From 3ee17585cd095e590096683395cfb9a017eac15e Mon Sep 17 00:00:00 2001
From: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
Date: Fri, 16 Oct 2020 15:51:57 +0100
Subject: Make LruCache register its own metrics (#8561)

rather than have everything that instantiates an LruCache manage metrics
separately, have LruCache do it itself.
---
 synapse/util/caches/lrucache.py | 46 +++++++++++++++++++++++++++++++----------
 1 file changed, 35 insertions(+), 11 deletions(-)

(limited to 'synapse/util/caches/lrucache.py')

diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py
index 33eae2b7c4..e4804f79e0 100644
--- a/synapse/util/caches/lrucache.py
+++ b/synapse/util/caches/lrucache.py
@@ -18,6 +18,7 @@ from functools import wraps
 from typing import Callable, Optional, Type, Union
 
 from synapse.config import cache as cache_config
+from synapse.util.caches import CacheMetric, register_cache
 from synapse.util.caches.treecache import TreeCache
 
 
@@ -43,27 +44,29 @@ class _Node:
 
 class LruCache:
     """
-    Least-recently-used cache.
+    Least-recently-used cache, supporting prometheus metrics and invalidation callbacks.
+
     Supports del_multi only if cache_type=TreeCache
     If cache_type=TreeCache, all keys must be tuples.
-
-    Can also set callbacks on objects when getting/setting which are fired
-    when that key gets invalidated/evicted.
     """
 
     def __init__(
         self,
         max_size: int,
+        cache_name: Optional[str] = None,
         keylen: int = 1,
         cache_type: Type[Union[dict, TreeCache]] = dict,
         size_callback: Optional[Callable] = None,
-        evicted_callback: Optional[Callable] = None,
+        metrics_collection_callback: Optional[Callable[[], None]] = None,
         apply_cache_factor_from_config: bool = True,
     ):
         """
         Args:
             max_size: The maximum amount of entries the cache can hold
 
+            cache_name: The name of this cache, for the prometheus metrics. If unset,
+                no metrics will be reported on this cache.
+
             keylen: The length of the tuple used as the cache key. Ignored unless
                 cache_type is `TreeCache`.
 
@@ -73,9 +76,13 @@ class LruCache:
 
             size_callback (func(V) -> int | None):
 
-            evicted_callback (func(int)|None):
-                if not None, called on eviction with the size of the evicted
-                entry
+            metrics_collection_callback:
+                metrics collection callback. This is called early in the metrics
+                collection process, before any of the metrics registered with the
+                prometheus Registry are collected, so can be used to update any dynamic
+                metrics.
+
+                Ignored if cache_name is None.
 
             apply_cache_factor_from_config (bool): If true, `max_size` will be
                 multiplied by a cache factor derived from the homeserver config
@@ -94,6 +101,19 @@ class LruCache:
         else:
             self.max_size = int(max_size)
 
+        if cache_name is not None:
+            metrics = register_cache(
+                "lru_cache",
+                cache_name,
+                self,
+                collect_callback=metrics_collection_callback,
+            )  # type: Optional[CacheMetric]
+        else:
+            metrics = None
+
+        # this is exposed for access from outside this class
+        self.metrics = metrics
+
         list_root = _Node(None, None, None, None)
         list_root.next_node = list_root
         list_root.prev_node = list_root
@@ -105,8 +125,8 @@ class LruCache:
                 todelete = list_root.prev_node
                 evicted_len = delete_node(todelete)
                 cache.pop(todelete.key, None)
-                if evicted_callback:
-                    evicted_callback(evicted_len)
+                if metrics:
+                    metrics.inc_evictions(evicted_len)
 
         def synchronized(f):
             @wraps(f)
@@ -169,13 +189,17 @@ class LruCache:
             return deleted_len
 
         @synchronized
-        def cache_get(key, default=None, callbacks=[]):
+        def cache_get(key, default=None, callbacks=[], update_metrics=True):
             node = cache.get(key, None)
             if node is not None:
                 move_node_to_front(node)
                 node.callbacks.update(callbacks)
+                if update_metrics and metrics:
+                    metrics.inc_hits()
                 return node.value
             else:
+                if update_metrics and metrics:
+                    metrics.inc_misses()
                 return default
 
