Track a histogram of state res durations

changelog.d/13036.feature

@@ -0,0 +1 @@
+Add metrics measuring the CPU and DB time spent in state resolution.

synapse/state/__init__.py

@@ -444,6 +444,15 @@ class _StateResMetrics:
     "expensive room for state resolution",
 )
 
+_cpu_times = Histogram(
+    "synapse_state_res_cpu_for_all_rooms_seconds",
+    "CPU time (utime+stime) spent computing a single state resolution",
+)
+_db_times = Histogram(
+    "synapse_state_res_db_for_all_rooms_seconds",
+    "Database time spent computing a single state resolution",
+)
+
 
 class StateResolutionHandler:
     """Responsible for doing state conflict resolution.
@@ -609,6 +618,9 @@ def _record_state_res_metrics(
         room_metrics.db_time += rusage.db_txn_duration_sec
         room_metrics.db_events += rusage.evt_db_fetch_count
 
+        _cpu_times.observe(rusage.ru_utime + rusage.ru_stime)
+        _db_times.observe(rusage.db_txn_duration_sec)
+
     def _report_metrics(self) -> None:
         if not self._state_res_metrics:
             # no state res has happened since the last iteration: don't bother logging.