Merge remote-tracking branch 'origin/develop' into 3218-official-prom

synapse/http/request_metrics.py

@@ -17,6 +17,7 @@
 import logging
 
 from prometheus_client.core import Counter, Histogram
+from synapse.metrics import LaterGauge
 
 from synapse.util.logcontext import LoggingContext
 
@@ -50,14 +51,75 @@ response_db_sched_duration = Counter("synapse_http_request_response_db_sched_dur
 response_size = Counter("synapse_http_request_response_size", "", ["method", "servlet", "tag"]
 )
 
+# In flight metrics are incremented while the requests are in flight, rather
+# than when the response was written.
+
+in_flight_requests_ru_utime = Counter("synapse_http_request_in_flight_requests_ru_utime_seconds", "", ["method", "servlet"])
+
+in_flight_requests_ru_stime = Counter("synapse_http_request_in_flight_requests_ru_stime_seconds", "", ["method", "servlet"])
+
+in_flight_requests_db_txn_count = Counter("synapse_http_request_in_flight_requests_db_txn_count", "", ["method", "servlet"])
+
+# seconds spent waiting for db txns, excluding scheduling time, when processing
+# this request
+in_flight_requests_db_txn_duration = Counter("synapse_http_request_in_flight_requests_db_txn_duration_seconds", "", ["method", "servlet"])
+
+# seconds spent waiting for a db connection, when processing this request
+in_flight_requests_db_sched_duration = Counter("synapse_http_request_in_flight_requests_db_sched_duration_seconds", "", ["method", "servlet"])
+
+# The set of all in flight requests, set[RequestMetrics]
+_in_flight_requests = set()
+
+
+def _collect_in_flight():
+    """Called just before metrics are collected, so we use it to update all
+    the in flight request metrics
+    """
+
+    for rm in _in_flight_requests:
+        rm.update_metrics()
+
+
+metrics.register_collector(_collect_in_flight)
+
+
+def _get_in_flight_counts():
+    """Returns a count of all in flight requests by (method, server_name)
+
+    Returns:
+        dict[tuple[str, str], int]
+    """
+
+    # Map from (method, name) -> int, the number of in flight requests of that
+    # type
+    counts = {}
+    for rm in _in_flight_requests:
+        key = (rm.method, rm.name,)
+        counts[key] = counts.get(key, 0) + 1
+
+    return counts
+
+LaterGauge(
+    "synapse_http_request_metrics_in_flight_requests_count", "",
+    ["method", "servlet"],
+    _get_in_flight_counts
+)
+
 
 class RequestMetrics(object):
-    def start(self, time_msec, name):
+    def start(self, time_msec, name, method):
         self.start = time_msec
         self.start_context = LoggingContext.current_context()
         self.name = name
+        self.method = method
+
+        self._request_stats = _RequestStats.from_context(self.start_context)
+
+        _in_flight_requests.add(self)
 
     def stop(self, time_msec, request):
+        _in_flight_requests.discard(self)
+
         context = LoggingContext.current_context()
 
         tag = ""
@@ -87,3 +149,76 @@ class RequestMetrics(object):
             context.db_sched_duration_ms / 1000.)
 
         response_size.labels(request.method, self.name, tag).inc(request.sentLength)
+
+        # We always call this at the end to ensure that we update the metrics
+        # regardless of whether a call to /metrics while the request was in
+        # flight.
+        self.update_metrics()
+
+    def update_metrics(self):
+        """Updates the in flight metrics with values from this request.
+        """
+        diff = self._request_stats.update(self.start_context)
+
+        in_flight_requests_ru_utime.labels(self.method, self.name).inc(diff.ru_utime)
+        in_flight_requests_ru_stime.labels(self.method, self.name).inc(diff.ru_stime)
+
+        in_flight_requests_db_txn_count.labels(self.method, self.name).inc(diff.db_txn_count)
+
+        in_flight_requests_db_txn_duration.labels(self.method, self.name).inc(diff.db_txn_duration_ms / 1000.)
+
+        in_flight_requests_db_sched_duration.labels(self.method, self.name).inc(diff.db_sched_duration_ms / 1000.)
+
+
+class _RequestStats(object):
+    """Keeps tracks of various metrics for an in flight request.
+    """
+
+    __slots__ = [
+        "ru_utime", "ru_stime",
+        "db_txn_count", "db_txn_duration_ms", "db_sched_duration_ms",
+    ]
+
+    def __init__(self, ru_utime, ru_stime, db_txn_count,
+                 db_txn_duration_ms, db_sched_duration_ms):
+        self.ru_utime = ru_utime
+        self.ru_stime = ru_stime
+        self.db_txn_count = db_txn_count
+        self.db_txn_duration_ms = db_txn_duration_ms
+        self.db_sched_duration_ms = db_sched_duration_ms
+
+    @staticmethod
+    def from_context(context):
+        ru_utime, ru_stime = context.get_resource_usage()
+
+        return _RequestStats(
+            ru_utime, ru_stime,
+            context.db_txn_count,
+            context.db_txn_duration_ms,
+            context.db_sched_duration_ms,
+        )
+
+    def update(self, context):
+        """Updates the current values and returns the difference between the
+        old and new values.
+
+        Returns:
+            _RequestStats: The difference between the old and new values
+        """
+        new = _RequestStats.from_context(context)
+
+        diff = _RequestStats(
+            new.ru_utime - self.ru_utime,
+            new.ru_stime - self.ru_stime,
+            new.db_txn_count - self.db_txn_count,
+            new.db_txn_duration_ms - self.db_txn_duration_ms,
+            new.db_sched_duration_ms - self.db_sched_duration_ms,
+        )
+
+        self.ru_utime = new.ru_utime
+        self.ru_stime = new.ru_stime
+        self.db_txn_count = new.db_txn_count
+        self.db_txn_duration_ms = new.db_txn_duration_ms
+        self.db_sched_duration_ms = new.db_sched_duration_ms
+
+        return diff