diff --git a/changelog.d/3956.bugfix b/changelog.d/3956.bugfix new file mode 100644 index 000000000..b0828c9fc --- /dev/null +++ b/changelog.d/3956.bugfix @@ -0,0 +1 @@ +Fix exceptions from metrics handler \ No newline at end of file diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py index 7b065b195..f37d5bec0 100644 --- a/synapse/util/caches/__init__.py +++ b/synapse/util/caches/__init__.py @@ -13,6 +13,7 @@ # See the License for the specific language governing permissions and # limitations under the License. +import logging import os import six @@ -20,6 +21,8 @@ from six.moves import intern from prometheus_client.core import REGISTRY, Gauge, GaugeMetricFamily +logger = logging.getLogger(__name__) + CACHE_SIZE_FACTOR = float(os.environ.get("SYNAPSE_CACHE_FACTOR", 0.5)) @@ -76,16 +79,20 @@ def register_cache(cache_type, cache_name, cache): return [] def collect(self): - if cache_type == "response_cache": - response_cache_size.labels(cache_name).set(len(cache)) - response_cache_hits.labels(cache_name).set(self.hits) - response_cache_evicted.labels(cache_name).set(self.evicted_size) - response_cache_total.labels(cache_name).set(self.hits + self.misses) - else: - cache_size.labels(cache_name).set(len(cache)) - cache_hits.labels(cache_name).set(self.hits) - cache_evicted.labels(cache_name).set(self.evicted_size) - cache_total.labels(cache_name).set(self.hits + self.misses) + try: + if cache_type == "response_cache": + response_cache_size.labels(cache_name).set(len(cache)) + response_cache_hits.labels(cache_name).set(self.hits) + response_cache_evicted.labels(cache_name).set(self.evicted_size) + response_cache_total.labels(cache_name).set(self.hits + self.misses) + else: + cache_size.labels(cache_name).set(len(cache)) + cache_hits.labels(cache_name).set(self.hits) + cache_evicted.labels(cache_name).set(self.evicted_size) + cache_total.labels(cache_name).set(self.hits + self.misses) + except Exception as e: + logger.warn("Error calculating metrics for %s: %s", cache_name, e) + raise yield GaugeMetricFamily("__unused", "") diff --git a/synapse/util/caches/expiringcache.py b/synapse/util/caches/expiringcache.py index 921a9c5b2..9af4ec4aa 100644 --- a/synapse/util/caches/expiringcache.py +++ b/synapse/util/caches/expiringcache.py @@ -16,6 +16,8 @@ import logging from collections import OrderedDict +from six import itervalues + from synapse.metrics.background_process_metrics import run_as_background_process from synapse.util.caches import register_cache @@ -54,8 +56,6 @@ class ExpiringCache(object): self.iterable = iterable - self._size_estimate = 0 - self.metrics = register_cache("expiring", cache_name, self) if not self._expiry_ms: @@ -74,16 +74,11 @@ class ExpiringCache(object): now = self._clock.time_msec() self._cache[key] = _CacheEntry(now, value) - if self.iterable: - self._size_estimate += len(value) - # Evict if there are now too many items while self._max_len and len(self) > self._max_len: _key, value = self._cache.popitem(last=False) if self.iterable: - removed_len = len(value.value) - self.metrics.inc_evictions(removed_len) - self._size_estimate -= removed_len + self.metrics.inc_evictions(len(value.value)) else: self.metrics.inc_evictions() @@ -134,7 +129,9 @@ class ExpiringCache(object): for k in keys_to_delete: value = self._cache.pop(k) if self.iterable: - self._size_estimate -= len(value.value) + self.metrics.inc_evictions(len(value.value)) + else: + self.metrics.inc_evictions() logger.debug( "[%s] _prune_cache before: %d, after len: %d", @@ -143,7 +140,7 @@ class ExpiringCache(object): def __len__(self): if self.iterable: - return self._size_estimate + return sum(len(entry.value) for entry in itervalues(self._cache)) else: return len(self._cache)