replacing portions

synapse/util/caches/__init__.py

@@ -13,29 +13,52 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import synapse.metrics
+from prometheus_client.core import GaugeMetricFamily, REGISTRY
+
 import os
 
 CACHE_SIZE_FACTOR = float(os.environ.get("SYNAPSE_CACHE_FACTOR", 0.5))
 
-metrics = synapse.metrics.get_metrics_for("synapse.util.caches")
-
 caches_by_name = {}
-# cache_counter = metrics.register_cache(
-#     "cache",
-#     lambda: {(name,): len(caches_by_name[name]) for name in caches_by_name.keys()},
-#     labels=["name"],
-# )
+collectors_by_name = {}
 
+def register_cache(name, cache_name, cache):
 
-def register_cache(name, cache):
-    caches_by_name[name] = cache
-    return metrics.register_cache(
-        "cache",
-        lambda: len(cache),
-        name,
-    )
+    # Check if the metric is already registered. Unregister it, if so.
+    metric_name = "synapse_util_caches_%s:%s" % (name, cache_name,)
+    if metric_name in collectors_by_name.keys():
+        REGISTRY.unregister(collectors_by_name[metric_name])
 
+    class CacheMetric(object):
+
+        hits = 0
+        misses = 0
+        evicted_size = 0
+
+        def inc_hits(self):
+            self.hits += 1
+
+        def inc_misses(self):
+            self.misses += 1
+
+        def inc_evictions(self, size=1):
+            self.evicted_size += size
+
+        def collect(self):
+            cache_size = len(cache)
+
+            gm = GaugeMetricFamily(metric_name, "", labels=["size", "hits", "misses", "total"])
+            gm.add_metric(["size"], cache_size)
+            gm.add_metric(["hits"], self.hits)
+            gm.add_metric(["misses"], self.misses)
+            gm.add_metric(["total"], self.hits + self.misses)
+            yield gm
+
+    metric = CacheMetric()
+    REGISTRY.register(metric)
+    caches_by_name[cache_name] = cache
+    collectors_by_name[metric_name] = metric
+    return metric
 
 KNOWN_KEYS = {
     key: key for key in

synapse/util/caches/descriptors.py

@@ -80,7 +80,7 @@ class Cache(object):
         self.name = name
         self.keylen = keylen
         self.thread = None
-        self.metrics = register_cache(name, self.cache)
+        self.metrics = register_cache("descriptor", name, self.cache)
 
     def _on_evicted(self, evicted_count):
         self.metrics.inc_evictions(evicted_count)

synapse/util/caches/dictionary_cache.py

@@ -55,7 +55,7 @@ class DictionaryCache(object):
             __slots__ = []
 
         self.sentinel = Sentinel()
-        self.metrics = register_cache(name, self.cache)
+        self.metrics = register_cache("dictionary", name, self.cache)
 
     def check_thread(self):
         expected_thread = self.thread

synapse/util/caches/expiringcache.py

@@ -52,12 +52,12 @@ class ExpiringCache(object):
 
         self._cache = OrderedDict()
 
-        self.metrics = register_cache(cache_name, self)
-
         self.iterable = iterable
 
         self._size_estimate = 0
 
+        self.metrics = register_cache("expiring", cache_name, self)
+
     def start(self):
         if not self._expiry_ms:
             # Don't bother starting the loop if things never expire

synapse/util/caches/response_cache.py

@@ -17,7 +17,7 @@ import logging
 from twisted.internet import defer
 
 from synapse.util.async import ObservableDeferred
-from synapse.util.caches import metrics as cache_metrics
+from synapse.util.caches import register_cache
 from synapse.util.logcontext import make_deferred_yieldable, run_in_background
 
 logger = logging.getLogger(__name__)
@@ -38,15 +38,16 @@ class ResponseCache(object):
         self.timeout_sec = timeout_ms / 1000.
 
         self._name = name
-        self._metrics = cache_metrics.register_cache(
-            "response_cache",
-            size_callback=lambda: self.size(),
-            cache_name=name,
+        self._metrics = register_cache(
+            "response_cache", name, self
         )
 
     def size(self):
         return len(self.pending_result_cache)
 
+    def __len__(self):
+        return self.size()
+
     def get(self, key):
         """Look up the given key.
 

synapse/util/caches/stream_change_cache.py

@@ -38,7 +38,7 @@ class StreamChangeCache(object):
         self._cache = sorteddict()
         self._earliest_known_stream_pos = current_stream_pos
         self.name = name
-        self.metrics = register_cache(self.name, self._cache)
+        self.metrics = register_cache("cache", self.name, self._cache)
 
         for entity, stream_pos in prefilled_cache.items():
             self.entity_has_changed(entity, stream_pos)

synapse/util/metrics.py

@@ -15,8 +15,8 @@
 
 from twisted.internet import defer
 
+from prometheus_client import Counter
 from synapse.util.logcontext import LoggingContext
-import synapse.metrics
 
 from functools import wraps
 import logging
@@ -24,66 +24,21 @@ import logging
 
 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"])
 
-# total number of times we have hit this block
-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_ru_utime = Counter("synapse_util_metrics_block_ru_utime_seconds", "", ["block_name"])
 
-block_timer = metrics.register_counter(
-    "block_time_seconds",
-    labels=["block_name"],
-    alternative_names=(
-        metrics.name_prefix + "_block_timer:total",
-    ),
-)
+block_ru_stime = Counter("synapse_util_metrics_block_ru_stime_seconds", "", ["block_name"])
 
-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",
-    ),
-)
+block_db_txn_count = Counter("synapse_util_metrics_block_db_txn_count", "", ["block_name"])
 
 # 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",
-    ),
-)
+block_db_txn_duration = Counter("synapse_util_metrics_block_db_txn_duration_seconds", "", ["block_name"])
 
 # seconds spent waiting for a db connection, in this block
-block_db_sched_duration = metrics.register_counter(
-    "block_db_sched_duration_seconds", labels=["block_name"],
-)
+block_db_sched_duration = Counter("synapse_util_metrics_block_db_sched_duration_seconds", "", ["block_name"])
 
 
 def measure_func(name):
@@ -132,8 +87,8 @@ class Measure(object):
 
         duration = self.clock.time_msec() - self.start
 
-        block_counter.inc(self.name)
-        block_timer.inc_by(duration, self.name)
+        block_counter.labels(self.name).inc()
+        block_timer.labels(self.name).inc(duration)
 
         context = LoggingContext.current_context()
 
@@ -150,19 +105,13 @@ class Measure(object):
 
         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_ms - self.db_txn_duration_ms) / 1000.,
-            self.name
-        )
-        block_db_sched_duration.inc_by(
-            (context.db_sched_duration_ms - self.db_sched_duration_ms) / 1000.,
-            self.name
-        )
+        block_ru_utime.labels(self.name).inc(ru_utime - self.ru_utime)
+        block_ru_stime.labels(self.name).inc(ru_stime - self.ru_stime)
+        block_db_txn_count.labels(self.name).inc(context.db_txn_count - self.db_txn_count)
+        block_db_txn_duration.labels(self.name).inc(
+            (context.db_txn_duration_ms - self.db_txn_duration_ms) / 1000.)
+        block_db_sched_duration.labels(self.name).inc(
+            (context.db_sched_duration_ms - self.db_sched_duration_ms) / 1000.)
 
         if self.created_context:
             self.start_context.__exit__(exc_type, exc_val, exc_tb)