Merge pull request #3256 from matrix-org/3218-official-prom

Switch to the Python Prometheus library

2 files changed, 55 insertions, 73 deletions

diff --git a/synapse/replication/tcp/protocol.py b/synapse/replication/tcp/protocol.py
index 7ca1588f6a..a6280aae70 100644
--- a/synapse/replication/tcp/protocol.py
+++ b/synapse/replication/tcp/protocol.py
@@ -60,22 +60,21 @@ from .commands import (
 )
 from .streams import STREAMS_MAP
 
+from synapse.metrics import LaterGauge
 from synapse.util.stringutils import random_string
-from synapse.metrics.metric import CounterMetric
 
-import logging
-import synapse.metrics
-import struct
-import fcntl
+from prometheus_client import Counter
 
-from six import iterkeys, iteritems
+from collections import defaultdict
 
-metrics = synapse.metrics.get_metrics_for(__name__)
+from six import iterkeys, iteritems
 
-connection_close_counter = metrics.register_counter(
-    "close_reason", labels=["reason_type"],
-)
+import logging
+import struct
+import fcntl
 
+connection_close_counter = Counter(
+    "synapse_replication_tcp_protocol_close_reason", "", ["reason_type"])
 
 # A list of all connected protocols. This allows us to send metrics about the
 # connections.
@@ -137,12 +136,8 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver):
         # The LoopingCall for sending pings.
         self._send_ping_loop = None
 
-        self.inbound_commands_counter = CounterMetric(
-            "inbound_commands", labels=["command"],
-        )
-        self.outbound_commands_counter = CounterMetric(
-            "outbound_commands", labels=["command"],
-        )
+        self.inbound_commands_counter = defaultdict(int)
+        self.outbound_commands_counter = defaultdict(int)
 
     def connectionMade(self):
         logger.info("[%s] Connection established", self.id())
@@ -202,7 +197,8 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver):
 
         self.last_received_command = self.clock.time_msec()
 
-        self.inbound_commands_counter.inc(cmd_name)
+        self.inbound_commands_counter[cmd_name] = (
+            self.inbound_commands_counter[cmd_name] + 1)
 
         cmd_cls = COMMAND_MAP[cmd_name]
         try:
@@ -252,8 +248,8 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver):
             self._queue_command(cmd)
             return
 
-        self.outbound_commands_counter.inc(cmd.NAME)
-
+        self.outbound_commands_counter[cmd.NAME] = (
+            self.outbound_commands_counter[cmd.NAME] + 1)
         string = "%s %s" % (cmd.NAME, cmd.to_line(),)
         if "\n" in string:
             raise Exception("Unexpected newline in command: %r", string)
@@ -318,9 +314,9 @@ class BaseReplicationStreamProtocol(LineOnlyReceiver):
     def connectionLost(self, reason):
         logger.info("[%s] Replication connection closed: %r", self.id(), reason)
         if isinstance(reason, Failure):
-            connection_close_counter.inc(reason.type.__name__)
+            connection_close_counter.labels(reason.type.__name__).inc()
         else:
-            connection_close_counter.inc(reason.__class__.__name__)
+            connection_close_counter.labels(reason.__class__.__name__).inc()
 
         try:
             # Remove us from list of connections to be monitored
@@ -519,7 +515,7 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol):
 
     def on_RDATA(self, cmd):
         stream_name = cmd.stream_name
-        inbound_rdata_count.inc(stream_name)
+        inbound_rdata_count.labels(stream_name).inc()
 
         try:
             row = STREAMS_MAP[stream_name].ROW_TYPE(*cmd.row)
@@ -567,14 +563,12 @@ class ClientReplicationStreamProtocol(BaseReplicationStreamProtocol):
 
 # The following simply registers metrics for the replication connections
 
-metrics.register_callback(
-    "pending_commands",
+pending_commands = LaterGauge(
+    "pending_commands", "", ["name", "conn_id"],
     lambda: {
         (p.name, p.conn_id): len(p.pending_commands)
         for p in connected_connections
-    },
-    labels=["name", "conn_id"],
-)
+    })
 
 
 def transport_buffer_size(protocol):
@@ -584,14 +578,12 @@ def transport_buffer_size(protocol):
     return 0
 
 
-metrics.register_callback(
-    "transport_send_buffer",
+transport_send_buffer = LaterGauge(
+    "synapse_replication_tcp_transport_send_buffer", "", ["name", "conn_id"],
     lambda: {
         (p.name, p.conn_id): transport_buffer_size(p)
         for p in connected_connections
-    },
-    labels=["name", "conn_id"],
-)
+    })
 
 
 def transport_kernel_read_buffer_size(protocol, read=True):
@@ -609,48 +601,38 @@ def transport_kernel_read_buffer_size(protocol, read=True):
     return 0
 
 
-metrics.register_callback(
-    "transport_kernel_send_buffer",
+tcp_transport_kernel_send_buffer = LaterGauge(
+    "synapse_replication_tcp_transport_kernel_send_buffer", "", ["name", "conn_id"],
     lambda: {
         (p.name, p.conn_id): transport_kernel_read_buffer_size(p, False)
         for p in connected_connections
-    },
-    labels=["name", "conn_id"],
-)
+    })
 