         @synchronized
-- 
cgit 1.5.1


From 0ec0bc3886bd72bdf2f64d455a7d777f4573a4f1 Mon Sep 17 00:00:00 2001
From: Richard van der Hoff <richard@matrix.org>
Date: Fri, 16 Oct 2020 15:56:39 +0100
Subject: type annotations for LruCache

---
 synapse/api/auth.py                     |  4 +-
 synapse/push/push_rule_evaluator.py     | 16 ++++----
 synapse/util/caches/deferred_cache.py   |  5 ++-
 synapse/util/caches/dictionary_cache.py | 22 ++++++----
 synapse/util/caches/lrucache.py         | 73 +++++++++++++++++++++++++++------
 5 files changed, 89 insertions(+), 31 deletions(-)

(limited to 'synapse/util/caches/lrucache.py')

diff --git a/synapse/api/auth.py b/synapse/api/auth.py
index eb6f418b13..bff87fabde 100644
--- a/synapse/api/auth.py
+++ b/synapse/api/auth.py
@@ -69,7 +69,9 @@ class Auth:
         self.store = hs.get_datastore()
         self.state = hs.get_state_handler()
 
-        self.token_cache = LruCache(10000, "token_cache")
+        self.token_cache = LruCache(
+            10000, "token_cache"
+        )  # type: LruCache[str, Tuple[str, bool]]
 
         self._auth_blocking = AuthBlocking(self.hs)
 
diff --git a/synapse/push/push_rule_evaluator.py b/synapse/push/push_rule_evaluator.py
index 4c95b149c5..854ffd625e 100644
--- a/synapse/push/push_rule_evaluator.py
+++ b/synapse/push/push_rule_evaluator.py
@@ -16,7 +16,7 @@
 
 import logging
 import re
-from typing import Any, Dict, List, Optional, Pattern, Union
+from typing import Any, Dict, List, Optional, Pattern, Tuple, Union
 
 from synapse.events import EventBase
 from synapse.types import UserID
@@ -173,19 +173,21 @@ class PushRuleEvaluatorForEvent:
         # Similar to _glob_matches, but do not treat display_name as a glob.
         r = regex_cache.get((display_name, False, True), None)
         if not r:
-            r = re.escape(display_name)
-            r = _re_word_boundary(r)
-            r = re.compile(r, flags=re.IGNORECASE)
+            r1 = re.escape(display_name)
+            r1 = _re_word_boundary(r1)
+            r = re.compile(r1, flags=re.IGNORECASE)
             regex_cache[(display_name, False, True)] = r
 
-        return r.search(body)
+        return bool(r.search(body))
 
     def _get_value(self, dotted_key: str) -> Optional[str]:
         return self._value_cache.get(dotted_key, None)
 
 
 # Caches (string, is_glob, word_boundary) -> regex for push. See _glob_matches
-regex_cache = LruCache(50000, "regex_push_cache")
+regex_cache = LruCache(
+    50000, "regex_push_cache"
+)  # type: LruCache[Tuple[str, bool, bool],Pattern]
 
 
 def _glob_matches(glob: str, value: str, word_boundary: bool = False) -> bool:
@@ -203,7 +205,7 @@ def _glob_matches(glob: str, value: str, word_boundary: bool = False) -> bool:
         if not r:
             r = _glob_to_re(glob, word_boundary)
             regex_cache[(glob, True, word_boundary)] = r
-        return r.search(value)
+        return bool(r.search(value))
     except re.error:
         logger.warning("Failed to parse glob to regex: %r", glob)
         return False
diff --git a/synapse/util/caches/deferred_cache.py b/synapse/util/caches/deferred_cache.py
index 91fdc8142d..4026e1f8fa 100644
--- a/synapse/util/caches/deferred_cache.py
+++ b/synapse/util/caches/deferred_cache.py
@@ -98,7 +98,7 @@ class DeferredCache(Generic[KT, VT]):
             size_callback=(lambda d: len(d)) if iterable else None,
             metrics_collection_callback=metrics_cb,
             apply_cache_factor_from_config=apply_cache_factor_from_config,
-        )
+        )  # type: LruCache[KT, VT]
 
         self.thread = None  # type: Optional[threading.Thread]
 
