Merge branch 'matrix-org:main' into presence-optimization

2024-05-20 06:13:48 +02:00 · 2024-03-04 16:25:41 -10:00 · 2024-03-04 16:25:41 -10:00 · b436d43133
parent 4872d21d72 09f15a3d3f
commit b436d43133
17 changed files with 180 additions and 78 deletions
--- a/.gitignore
+++ b/.gitignore
@ -77,4 +77,7 @@ media_store/
 build

 # golang workspaces
-go.work*
+go.work*
+
+# helm chart
+helm/dendrite/charts/
--- a/docs/FAQ.md
+++ b/docs/FAQ.md
@ -117,6 +117,7 @@ The list of files that need to be stored is:
 - matrix-key.pem
 - dendrite.yaml
 - the postgres or sqlite DB
+- the jetstream directory
 - the media store
 - the search index (although this can be regenerated)

--- a/docs/Gemfile.lock
+++ b/docs/Gemfile.lock
@ -231,9 +231,9 @@ GEM
      jekyll-seo-tag (~> 2.1)
    minitest (5.17.0)
    multipart-post (2.1.1)
-    nokogiri (1.14.3-arm64-darwin)
+    nokogiri (1.16.2-arm64-darwin)
      racc (~> 1.4)
-    nokogiri (1.14.3-x86_64-linux)
+    nokogiri (1.16.2-x86_64-linux)
      racc (~> 1.4)
    octokit (4.22.0)
      faraday (>= 0.9)
@ -241,7 +241,7 @@ GEM
    pathutil (0.16.2)
      forwardable-extended (~> 2.6)
    public_suffix (4.0.7)
-    racc (1.6.2)
+    racc (1.7.3)
    rb-fsevent (0.11.1)
    rb-inotify (0.10.1)
      ffi (~> 1.0)
--- a/federationapi/federationapi.go
+++ b/federationapi/federationapi.go
@ -113,10 +113,7 @@ func NewInternalAPI(
 		_ = federationDB.RemoveAllServersFromBlacklist()
 	}

-	stats := statistics.NewStatistics(
-		federationDB,
-		cfg.FederationMaxRetries+1,
-		cfg.P2PFederationRetriesUntilAssumedOffline+1)
+	stats := statistics.NewStatistics(federationDB, cfg.FederationMaxRetries+1, cfg.P2PFederationRetriesUntilAssumedOffline+1, cfg.EnableRelays)

 	js, nats := natsInstance.Prepare(processContext, &cfg.Matrix.JetStream)

--- a/federationapi/internal/federationclient_test.go
+++ b/federationapi/internal/federationclient_test.go
@ -61,7 +61,7 @@ func TestFederationClientQueryKeys(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -92,7 +92,7 @@ func TestFederationClientQueryKeysBlacklisted(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -122,7 +122,7 @@ func TestFederationClientQueryKeysFailure(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{shouldFail: true}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -152,7 +152,7 @@ func TestFederationClientClaimKeys(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -183,7 +183,7 @@ func TestFederationClientClaimKeysBlacklisted(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
--- a/federationapi/internal/perform_test.go
+++ b/federationapi/internal/perform_test.go
@ -66,7 +66,7 @@ func TestPerformWakeupServers(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, true)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -112,7 +112,7 @@ func TestQueryRelayServers(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -153,7 +153,7 @@ func TestRemoveRelayServers(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -193,7 +193,7 @@ func TestPerformDirectoryLookup(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
@ -232,7 +232,7 @@ func TestPerformDirectoryLookupRelaying(t *testing.T) {
 		},
 	}
 	fedClient := &testFedClient{}
-	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(testDB, FailuresUntilBlacklist, FailuresUntilAssumedOffline, true)
 	queues := queue.NewOutgoingQueues(
 		testDB, process.NewProcessContext(),
 		false,
--- a/federationapi/queue/queue_test.go
+++ b/federationapi/queue/queue_test.go
@ -117,7 +117,7 @@ func testSetup(failuresUntilBlacklist uint32, failuresUntilAssumedOffline uint32
 		txRelayCount:         *atomic.NewUint32(0),
 	}

