Merge pull request #3066 from matrix-org/rav/remove_redundant_metrics

Remove redundant metrics which were deprecated in 0.27.0.

CHANGES.rst

@@ -1,3 +1,12 @@
+Changes in synapse v0.28.0 (2018-xx-xx)
+=======================================
+
+As previously advised, this release removes a number of redundant Prometheus
+metrics. Administrators may need to update their dashboards and alerting rules
+to use the updated metric names, if they have not already done so. See
+`docs/metrics-howto.rst <docs/metrics-howto.rst#deprecated-metrics-removed-in-0-28-0>`_
+for more details.
+
 Changes in synapse v0.27.2 (2018-03-26)
 =======================================
 

UPGRADE.rst

@@ -52,7 +52,7 @@ Upgrading to $NEXT_VERSION
 ====================
 
 This release expands the anonymous usage stats sent if the opt-in
-``report_stats`` configuration is set to ``true``. We now capture RSS memory 
+``report_stats`` configuration is set to ``true``. We now capture RSS memory
 and cpu use at a very coarse level. This requires administrators to install
 the optional ``psutil`` python module.
 
@@ -60,6 +60,13 @@ We would appreciate it if you could assist by ensuring this module is available
 and ``report_stats`` is enabled. This will let us see if performance changes to
 synapse are having an impact to the general community.
 
+This release also removes a number of redundant Prometheus metrics.
+Administrators may need to update their dashboards and alerting rules to use
+the updated metric names, if they have not already done so. See
+`docs/metrics-howto.rst <docs/metrics-howto.rst#deprecated-metrics-removed-in-0-28-0>`_
+for more details.
+
+
 Upgrading to v0.15.0
 ====================
 

docs/metrics-howto.rst

@@ -34,6 +34,17 @@ How to monitor Synapse metrics using Prometheus
    Restart prometheus.
 
 
+Deprecated metrics removed in 0.28.0
+------------------------------------
+
+Synapse 0.28.0 removes all of the metrics deprecated by 0.27.0, which are those
+listed under "Old name" below. This has been done to reduce the bandwidth used
+by gathering metrics and the storage requirements for the Prometheus server, as
+well as reducing CPU overhead for both Synapse and Prometheus.
+
+Administrators should update any alerts or monitoring dashboards to use the
+"New name" listed below.
+
 Block and response metrics renamed for 0.27.0
 ---------------------------------------------
 

synapse/http/server.py

@@ -47,17 +47,6 @@ metrics = synapse.metrics.get_metrics_for(__name__)
 response_count = metrics.register_counter(
     "response_count",
     labels=["method", "servlet", "tag"],
-    alternative_names=(
-        # the following are all deprecated aliases for the same metric
-        metrics.name_prefix + x for x in (
-            "_requests",
-            "_response_time:count",
-            "_response_ru_utime:count",
-            "_response_ru_stime:count",
-            "_response_db_txn_count:count",
-            "_response_db_txn_duration:count",
-        )
-    )
 )
 
 requests_counter = metrics.register_counter(
@@ -73,39 +62,24 @@ outgoing_responses_counter = metrics.register_counter(
 response_timer = metrics.register_counter(
     "response_time_seconds",
     labels=["method", "servlet", "tag"],
-    alternative_names=(
-        metrics.name_prefix + "_response_time:total",
-    ),
 )
 
 response_ru_utime = metrics.register_counter(
     "response_ru_utime_seconds", labels=["method", "servlet", "tag"],
-    alternative_names=(
-        metrics.name_prefix + "_response_ru_utime:total",
-    ),
 )
 
 response_ru_stime = metrics.register_counter(
     "response_ru_stime_seconds", labels=["method", "servlet", "tag"],
-    alternative_names=(
-        metrics.name_prefix + "_response_ru_stime:total",
-    ),
 )
 
 response_db_txn_count = metrics.register_counter(
     "response_db_txn_count", labels=["method", "servlet", "tag"],
-    alternative_names=(
-        metrics.name_prefix + "_response_db_txn_count:total",
-    ),
 )
 
 # seconds spent waiting for db txns, excluding scheduling time, when processing
 # this request
 response_db_txn_duration = metrics.register_counter(
     "response_db_txn_duration_seconds", labels=["method", "servlet", "tag"],
-    alternative_names=(
-        metrics.name_prefix + "_response_db_txn_duration:total",
-    ),
 )
 
 # seconds spent waiting for a db connection, when processing this request

synapse/util/metrics.py

@@ -31,53 +31,28 @@ metrics = synapse.metrics.get_metrics_for(__name__)
 block_counter = metrics.register_counter(
     "block_count",
     labels=["block_name"],
-    alternative_names=(
-        # the following are all deprecated aliases for the same metric
-        metrics.name_prefix + x for x in (
-            "_block_timer:count",
-            "_block_ru_utime:count",
-            "_block_ru_stime:count",
-            "_block_db_txn_count:count",
-            "_block_db_txn_duration:count",
-        )
-    )
 )
 
 block_timer = metrics.register_counter(
     "block_time_seconds",
     labels=["block_name"],
-    alternative_names=(
-        metrics.name_prefix + "_block_timer:total",
-    ),
 )
 
 block_ru_utime = metrics.register_counter(
     "block_ru_utime_seconds", labels=["block_name"],
-    alternative_names=(
-        metrics.name_prefix + "_block_ru_utime:total",
-    ),
 )
 
 block_ru_stime = metrics.register_counter(
     "block_ru_stime_seconds", labels=["block_name"],
-    alternative_names=(
-        metrics.name_prefix + "_block_ru_stime:total",
-    ),
 )
 
 block_db_txn_count = metrics.register_counter(
     "block_db_txn_count", labels=["block_name"],
-    alternative_names=(
-        metrics.name_prefix + "_block_db_txn_count:total",
-    ),
 )
 
 # seconds spent waiting for db txns, excluding scheduling time, in this block
 block_db_txn_duration = metrics.register_counter(
     "block_db_txn_duration_seconds", labels=["block_name"],
-    alternative_names=(
-        metrics.name_prefix + "_block_db_txn_duration:total",
-    ),
 )
 
 # seconds spent waiting for a db connection, in this block