@@ -240,11 +240,12 @@ class DeferredCache(Generic[KT, VT]):
         self.check_thread()
         if not isinstance(key, tuple):
             raise TypeError("The cache key must be a tuple not %r" % (type(key),))
+        key = cast(KT, key)
         self.cache.del_multi(key)
 
         # if we have a pending lookup for this key, remove it from the
         # _pending_deferred_cache, as above
-        entry_dict = self._pending_deferred_cache.pop(cast(KT, key), None)
+        entry_dict = self._pending_deferred_cache.pop(key, None)
         if entry_dict is not None:
             for entry in iterate_tree_cache_entry(entry_dict):
                 entry.invalidate()
diff --git a/synapse/util/caches/dictionary_cache.py b/synapse/util/caches/dictionary_cache.py
index 8b426c005b..588d2d49f2 100644
--- a/synapse/util/caches/dictionary_cache.py
+++ b/synapse/util/caches/dictionary_cache.py
@@ -12,10 +12,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
+import enum
 import logging
 import threading
 from collections import namedtuple
+from typing import Any
 
 from synapse.util.caches.lrucache import LruCache
 
@@ -38,23 +39,26 @@ class DictionaryEntry(namedtuple("DictionaryEntry", ("full", "known_absent", "va
         return len(self.value)
 
 
+class _Sentinel(enum.Enum):
+    # defining a sentinel in this way allows mypy to correctly handle the
+    # type of a dictionary lookup.
+    sentinel = object()
+
+
 class DictionaryCache:
     """Caches key -> dictionary lookups, supporting caching partial dicts, i.e.
     fetching a subset of dictionary keys for a particular key.
     """
 
     def __init__(self, name, max_entries=1000):
-        self.cache = LruCache(max_size=max_entries, cache_name=name, size_callback=len)
+        self.cache = LruCache(
+            max_size=max_entries, cache_name=name, size_callback=len
+        )  # type: LruCache[Any, DictionaryEntry]
 
         self.name = name
         self.sequence = 0
         self.thread = None
 
-        class Sentinel:
-            __slots__ = []
-
-        self.sentinel = Sentinel()
-
     def check_thread(self):
         expected_thread = self.thread
         if expected_thread is None:
@@ -76,8 +80,8 @@ class DictionaryCache:
         Returns:
             DictionaryEntry
         """
-        entry = self.cache.get(key, self.sentinel)
-        if entry is not self.sentinel:
+        entry = self.cache.get(key, _Sentinel.sentinel)
+        if entry is not _Sentinel.sentinel:
             if dict_keys is None:
                 return DictionaryEntry(
                     entry.full, entry.known_absent, dict(entry.value)
diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py
index e4804f79e0..0eed53d3f4 100644
--- a/synapse/util/caches/lrucache.py
+++ b/synapse/util/caches/lrucache.py
@@ -15,12 +15,30 @@
 
 import threading
 from functools import wraps
-from typing import Callable, Optional, Type, Union
+from typing import (
+    Any,
+    Callable,
+    Generic,
+    Iterable,
+    Optional,
+    Type,
+    TypeVar,
+    Union,
+    cast,
+    overload,
+)
+
+from typing_extensions import Literal
 
 from synapse.config import cache as cache_config
 from synapse.util.caches import CacheMetric, register_cache
 from synapse.util.caches.treecache import TreeCache
 
+T = TypeVar("T")
+FT = TypeVar("FT", bound=Callable[..., Any])
+KT = TypeVar("KT")
+VT = TypeVar("VT")
+
 
 def enumerate_leaves(node, depth):
     if depth == 0:
@@ -42,7 +60,7 @@ class _Node:
         self.callbacks = callbacks
 
 
-class LruCache:
+class LruCache(Generic[KT, VT]):
     """
     Least-recently-used cache, supporting prometheus metrics and invalidation callbacks.
 