-	stats := statistics.NewStatistics(db, failuresUntilBlacklist, failuresUntilAssumedOffline)
+	stats := statistics.NewStatistics(db, failuresUntilBlacklist, failuresUntilAssumedOffline, false)
 	signingInfo := []*fclient.SigningIdentity{
 		{
 			KeyID:      "ed21019:auto",
--- a/federationapi/statistics/statistics.go
+++ b/federationapi/statistics/statistics.go
@ -34,12 +34,15 @@ type Statistics struct {
 	// mark the destination as offline. At this point we should attempt
 	// to send messages to the user's async relay servers if we know them.
 	FailuresUntilAssumedOffline uint32
+
+	enableRelays bool
 }

 func NewStatistics(
 	db storage.Database,
 	failuresUntilBlacklist uint32,
 	failuresUntilAssumedOffline uint32,
+	enableRelays bool,
 ) Statistics {
 	return Statistics{
 		DB:                          db,
@ -47,6 +50,7 @@ func NewStatistics(
 		FailuresUntilAssumedOffline: failuresUntilAssumedOffline,
 		backoffTimers:               make(map[spec.ServerName]*time.Timer),
 		servers:                     make(map[spec.ServerName]*ServerStatistics),
+		enableRelays:                enableRelays,
 	}
 }

@ -73,6 +77,13 @@ func (s *Statistics) ForServer(serverName spec.ServerName) *ServerStatistics {
 		} else {
 			server.blacklisted.Store(blacklisted)
 		}
+
+		// Don't bother hitting the database 2 additional times
+		// if we don't want to use relays.
+		if !s.enableRelays {
+			return server
+		}
+
 		assumedOffline, err := s.DB.IsServerAssumedOffline(context.Background(), serverName)
 		if err != nil {
 			logrus.WithError(err).Errorf("Failed to get assumed offline entry %q", serverName)
--- a/federationapi/statistics/statistics_test.go
+++ b/federationapi/statistics/statistics_test.go
@ -16,7 +16,7 @@ const (
 )

 func TestBackoff(t *testing.T) {
-	stats := NewStatistics(nil, FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := NewStatistics(nil, FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	server := ServerStatistics{
 		statistics: &stats,
 		serverName: "test.com",
@ -106,7 +106,7 @@ func TestBackoff(t *testing.T) {
 }

 func TestRelayServersListing(t *testing.T) {
-	stats := NewStatistics(test.NewInMemoryFederationDatabase(), FailuresUntilBlacklist, FailuresUntilAssumedOffline)
+	stats := NewStatistics(test.NewInMemoryFederationDatabase(), FailuresUntilBlacklist, FailuresUntilAssumedOffline, false)
 	server := ServerStatistics{statistics: &stats}
 	server.AddRelayServers([]spec.ServerName{"relay1", "relay1", "relay2"})
 	relayServers := server.KnownRelayServers()
--- a/helm/dendrite/Chart.yaml
+++ b/helm/dendrite/Chart.yaml
@ -1,9 +1,10 @@
 apiVersion: v2
 name: dendrite
-version: "0.13.7"
+version: "0.14.0"
 appVersion: "0.13.6"
 description: Dendrite Matrix Homeserver
 type: application
+icon: https://avatars.githubusercontent.com/u/8418310?s=48&v=4
 keywords:
  - matrix
  - chat
@ -13,7 +14,7 @@ home: https://github.com/matrix-org/dendrite
 sources:
  - https://github.com/matrix-org/dendrite
 dependencies:
- name: postgresql
-  version: 12.1.7
-  repository: https://charts.bitnami.com/bitnami
-  condition: postgresql.enabled
+  - name: postgresql
+    version: 14.2.3
+    repository: https://charts.bitnami.com/bitnami
+    condition: postgresql.enabled
--- a/helm/dendrite/README.md
+++ b/helm/dendrite/README.md
@ -1,7 +1,7 @@

 # dendrite

-![Version: 0.13.7](https://img.shields.io/badge/Version-0.13.7-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: 0.13.6](https://img.shields.io/badge/AppVersion-0.13.6-informational?style=flat-square)
+![Version: 0.14.0](https://img.shields.io/badge/Version-0.14.0-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: 0.13.6](https://img.shields.io/badge/AppVersion-0.13.6-informational?style=flat-square)
 Dendrite Matrix Homeserver