 
-metrics.register_callback(
-    "transport_kernel_read_buffer",
+tcp_transport_kernel_read_buffer = LaterGauge(
+    "synapse_replication_tcp_transport_kernel_read_buffer", "", ["name", "conn_id"],
     lambda: {
         (p.name, p.conn_id): transport_kernel_read_buffer_size(p, True)
         for p in connected_connections
-    },
-    labels=["name", "conn_id"],
-)
+    })
 
 
-metrics.register_callback(
-    "inbound_commands",
+tcp_inbound_commands = LaterGauge(
+    "synapse_replication_tcp_inbound_commands", "", ["command", "name", "conn_id"],
     lambda: {
         (k[0], p.name, p.conn_id): count
         for p in connected_connections
         for k, count in iteritems(p.inbound_commands_counter.counts)
-    },
-    labels=["command", "name", "conn_id"],
-)
+    })
 
-metrics.register_callback(
-    "outbound_commands",
+tcp_outbound_commands = LaterGauge(
+    "synapse_replication_tcp_outbound_commands", "", ["command", "name", "conn_id"],
     lambda: {
         (k[0], p.name, p.conn_id): count
         for p in connected_connections
         for k, count in iteritems(p.outbound_commands_counter.counts)
-    },
-    labels=["command", "name", "conn_id"],
-)
+    })
 
 # number of updates received for each RDATA stream
-inbound_rdata_count = metrics.register_counter(
-    "inbound_rdata_count",
-    labels=["stream_name"],
-)
+inbound_rdata_count = Counter("synapse_replication_tcp_inbound_rdata_count", "",
+                              ["stream_name"])
diff --git a/synapse/replication/tcp/resource.py b/synapse/replication/tcp/resource.py
index 1969072ab9..63bd6d2652 100644
--- a/synapse/replication/tcp/resource.py
+++ b/synapse/replication/tcp/resource.py
@@ -22,21 +22,21 @@ from .streams import STREAMS_MAP, FederationStream
 from .protocol import ServerReplicationStreamProtocol
 
 from synapse.util.metrics import Measure, measure_func
+from synapse.metrics import LaterGauge
 
 import logging
-import synapse.metrics
 
+from prometheus_client import Counter
 from six import itervalues
 
-metrics = synapse.metrics.get_metrics_for(__name__)
-stream_updates_counter = metrics.register_counter(
-    "stream_updates", labels=["stream_name"]
-)
-user_sync_counter = metrics.register_counter("user_sync")
-federation_ack_counter = metrics.register_counter("federation_ack")
-remove_pusher_counter = metrics.register_counter("remove_pusher")
-invalidate_cache_counter = metrics.register_counter("invalidate_cache")
-user_ip_cache_counter = metrics.register_counter("user_ip_cache")
+stream_updates_counter = Counter("synapse_replication_tcp_resource_stream_updates",
+                                 "", ["stream_name"])
+user_sync_counter = Counter("synapse_replication_tcp_resource_user_sync", "")
+federation_ack_counter = Counter("synapse_replication_tcp_resource_federation_ack", "")
+remove_pusher_counter = Counter("synapse_replication_tcp_resource_remove_pusher", "")
+invalidate_cache_counter = Counter("synapse_replication_tcp_resource_invalidate_cache",
+                                   "")
+user_ip_cache_counter = Counter("synapse_replication_tcp_resource_user_ip_cache", "")
 
 logger = logging.getLogger(__name__)
 
@@ -75,7 +75,8 @@ class ReplicationStreamer(object):
         # Current connections.
         self.connections = []
 
-        metrics.register_callback("total_connections", lambda: len(self.connections))
+        LaterGauge("synapse_replication_tcp_resource_total_connections", "", [],
+                   lambda: len(self.connections))
 
         # List of streams that clients can subscribe to.
         # We only support federation stream if federation sending hase been
@@ -87,17 +88,16 @@ class ReplicationStreamer(object):
 
         self.streams_by_name = {stream.NAME: stream for stream in self.streams}
 
-        metrics.register_callback(
-            "connections_per_stream",
+        LaterGauge(
+            "synapse_replication_tcp_resource_connections_per_stream", "",
+            ["stream_name"],
             lambda: {
                 (stream_name,): len([
                     conn for conn in self.connections
                     if stream_name in conn.replication_streams
                 ])
                 for stream_name in self.streams_by_name
-            },
-            labels=["stream_name"],
-        )
+            })
 
         self.federation_sender = None
         if not hs.config.send_federation:
@@ -177,7 +177,7 @@ class ReplicationStreamer(object):
                             logger.info(
                                 "Streaming: %s -> %s", stream.NAME, updates[-1][0]
                             )
-                            stream_updates_counter.inc_by(len(updates), stream.NAME)
+                            stream_updates_counter.labels(stream.NAME).inc(len(updates))
 
                         # Some streams return multiple rows with the same stream IDs,
                         # we need to make sure they get sent out in batches. We do