@@ -128,13 +146,13 @@ class LruCache:
                 if metrics:
                     metrics.inc_evictions(evicted_len)
 
-        def synchronized(f):
+        def synchronized(f: FT) -> FT:
             @wraps(f)
             def inner(*args, **kwargs):
                 with lock:
                     return f(*args, **kwargs)
 
-            return inner
+            return cast(FT, inner)
 
         cached_cache_len = [0]
         if size_callback is not None:
@@ -188,8 +206,31 @@ class LruCache:
             node.callbacks.clear()
             return deleted_len
 
+        @overload
+        def cache_get(
+            key: KT,
+            default: Literal[None] = None,
+            callbacks: Iterable[Callable[[], None]] = ...,
+            update_metrics: bool = ...,
+        ) -> Optional[VT]:
+            ...
+
+        @overload
+        def cache_get(
+            key: KT,
+            default: T,
+            callbacks: Iterable[Callable[[], None]] = ...,
+            update_metrics: bool = ...,
+        ) -> Union[T, VT]:
+            ...
+
         @synchronized
-        def cache_get(key, default=None, callbacks=[], update_metrics=True):
+        def cache_get(
+            key: KT,
+            default=None,
+            callbacks: Iterable[Callable[[], None]] = [],
+            update_metrics: bool = True,
+        ):
             node = cache.get(key, None)
             if node is not None:
                 move_node_to_front(node)
@@ -203,7 +244,7 @@ class LruCache:
                 return default
 
         @synchronized
-        def cache_set(key, value, callbacks=[]):
+        def cache_set(key: KT, value: VT, callbacks: Iterable[Callable[[], None]] = []):
             node = cache.get(key, None)
             if node is not None:
                 # We sometimes store large objects, e.g. dicts, which cause
@@ -232,7 +273,7 @@ class LruCache:
             evict()
 
         @synchronized
-        def cache_set_default(key, value):
+        def cache_set_default(key: KT, value: VT) -> VT:
             node = cache.get(key, None)
             if node is not None:
                 return node.value
@@ -241,8 +282,16 @@ class LruCache:
                 evict()
                 return value
 
+        @overload
+        def cache_pop(key: KT, default: Literal[None] = None) -> Union[None, VT]:
+            ...
+
+        @overload
+        def cache_pop(key: KT, default: T) -> Union[T, VT]:
+            ...
+
         @synchronized
-        def cache_pop(key, default=None):
+        def cache_pop(key: KT, default=None):
             node = cache.get(key, None)
             if node:
                 delete_node(node)
@@ -252,18 +301,18 @@ class LruCache:
                 return default
 
         @synchronized
-        def cache_del_multi(key):
+        def cache_del_multi(key: KT) -> None:
             """
             This will only work if constructed with cache_type=TreeCache
             """
             popped = cache.pop(key)
             if popped is None:
                 return
-            for leaf in enumerate_leaves(popped, keylen - len(key)):
+            for leaf in enumerate_leaves(popped, keylen - len(cast(tuple, key))):
                 delete_node(leaf)
 
         @synchronized
-        def cache_clear():
+        def cache_clear() -> None:
             list_root.next_node = list_root
             list_root.prev_node = list_root
             for node in cache.values():
@@ -274,7 +323,7 @@ class LruCache:
                 cached_cache_len[0] = 0
 
         @synchronized
-        def cache_contains(key):
+        def cache_contains(key: KT) -> bool:
             return key in cache
 
         self.sentinel = object()
-- 
cgit 1.5.1


From 995cc615a01bb11b70dbf8fdd0eb7f8b3d1fdc1e Mon Sep 17 00:00:00 2001
From: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
Date: Fri, 16 Oct 2020 16:14:42 +0100
Subject: Apply suggestions from code review

Co-authored-by: Patrick Cloke <clokep@users.noreply.github.com>
---
 synapse/push/push_rule_evaluator.py | 2 +-
 synapse/util/caches/lrucache.py     | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