 Status: **NOT PRODUCTION READY**
@ -37,7 +37,7 @@ Create a folder `appservices` and place your configurations in there.  The confi

 | Repository | Name | Version |
 |------------|------|---------|
-| https://charts.bitnami.com/bitnami | postgresql | 12.1.7 |
+| https://charts.bitnami.com/bitnami | postgresql | 14.2.3 |
 ## Values

 | Key | Type | Default | Description |
@ -48,16 +48,19 @@ Create a folder `appservices` and place your configurations in there.  The confi
 | signing_key.create | bool | `true` | Create a new signing key, if not exists |
 | signing_key.existingSecret | string | `""` | Use an existing secret |
 | resources | object | sets some sane default values | Default resource requests/limits. |
-| persistence.jetstream | object | `{"capacity":"1Gi","existingClaim":""}` | The storage class to use for volume claims. Used unless specified at the specific component. Defaults to the cluster default storage class. # If defined, storageClassName: <storageClass> # If set to "-", storageClassName: "", which disables dynamic provisioning # If undefined (the default) or set to null, no storageClassName spec is #   set, choosing the default provisioner.  (gp2 on AWS, standard on #   GKE, AWS & OpenStack) # storageClass: "" |
+| persistence.storageClass | string | `nil` | The storage class to use for volume claims. Used unless specified at the specific component. Defaults to the cluster default storage class. If defined, storageClassName: <storageClass> If set to "-", storageClassName: "", which disables dynamic provisioning If undefined (the default) or set to null, no storageClassName spec is   set, choosing the default provisioner.  (gp2 on AWS, standard on   GKE, AWS & OpenStack)  |
 | persistence.jetstream.existingClaim | string | `""` | Use an existing volume claim for jetstream |
 | persistence.jetstream.capacity | string | `"1Gi"` | PVC Storage Request for the jetstream volume |
+| persistence.jetstream.storageClass | string | `nil` | The storage class to use for volume claims. Defaults to persistence.storageClass If defined, storageClassName: <storageClass> If set to "-", storageClassName: "", which disables dynamic provisioning If undefined (the default) or set to null, no storageClassName spec is   set, choosing the default provisioner.  (gp2 on AWS, standard on   GKE, AWS & OpenStack) |
 | persistence.media.existingClaim | string | `""` | Use an existing volume claim for media files |
 | persistence.media.capacity | string | `"1Gi"` | PVC Storage Request for the media volume |
+| persistence.media.storageClass | string | `nil` | The storage class to use for volume claims. Defaults to persistence.storageClass If defined, storageClassName: <storageClass> If set to "-", storageClassName: "", which disables dynamic provisioning If undefined (the default) or set to null, no storageClassName spec is   set, choosing the default provisioner.  (gp2 on AWS, standard on   GKE, AWS & OpenStack) |
 | persistence.search.existingClaim | string | `""` | Use an existing volume claim for the fulltext search index |
 | persistence.search.capacity | string | `"1Gi"` | PVC Storage Request for the search volume |
+| persistence.search.storageClass | string | `nil` | The storage class to use for volume claims. Defaults to persistence.storageClass If defined, storageClassName: <storageClass> If set to "-", storageClassName: "", which disables dynamic provisioning If undefined (the default) or set to null, no storageClassName spec is   set, choosing the default provisioner.  (gp2 on AWS, standard on   GKE, AWS & OpenStack) |
 | extraVolumes | list | `[]` | Add additional volumes to the Dendrite Pod |
 | extraVolumeMounts | list | `[]` | Configure additional mount points volumes in the Dendrite Pod |
-| strategy.type | string | `"RollingUpdate"` | Strategy to use for rolling updates (e.g. Recreate, RollingUpdate) If you are using ReadWriteOnce volumes, you should probably use Recreate |
