Track a histogram of state res durations (#13036)
2 files changed, 13 insertions, 0 deletions
diff --git a/changelog.d/13036.feature b/changelog.d/13036.feature
new file mode 100644
index 0000000000..71e5a29fe9
--- /dev/null
+++ b/changelog.d/13036.feature
@@ -0,0 +1 @@
+Add metrics measuring the CPU and DB time spent in state resolution.
diff --git a/synapse/state/__init__.py b/synapse/state/__init__.py
index da25f20ae5..9d3fe66100 100644
--- a/synapse/state/__init__.py
+++ b/synapse/state/__init__.py
@@ -444,6 +444,15 @@ _biggest_room_by_db_counter = Counter(
"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 @@ class StateResolutionHandler:
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.
|