(limited to 'synapse/util/caches/lrucache.py')

diff --git a/synapse/push/push_rule_evaluator.py b/synapse/push/push_rule_evaluator.py
index 854ffd625e..2ce9e444ab 100644
--- a/synapse/push/push_rule_evaluator.py
+++ b/synapse/push/push_rule_evaluator.py
@@ -187,7 +187,7 @@ class PushRuleEvaluatorForEvent:
 # Caches (string, is_glob, word_boundary) -> regex for push. See _glob_matches
 regex_cache = LruCache(
     50000, "regex_push_cache"
-)  # type: LruCache[Tuple[str, bool, bool],Pattern]
+)  # type: LruCache[Tuple[str, bool, bool], Pattern]
 
 
 def _glob_matches(glob: str, value: str, word_boundary: bool = False) -> bool:
diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py
index 0eed53d3f4..1a2c2d4c0b 100644
--- a/synapse/util/caches/lrucache.py
+++ b/synapse/util/caches/lrucache.py
@@ -283,7 +283,7 @@ class LruCache(Generic[KT, VT]):
                 return value
 
         @overload
-        def cache_pop(key: KT, default: Literal[None] = None) -> Union[None, VT]:
+        def cache_pop(key: KT, default: Literal[None] = None) -> Optional[VT]:
             ...
 
         @overload
-- 
cgit 1.5.1


From 6d7b22041ddf5ecceaf230404ef00c4d0b432727 Mon Sep 17 00:00:00 2001
From: Richard van der Hoff <richard@matrix.org>
Date: Fri, 16 Oct 2020 16:21:43 +0100
Subject: review comments

---
 synapse/util/caches/lrucache.py | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

(limited to 'synapse/util/caches/lrucache.py')

diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py
index 1a2c2d4c0b..4e95dd9bf3 100644
--- a/synapse/util/caches/lrucache.py
+++ b/synapse/util/caches/lrucache.py
@@ -34,11 +34,16 @@ from synapse.config import cache as cache_config
 from synapse.util.caches import CacheMetric, register_cache
 from synapse.util.caches.treecache import TreeCache
 
-T = TypeVar("T")
+# Function type: the type used for invalidation callbacks
 FT = TypeVar("FT", bound=Callable[..., Any])
+
+# Key and Value type for the cache
 KT = TypeVar("KT")
 VT = TypeVar("VT")
 
+# a general type var, distinct from either KT or VT
+T = TypeVar("T")
+
 
 def enumerate_leaves(node, depth):
     if depth == 0:
@@ -227,7 +232,7 @@ class LruCache(Generic[KT, VT]):
         @synchronized
         def cache_get(
             key: KT,
-            default=None,
+            default: Optional[T] = None,
             callbacks: Iterable[Callable[[], None]] = [],
             update_metrics: bool = True,
         ):
@@ -291,7 +296,7 @@ class LruCache(Generic[KT, VT]):
             ...
 
         @synchronized
-        def cache_pop(key: KT, default=None):
+        def cache_pop(key: KT, default: Optional[T] = None):
             node = cache.get(key, None)
             if node:
                 delete_node(node)
-- 
cgit 1.5.1


From 97647b33c248f25571bae617365d95434e6a3d5f Mon Sep 17 00:00:00 2001
From: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
Date: Mon, 19 Oct 2020 12:20:29 +0100
Subject: Replace DeferredCache with LruCache where possible (#8563)

Most of these uses don't need a full-blown DeferredCache; LruCache is lighter and more appropriate.
---
 changelog.d/8563.misc                           |  1 +
 synapse/replication/slave/storage/client_ips.py | 10 +++++-----
 synapse/storage/_base.py                        | 12 +++++++-----
 synapse/storage/databases/main/client_ips.py    |  8 ++++----
 synapse/storage/databases/main/devices.py       |  8 ++++----
 synapse/storage/databases/main/events.py        |  4 +---
 synapse/storage/databases/main/events_worker.py | 11 +++++------
 synapse/util/caches/lrucache.py                 |  3 +++
 8 files changed, 30 insertions(+), 27 deletions(-)
 create mode 100644 changelog.d/8563.misc