+| strategy.type | string | `"Recreate"` | Strategy to use for rolling updates (e.g. Recreate, RollingUpdate) If you are using ReadWriteOnce volumes, you should probably use Recreate |
 | strategy.rollingUpdate.maxUnavailable | string | `"25%"` | Maximum number of pods that can be unavailable during the update process |
 | strategy.rollingUpdate.maxSurge | string | `"25%"` | Maximum number of pods that can be scheduled above the desired number of pods |
 | dendrite_config.version | int | `2` |  |
@ -139,7 +142,7 @@ Create a folder `appservices` and place your configurations in there.  The confi
 | dendrite_config.logging | list | `[{"level":"info","type":"std"}]` | Default logging configuration |
 | postgresql.enabled | bool | See value.yaml | Enable and configure postgres as the database for dendrite. |
 | postgresql.image.repository | string | `"bitnami/postgresql"` |  |
-| postgresql.image.tag | string | `"15.1.0"` |  |
+| postgresql.image.tag | string | `"16.2.0"` |  |
 | postgresql.auth.username | string | `"dendrite"` |  |
 | postgresql.auth.password | string | `"changeme"` |  |
 | postgresql.auth.database | string | `"dendrite"` |  |
@ -186,3 +189,5 @@ grafana:
 ```
 PS: The label `release=kube-prometheus-stack` is setup with the helmchart of the Prometheus Operator. For Grafana Dashboards it may be necessary to enable scanning in the correct namespaces (or ALL), enabled by `sidecar.dashboards.searchNamespace` in [Helmchart of grafana](https://artifacthub.io/packages/helm/grafana/grafana) (which is part of PrometheusOperator, so `grafana.sidecar.dashboards.searchNamespace`)

+----------------------------------------------
+Autogenerated from chart metadata using [helm-docs v1.13.0](https://github.com/norwoodj/helm-docs/releases/v1.13.0)
--- a/helm/dendrite/templates/ingress.yaml
+++ b/helm/dendrite/templates/ingress.yaml
@ -4,6 +4,7 @@
 {{- $wellKnownServerHost := default $serverNameHost (regexFind "^(\\[.+\\])?[^:]*" .Values.dendrite_config.global.well_known_server_name) -}}
 {{- $wellKnownClientHost := default $serverNameHost (regexFind "//(\\[.+\\])?[^:/]*" .Values.dendrite_config.global.well_known_client_name | trimAll "/") -}}
 {{- $allHosts := list $serverNameHost $wellKnownServerHost $wellKnownClientHost | uniq -}}
+
 {{- if semverCompare ">=1.19-0" .Capabilities.KubeVersion.GitVersion -}}
 apiVersion: networking.k8s.io/v1
 {{- else if semverCompare ">=1.14-0" .Capabilities.KubeVersion.GitVersion -}}
@ -56,7 +57,7 @@ spec:
              service:
                name: {{ $fullName }}
                port:
-                  name: http
+                  number: {{ $.Values.service.port }}
              {{- else }}
              serviceName: {{ $fullName }}
              servicePort: http
@ -72,7 +73,7 @@ spec:
              service:
                name: {{ $fullName }}
                port:
-                  name: http
+                  number: {{ $.Values.service.port }}
              {{- else }}
              serviceName: {{ $fullName }}
              servicePort: http
@ -88,7 +89,7 @@ spec:
              service:
                name: {{ $fullName }}
                port:
-                  name: http
+                  number: {{ $.Values.service.port }}
              {{- else }}
              serviceName: {{ $fullName }}
              servicePort: http
@ -105,7 +106,7 @@ spec:
              service:
                name: {{ $fullName }}
                port:
-                  name: http
+                  number: {{ $.Values.service.port }}
              {{- else }}
              serviceName: {{ $fullName }}
              servicePort: http
--- a/helm/dendrite/templates/service.yaml
+++ b/helm/dendrite/templates/service.yaml
@ -14,4 +14,4 @@ spec:
    - name: http
      protocol: TCP
      port: {{ .Values.service.port }}
-      targetPort: 8008
+      targetPort: http
--- a/helm/dendrite/values.yaml
+++ b/helm/dendrite/values.yaml
