diff --git a/synapse/util/metrics.py b/synapse/util/metrics.py
index 4ea930d3e8..97f1267380 100644
--- a/synapse/util/metrics.py
+++ b/synapse/util/metrics.py
@@ -13,40 +13,37 @@
# See the License for the specific language governing permissions and
# limitations under the License.
-from twisted.internet import defer
+import logging
+from functools import wraps
-from synapse.util.logcontext import LoggingContext
-import synapse.metrics
+from prometheus_client import Counter
-from functools import wraps
-import logging
+from twisted.internet import defer
+from synapse.util.logcontext import LoggingContext
logger = logging.getLogger(__name__)
+block_counter = Counter("synapse_util_metrics_block_count", "", ["block_name"])
-metrics = synapse.metrics.get_metrics_for(__name__)
+block_timer = Counter("synapse_util_metrics_block_time_seconds", "", ["block_name"])
-block_timer = metrics.register_distribution(
- "block_timer",
- labels=["block_name"]
-)
+block_ru_utime = Counter(
+ "synapse_util_metrics_block_ru_utime_seconds", "", ["block_name"])
-block_ru_utime = metrics.register_distribution(
- "block_ru_utime", labels=["block_name"]
-)
+block_ru_stime = Counter(
+ "synapse_util_metrics_block_ru_stime_seconds", "", ["block_name"])
-block_ru_stime = metrics.register_distribution(
- "block_ru_stime", labels=["block_name"]
-)
+block_db_txn_count = Counter(
+ "synapse_util_metrics_block_db_txn_count", "", ["block_name"])
-block_db_txn_count = metrics.register_distribution(
- "block_db_txn_count", labels=["block_name"]
-)
+# seconds spent waiting for db txns, excluding scheduling time, in this block
+block_db_txn_duration = Counter(
+ "synapse_util_metrics_block_db_txn_duration_seconds", "", ["block_name"])
-block_db_txn_duration = metrics.register_distribution(
- "block_db_txn_duration", labels=["block_name"]
-)
+# seconds spent waiting for a db connection, in this block
+block_db_sched_duration = Counter(
+ "synapse_util_metrics_block_db_sched_duration_seconds", "", ["block_name"])
def measure_func(name):
@@ -63,8 +60,9 @@ def measure_func(name):
class Measure(object):
__slots__ = [
- "clock", "name", "start_context", "start", "new_context", "ru_utime",
- "ru_stime", "db_txn_count", "db_txn_duration", "created_context"
+ "clock", "name", "start_context", "start",
+ "created_context",
+ "start_usage",
]
def __init__(self, clock, name):
@@ -75,23 +73,23 @@ class Measure(object):
self.created_context = False
def __enter__(self):
- self.start = self.clock.time_msec()
+ self.start = self.clock.time()
self.start_context = LoggingContext.current_context()
if not self.start_context:
self.start_context = LoggingContext("Measure")
self.start_context.__enter__()
self.created_context = True
- self.ru_utime, self.ru_stime = self.start_context.get_resource_usage()
- self.db_txn_count = self.start_context.db_txn_count
- self.db_txn_duration = self.start_context.db_txn_duration
+ self.start_usage = self.start_context.get_resource_usage()
def __exit__(self, exc_type, exc_val, exc_tb):
if isinstance(exc_type, Exception) or not self.start_context:
return
- duration = self.clock.time_msec() - self.start
- block_timer.inc_by(duration, self.name)
+ duration = self.clock.time() - self.start
+
+ block_counter.labels(self.name).inc()
+ block_timer.labels(self.name).inc(duration)
context = LoggingContext.current_context()
@@ -106,16 +104,19 @@ class Measure(object):
logger.warn("Expected context. (%r)", self.name)
return
- ru_utime, ru_stime = context.get_resource_usage()
-
- block_ru_utime.inc_by(ru_utime - self.ru_utime, self.name)
- block_ru_stime.inc_by(ru_stime - self.ru_stime, self.name)
- block_db_txn_count.inc_by(
- context.db_txn_count - self.db_txn_count, self.name
- )
- block_db_txn_duration.inc_by(
- context.db_txn_duration - self.db_txn_duration, self.name
- )
+ current = context.get_resource_usage()
+ usage = current - self.start_usage
+ try:
+ block_ru_utime.labels(self.name).inc(usage.ru_utime)
+ block_ru_stime.labels(self.name).inc(usage.ru_stime)
+ block_db_txn_count.labels(self.name).inc(usage.db_txn_count)
+ block_db_txn_duration.labels(self.name).inc(usage.db_txn_duration_sec)
+ block_db_sched_duration.labels(self.name).inc(usage.db_sched_duration_sec)
+ except ValueError:
+ logger.warn(
+ "Failed to save metrics! OLD: %r, NEW: %r",
+ self.start_usage, current
+ )
if self.created_context:
self.start_context.__exit__(exc_type, exc_val, exc_tb)
|