(limited to 'synapse/util/caches/lrucache.py')

diff --git a/changelog.d/8563.misc b/changelog.d/8563.misc
new file mode 100644
index 0000000000..eeba8e5fee
--- /dev/null
+++ b/changelog.d/8563.misc
@@ -0,0 +1 @@
+Replace `DeferredCache` with the lighter-weight `LruCache` where possible.
diff --git a/synapse/replication/slave/storage/client_ips.py b/synapse/replication/slave/storage/client_ips.py
index 4b0ea0cc01..0f5b7adef7 100644
--- a/synapse/replication/slave/storage/client_ips.py
+++ b/synapse/replication/slave/storage/client_ips.py
@@ -15,7 +15,7 @@
 
 from synapse.storage.database import DatabasePool
 from synapse.storage.databases.main.client_ips import LAST_SEEN_GRANULARITY
-from synapse.util.caches.deferred_cache import DeferredCache
+from synapse.util.caches.lrucache import LruCache
 
 from ._base import BaseSlavedStore
 
@@ -24,9 +24,9 @@ class SlavedClientIpStore(BaseSlavedStore):
     def __init__(self, database: DatabasePool, db_conn, hs):
         super().__init__(database, db_conn, hs)
 
-        self.client_ip_last_seen = DeferredCache(
-            name="client_ip_last_seen", keylen=4, max_entries=50000
-        )  # type: DeferredCache[tuple, int]
+        self.client_ip_last_seen = LruCache(
+            cache_name="client_ip_last_seen", keylen=4, max_size=50000
+        )  # type: LruCache[tuple, int]
 
     async def insert_client_ip(self, user_id, access_token, ip, user_agent, device_id):
         now = int(self._clock.time_msec())
@@ -41,7 +41,7 @@ class SlavedClientIpStore(BaseSlavedStore):
         if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY:
             return
 
-        self.client_ip_last_seen.prefill(key, now)
+        self.client_ip_last_seen.set(key, now)
 
         self.hs.get_tcp_replication().send_user_ip(
             user_id, access_token, ip, user_agent, device_id, now
diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py
index ab49d227de..2b196ded1b 100644
--- a/synapse/storage/_base.py
+++ b/synapse/storage/_base.py
@@ -76,14 +76,16 @@ class SQLBaseStore(metaclass=ABCMeta):
         """
 
         try:
-            if key is None:
-                getattr(self, cache_name).invalidate_all()
-            else:
-                getattr(self, cache_name).invalidate(tuple(key))
+            cache = getattr(self, cache_name)
         except AttributeError:
             # We probably haven't pulled in the cache in this worker,
             # which is fine.
-            pass
+            return
+
+        if key is None:
+            cache.invalidate_all()
+        else:
+            cache.invalidate(tuple(key))
 
 
 def db_to_json(db_content):
diff --git a/synapse/storage/databases/main/client_ips.py b/synapse/storage/databases/main/client_ips.py
index 9e66e6648a..339bd691a4 100644
--- a/synapse/storage/databases/main/client_ips.py
+++ b/synapse/storage/databases/main/client_ips.py
@@ -19,7 +19,7 @@ from typing import Dict, Optional, Tuple
 from synapse.metrics.background_process_metrics import wrap_as_background_process
 from synapse.storage._base import SQLBaseStore
 from synapse.storage.database import DatabasePool, make_tuple_comparison_clause
-from synapse.util.caches.deferred_cache import DeferredCache
+from synapse.util.caches.lrucache import LruCache
 
 logger = logging.getLogger(__name__)
 
@@ -410,8 +410,8 @@ class ClientIpWorkerStore(ClientIpBackgroundUpdateStore):
 class ClientIpStore(ClientIpWorkerStore):
     def __init__(self, database: DatabasePool, db_conn, hs):
 
-        self.client_ip_last_seen = DeferredCache(
-            name="client_ip_last_seen", keylen=4, max_entries=50000
+        self.client_ip_last_seen = LruCache(
+            cache_name="client_ip_last_seen", keylen=4, max_size=50000
         )
 
         super().__init__(database, db_conn, hs)
@@ -442,7 +442,7 @@ class ClientIpStore(ClientIpWorkerStore):
         if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY:
             return
 