@ -26,13 +26,13 @@ persistence:
  # -- The storage class to use for volume claims.
  # Used unless specified at the specific component.
  # Defaults to the cluster default storage class.
-  ## If defined, storageClassName: <storageClass>
-  ## If set to "-", storageClassName: "", which disables dynamic provisioning
-  ## If undefined (the default) or set to null, no storageClassName spec is
-  ##   set, choosing the default provisioner.  (gp2 on AWS, standard on
-  ##   GKE, AWS & OpenStack)
-  ##
-  # storageClass: ""
+  # If defined, storageClassName: <storageClass>
+  # If set to "-", storageClassName: "", which disables dynamic provisioning
+  # If undefined (the default) or set to null, no storageClassName spec is
+  #   set, choosing the default provisioner.  (gp2 on AWS, standard on
+  #   GKE, AWS & OpenStack)
+  #
+  storageClass:
  jetstream:
    # -- Use an existing volume claim for jetstream
    existingClaim: ""
@ -40,13 +40,12 @@ persistence:
    capacity: "1Gi"
    # -- The storage class to use for volume claims.
    # Defaults to persistence.storageClass
-    ## If defined, storageClassName: <storageClass>
-    ## If set to "-", storageClassName: "", which disables dynamic provisioning
-    ## If undefined (the default) or set to null, no storageClassName spec is
-    ##   set, choosing the default provisioner.  (gp2 on AWS, standard on
-    ##   GKE, AWS & OpenStack)
-    ##
-    # storageClass: ""
+    # If defined, storageClassName: <storageClass>
+    # If set to "-", storageClassName: "", which disables dynamic provisioning
+    # If undefined (the default) or set to null, no storageClassName spec is
+    #   set, choosing the default provisioner.  (gp2 on AWS, standard on
+    #   GKE, AWS & OpenStack)
+    storageClass:
  media:
    # -- Use an existing volume claim for media files
    existingClaim: ""
@ -54,13 +53,12 @@ persistence:
    capacity: "1Gi"
    # -- The storage class to use for volume claims.
    # Defaults to persistence.storageClass
-    ## If defined, storageClassName: <storageClass>
-    ## If set to "-", storageClassName: "", which disables dynamic provisioning
-    ## If undefined (the default) or set to null, no storageClassName spec is
-    ##   set, choosing the default provisioner.  (gp2 on AWS, standard on
-    ##   GKE, AWS & OpenStack)
-    ##
-    # storageClass: ""
+    # If defined, storageClassName: <storageClass>
+    # If set to "-", storageClassName: "", which disables dynamic provisioning
+    # If undefined (the default) or set to null, no storageClassName spec is
+    #   set, choosing the default provisioner.  (gp2 on AWS, standard on
+    #   GKE, AWS & OpenStack)
+    storageClass:
  search:
    # -- Use an existing volume claim for the fulltext search index
    existingClaim: ""
@ -68,13 +66,12 @@ persistence:
    capacity: "1Gi"
    # -- The storage class to use for volume claims.
    # Defaults to persistence.storageClass
-    ## If defined, storageClassName: <storageClass>
-    ## If set to "-", storageClassName: "", which disables dynamic provisioning
-    ## If undefined (the default) or set to null, no storageClassName spec is
-    ##   set, choosing the default provisioner.  (gp2 on AWS, standard on
-    ##   GKE, AWS & OpenStack)
-    ##
-    # storageClass: ""
+    # If defined, storageClassName: <storageClass>
+    # If set to "-", storageClassName: "", which disables dynamic provisioning
+    # If undefined (the default) or set to null, no storageClassName spec is
+    #   set, choosing the default provisioner.  (gp2 on AWS, standard on
+    #   GKE, AWS & OpenStack)
+    storageClass:

 # -- Add additional volumes to the Dendrite Pod
 extraVolumes: []
@ -92,7 +89,7 @@ extraVolumeMounts: []
 strategy:
  # -- Strategy to use for rolling updates (e.g. Recreate, RollingUpdate)
  # If you are using ReadWriteOnce volumes, you should probably use Recreate
-  type: RollingUpdate
+  type: Recreate
  rollingUpdate:
    # -- Maximum number of pods that can be unavailable during the update process
    maxUnavailable: 25%
@ -378,7 +375,7 @@ postgresql:
  enabled: false
  image:
    repository: bitnami/postgresql
-    tag: "15.1.0"
+    tag: "16.2.0"
  auth:
    username: dendrite
    password: changeme
--- a/roomserver/acls/acls.go
+++ b/roomserver/acls/acls.go
@ -41,15 +41,21 @@ type ServerACLDatabase interface {
 }

 type ServerACLs struct {
-	acls      map[string]*serverACL // room ID -> ACL
-	aclsMutex sync.RWMutex          // protects the above
+	acls               map[string]*serverACL      // room ID -> ACL
+	aclsMutex          sync.RWMutex               // protects the above
+	aclRegexCache      map[string]**regexp.Regexp // Cache from "serverName" -> pointer to a regex
+	aclRegexCacheMutex sync.RWMutex               // protects the above
 }

 func NewServerACLs(db ServerACLDatabase) *ServerACLs {
 	ctx := context.TODO()
 	acls := &ServerACLs{
 		acls: make(map[string]*serverACL),
+		// Be generous when creating the cache, as in reality
+		// there are hundreds of servers in an ACL.
+		aclRegexCache: make(map[string]**regexp.Regexp, 100),
 	}
+
 	// Look up all of the rooms that the current state server knows about.
 	rooms, err := db.GetKnownRooms(ctx)
 	if err != nil {
@ -67,6 +73,7 @@ func NewServerACLs(db ServerACLDatabase) *ServerACLs {
 	for _, event := range events {
 		acls.OnServerACLUpdate(event)
 	}
+
 	return acls
 }

@ -78,8 +85,8 @@ type ServerACL struct {

 type serverACL struct {
 	ServerACL
-	allowedRegexes []*regexp.Regexp
-	deniedRegexes  []*regexp.Regexp
+	allowedRegexes []**regexp.Regexp
+	deniedRegexes  []**regexp.Regexp
 }

 func compileACLRegex(orig string) (*regexp.Regexp, error) {
@ -89,6 +96,25 @@ func compileACLRegex(orig string) (*regexp.Regexp, error) {
 	return regexp.Compile(escaped)
 }

+// cachedCompileACLRegex is a wrapper around compileACLRegex with added caching
+func (s *ServerACLs) cachedCompileACLRegex(orig string) (**regexp.Regexp, error) {
+	s.aclRegexCacheMutex.RLock()
+	re, ok := s.aclRegexCache[orig]
+	if ok {
+		s.aclRegexCacheMutex.RUnlock()
+		return re, nil
+	}
+	s.aclRegexCacheMutex.RUnlock()
+	compiled, err := compileACLRegex(orig)
+	if err != nil {
+		return nil, err
+	}
+	s.aclRegexCacheMutex.Lock()
+	defer s.aclRegexCacheMutex.Unlock()
+	s.aclRegexCache[orig] = &compiled
+	return &compiled, nil
+}
+
 func (s *ServerACLs) OnServerACLUpdate(strippedEvent tables.StrippedEvent) {
 	acls := &serverACL{}
 	if err := json.Unmarshal([]byte(strippedEvent.ContentValue), &acls.ServerACL); err != nil {
@ -100,14 +126,14 @@ func (s *ServerACLs) OnServerACLUpdate(strippedEvent tables.StrippedEvent) {
 	// special characters and then replace * and ? with their regex counterparts.
 	// https://matrix.org/docs/spec/client_server/r0.6.1#m-room-server-acl
 	for _, orig := range acls.Allowed {
-		if expr, err := compileACLRegex(orig); err != nil {
+		if expr, err := s.cachedCompileACLRegex(orig); err != nil {
 			logrus.WithError(err).Errorf("Failed to compile allowed regex")
 		} else {
 			acls.allowedRegexes = append(acls.allowedRegexes, expr)
 		}
 	}
 	for _, orig := range acls.Denied {
-		if expr, err := compileACLRegex(orig); err != nil {
+		if expr, err := s.cachedCompileACLRegex(orig); err != nil {
 			logrus.WithError(err).Errorf("Failed to compile denied regex")
 		} else {
 			acls.deniedRegexes = append(acls.deniedRegexes, expr)
@ -118,6 +144,11 @@ func (s *ServerACLs) OnServerACLUpdate(strippedEvent tables.StrippedEvent) {
 		"num_allowed":       len(acls.allowedRegexes),
 		"num_denied":        len(acls.deniedRegexes),
 	}).Debugf("Updating server ACLs for %q", strippedEvent.RoomID)
+
+	// Clear out Denied and Allowed, now that we have the compiled regexes.
+	// They are not needed anymore from this point on.
+	acls.Denied = nil
+	acls.Allowed = nil
 	s.aclsMutex.Lock()
 	defer s.aclsMutex.Unlock()
 	s.acls[strippedEvent.RoomID] = acls
