limits can be enforced with JSON maps keyed by policy ID:
 `DW_PERF_MAX_SERVER_CACHE_KEYS_BY_POLICY` for maximum observed keys and

scripts/perf/server_soak.py

@@ -51,6 +51,7 @@ def __init__(self) -> None:
             "redis_db_keys": 0,
             "redis_polling_keys": 0,
             "redis_server_keys": 0,
+            "redis_server_keys_by_policy": {policy_id: 0 for policy_id in SERVER_CACHE_KEY_PATTERNS},
             "assertion_failed": 0,
         }
 
@@ -71,6 +72,12 @@ def update_sample(self, sample: dict[str, Any]) -> None:
             self.latest["redis_db_keys"] = int(sample.get("redis_db_keys") or 0)
             self.latest["redis_polling_keys"] = int(sample.get("redis_polling_keys") or 0)
             self.latest["redis_server_keys"] = int(sample.get("redis_server_keys") or 0)
+            by_policy = sample.get("redis_server_keys_by_policy")
+            if isinstance(by_policy, dict):
+                self.latest["redis_server_keys_by_policy"] = {
+                    policy_id: int(by_policy.get(policy_id) or 0)
+                    for policy_id in SERVER_CACHE_KEY_PATTERNS
+                }
 
     def mark_assertion_failed(self) -> None:
         with self.lock:
@@ -106,6 +113,12 @@ def prometheus(self) -> str:
                     "# HELP dw_perf_redis_server_keys Redis keys owned by the Durable Workflow server cache namespace.",
                     "# TYPE dw_perf_redis_server_keys gauge",
                     f"dw_perf_redis_server_keys {self.latest['redis_server_keys']}",
+                    "# HELP dw_perf_redis_server_keys_by_policy Redis keys owned by the Durable Workflow server cache namespace by bounded-growth policy.",
+                    "# TYPE dw_perf_redis_server_keys_by_policy gauge",
+                    *[
+                        f'dw_perf_redis_server_keys_by_policy{{policy="{policy_id}"}} {count}'
+                        for policy_id, count in sorted(self.latest["redis_server_keys_by_policy"].items())
+                    ],
                     "# HELP dw_perf_redis_db_keys Redis DBSIZE count.",
                     "# TYPE dw_perf_redis_db_keys gauge",
                     f"dw_perf_redis_db_keys {self.latest['redis_db_keys']}",

tests/Unit/ServerPerfHarnessContractTest.php

@@ -25,6 +25,7 @@ public function test_soak_summary_records_trusted_evidence_fields(): void
             'final_server_cache_keys',
             'max_server_cache_keys_by_policy',
             'final_server_cache_keys_by_policy',
+            'dw_perf_redis_server_keys_by_policy',
             'DW_PERF_MAX_SERVER_CACHE_KEYS_BY_POLICY',
             'DW_PERF_MAX_FINAL_SERVER_CACHE_KEYS_BY_POLICY',
             'parse_policy_thresholds',