-        self.client_ip_last_seen.prefill(key, now)
+        self.client_ip_last_seen.set(key, now)
 
         self._batch_row_update[key] = (user_agent, device_id, now)
 
diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py
index e662a20d24..dfb4f87b8f 100644
--- a/synapse/storage/databases/main/devices.py
+++ b/synapse/storage/databases/main/devices.py
@@ -34,8 +34,8 @@ from synapse.storage.database import (
 )
 from synapse.types import Collection, JsonDict, get_verify_key_from_cross_signing_key
 from synapse.util import json_decoder, json_encoder
-from synapse.util.caches.deferred_cache import DeferredCache
 from synapse.util.caches.descriptors import cached, cachedList
+from synapse.util.caches.lrucache import LruCache
 from synapse.util.iterutils import batch_iter
 from synapse.util.stringutils import shortstr
 
@@ -1005,8 +1005,8 @@ class DeviceStore(DeviceWorkerStore, DeviceBackgroundUpdateStore):
 
         # Map of (user_id, device_id) -> bool. If there is an entry that implies
         # the device exists.
-        self.device_id_exists_cache = DeferredCache(
-            name="device_id_exists", keylen=2, max_entries=10000
+        self.device_id_exists_cache = LruCache(
+            cache_name="device_id_exists", keylen=2, max_size=10000
         )
 
     async def store_device(
@@ -1052,7 +1052,7 @@ class DeviceStore(DeviceWorkerStore, DeviceBackgroundUpdateStore):
                 )
                 if hidden:
                     raise StoreError(400, "The device ID is in use", Codes.FORBIDDEN)
-            self.device_id_exists_cache.prefill(key, True)
+            self.device_id_exists_cache.set(key, True)
             return inserted
         except StoreError:
             raise
diff --git a/synapse/storage/databases/main/events.py b/synapse/storage/databases/main/events.py
index ba3b1769b0..87808c1483 100644
--- a/synapse/storage/databases/main/events.py
+++ b/synapse/storage/databases/main/events.py
@@ -1051,9 +1051,7 @@ class PersistEventsStore:
 
         def prefill():
             for cache_entry in to_prefill:
-                self.store._get_event_cache.prefill(
-                    (cache_entry[0].event_id,), cache_entry
-                )
+                self.store._get_event_cache.set((cache_entry[0].event_id,), cache_entry)
 
         txn.call_after(prefill)
 
diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py
index 0ad9a19b3d..c342df2a8b 100644
--- a/synapse/storage/databases/main/events_worker.py
+++ b/synapse/storage/databases/main/events_worker.py
@@ -42,8 +42,8 @@ from synapse.storage.database import DatabasePool
 from synapse.storage.engines import PostgresEngine
 from synapse.storage.util.id_generators import MultiWriterIdGenerator, StreamIdGenerator
 from synapse.types import Collection, get_domain_from_id
-from synapse.util.caches.deferred_cache import DeferredCache
 from synapse.util.caches.descriptors import cached
+from synapse.util.caches.lrucache import LruCache
 from synapse.util.iterutils import batch_iter
 from synapse.util.metrics import Measure
 
@@ -146,11 +146,10 @@ class EventsWorkerStore(SQLBaseStore):
                 self._cleanup_old_transaction_ids,
             )
 
-        self._get_event_cache = DeferredCache(
-            "*getEvent*",
+        self._get_event_cache = LruCache(
+            cache_name="*getEvent*",
             keylen=3,
-            max_entries=hs.config.caches.event_cache_size,
-            apply_cache_factor_from_config=False,
+            max_size=hs.config.caches.event_cache_size,
         )
 
         self._event_fetch_lock = threading.Condition()
@@ -749,7 +748,7 @@ class EventsWorkerStore(SQLBaseStore):
                 event=original_ev, redacted_event=redacted_event
             )
 