@ -150,14 +181,14 @@ func (s *ServerACLs) IsServerBannedFromRoom(serverName spec.ServerName, roomID s
 	// Check if the hostname matches one of the denied regexes. If it does then
 	// the server is banned from the room.
 	for _, expr := range acls.deniedRegexes {
-		if expr.MatchString(string(serverName)) {
+		if (*expr).MatchString(string(serverName)) {
 			return true
 		}
 	}
 	// Check if the hostname matches one of the allowed regexes. If it does then
 	// the server is NOT banned from the room.
 	for _, expr := range acls.allowedRegexes {
-		if expr.MatchString(string(serverName)) {
+		if (*expr).MatchString(string(serverName)) {
 			return false
 		}
 	}
--- a/roomserver/acls/acls_test.go
+++ b/roomserver/acls/acls_test.go
@ -15,8 +15,14 @@
 package acls

 import (
+	"context"
 	"regexp"
 	"testing"
+
+	"github.com/matrix-org/dendrite/roomserver/storage/tables"
+	"github.com/matrix-org/gomatrixserverlib"
+	"github.com/matrix-org/gomatrixserverlib/spec"
+	"github.com/stretchr/testify/assert"
 )

 func TestOpenACLsWithBlacklist(t *testing.T) {
@ -38,8 +44,8 @@ func TestOpenACLsWithBlacklist(t *testing.T) {
 		ServerACL: ServerACL{
 			AllowIPLiterals: true,
 		},
-		allowedRegexes: []*regexp.Regexp{allowRegex},
-		deniedRegexes:  []*regexp.Regexp{denyRegex},
+		allowedRegexes: []**regexp.Regexp{&allowRegex},
+		deniedRegexes:  []**regexp.Regexp{&denyRegex},
 	}

 	if acls.IsServerBannedFromRoom("1.2.3.4", roomID) {
@ -77,8 +83,8 @@ func TestDefaultACLsWithWhitelist(t *testing.T) {
 		ServerACL: ServerACL{
 			AllowIPLiterals: false,
 		},
-		allowedRegexes: []*regexp.Regexp{allowRegex},
-		deniedRegexes:  []*regexp.Regexp{},
+		allowedRegexes: []**regexp.Regexp{&allowRegex},
+		deniedRegexes:  []**regexp.Regexp{},
 	}

 	if !acls.IsServerBannedFromRoom("1.2.3.4", roomID) {
@ -103,3 +109,45 @@ func TestDefaultACLsWithWhitelist(t *testing.T) {
 		t.Fatal("Expected qux.com:4567 to be allowed but wasn't")
 	}
 }
+
+var (
+	content1 = `{"allow":["*"],"allow_ip_literals":false,"deny":["hello.world", "*.hello.world"]}`
+)
+
+type dummyACLDB struct{}
+
+func (d dummyACLDB) GetKnownRooms(ctx context.Context) ([]string, error) {
+	return []string{"1", "2"}, nil
+}
+
+func (d dummyACLDB) GetBulkStateContent(ctx context.Context, roomIDs []string, tuples []gomatrixserverlib.StateKeyTuple, allowWildcards bool) ([]tables.StrippedEvent, error) {
+	return []tables.StrippedEvent{
+		{
+			RoomID:       "1",
+			ContentValue: content1,
+		},
+		{
+			RoomID:       "2",
+			ContentValue: content1,
+		},
+	}, nil
+}
+
+func TestCachedRegex(t *testing.T) {
+	db := dummyACLDB{}
+	wantBannedServer := spec.ServerName("hello.world")
+
+	acls := NewServerACLs(db)
+
+	// Check that hello.world is banned in room 1
+	banned := acls.IsServerBannedFromRoom(wantBannedServer, "1")
+	assert.True(t, banned)
+
+	// Check that hello.world is banned in room 2
+	banned = acls.IsServerBannedFromRoom(wantBannedServer, "2")
+	assert.True(t, banned)
+
+	// Check that matrix.hello.world is banned in room 2
+	banned = acls.IsServerBannedFromRoom("matrix."+wantBannedServer, "2")
+	assert.True(t, banned)
+}
--- a/setup/config/config_federationapi.go
+++ b/setup/config/config_federationapi.go
@ -18,6 +18,13 @@ type FederationAPI struct {
 	// The default value is 16 if not specified, which is circa 18 hours.
 	FederationMaxRetries uint32 `yaml:"send_max_retries"`

+	// P2P Feature: Whether relaying to specific nodes should be enabled.
+	// Defaults to false.
+	// Note: Enabling relays introduces a huge startup delay, if you are not using
+	// relays and have many servers to re-hydrate on start. Only enable this
+	// if you are using relays!
+	EnableRelays bool `yaml:"enable_relays"`
+
 	// P2P Feature: How many consecutive failures that we should tolerate when
 	// sending federation requests to a specific server until we should assume they
 	// are offline. If we assume they are offline then we will attempt to send