-            self._get_event_cache.prefill((event_id,), cache_entry)
+            self._get_event_cache.set((event_id,), cache_entry)
             result_map[event_id] = cache_entry
 
         return result_map
diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py
index 4e95dd9bf3..3b471d8fd3 100644
--- a/synapse/util/caches/lrucache.py
+++ b/synapse/util/caches/lrucache.py
@@ -337,6 +337,9 @@ class LruCache(Generic[KT, VT]):
         self.set = cache_set
         self.setdefault = cache_set_default
         self.pop = cache_pop
+        # `invalidate` is exposed for consistency with DeferredCache, so that it can be
+        # invalidated by the cache invalidation replication stream.
+        self.invalidate = cache_pop
         if cache_type is TreeCache:
             self.del_multi = cache_del_multi
         self.len = synchronized(cache_len)
-- 
cgit 1.5.1


From 96e7d3c4a0feec6d19b873fd550bcfffd485d910 Mon Sep 17 00:00:00 2001
From: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
Date: Mon, 19 Oct 2020 21:13:50 +0100
Subject: Fix 'LruCache' object has no attribute '_on_resize' (#8591)

We need to make sure we are readu for the `set_cache_factor` callback.
---
 changelog.d/8591.misc           |  1 +
 synapse/util/caches/lrucache.py | 10 +++++++++-
 tests/util/test_lrucache.py     |  8 +++++++-
 3 files changed, 17 insertions(+), 2 deletions(-)
 create mode 100644 changelog.d/8591.misc

(limited to 'synapse/util/caches/lrucache.py')

diff --git a/changelog.d/8591.misc b/changelog.d/8591.misc
new file mode 100644
index 0000000000..8f16bc3e7e
--- /dev/null
+++ b/changelog.d/8591.misc
@@ -0,0 +1 @@
+ Move metric registration code down into `LruCache`.
diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py
index 3b471d8fd3..60bb6ff642 100644
--- a/synapse/util/caches/lrucache.py
+++ b/synapse/util/caches/lrucache.py
@@ -124,6 +124,10 @@ class LruCache(Generic[KT, VT]):
         else:
             self.max_size = int(max_size)
 
+        # register_cache might call our "set_cache_factor" callback; there's nothing to
+        # do yet when we get resized.
+        self._on_resize = None  # type: Optional[Callable[[],None]]
+
         if cache_name is not None:
             metrics = register_cache(
                 "lru_cache",
@@ -332,7 +336,10 @@ class LruCache(Generic[KT, VT]):
             return key in cache
 
         self.sentinel = object()
+
+        # make sure that we clear out any excess entries after we get resized.
         self._on_resize = evict
+
         self.get = cache_get
         self.set = cache_set
         self.setdefault = cache_set_default
@@ -383,6 +390,7 @@ class LruCache(Generic[KT, VT]):
         new_size = int(self._original_max_size * factor)
         if new_size != self.max_size:
             self.max_size = new_size
-            self._on_resize()
+            if self._on_resize:
+                self._on_resize()
             return True
         return False
diff --git a/tests/util/test_lrucache.py b/tests/util/test_lrucache.py
index f12834edab..a739a6aaaf 100644
--- a/tests/util/test_lrucache.py
+++ b/tests/util/test_lrucache.py
@@ -19,7 +19,8 @@ from mock import Mock
 from synapse.util.caches.lrucache import LruCache
 from synapse.util.caches.treecache import TreeCache
 
-from .. import unittest
+from tests import unittest
+from tests.unittest import override_config
 
 
 class LruCacheTestCase(unittest.HomeserverTestCase):
@@ -83,6 +84,11 @@ class LruCacheTestCase(unittest.HomeserverTestCase):
         cache.clear()
         self.assertEquals(len(cache), 0)
 
+    @override_config({"caches": {"per_cache_factors": {"mycache": 10}}})
+    def test_special_size(self):
+        cache = LruCache(10, "mycache")
+        self.assertEqual(cache.max_size, 100)
+
 
 class LruCacheCallbacksTestCase(unittest.HomeserverTestCase):
     def test_get(self):
-- 
cgit 1.5.1