[wip] Fixing memory leaks

Sources/DistributedActorsTestKit/Cluster/ClusteredActorSystemsXCTestCase.swift

@@ -165,6 +165,10 @@ open class ClusteredActorSystemsXCTestCase: XCTestCase {
             self.printAllCapturedLogs()
         }
 
+        self.lock.withLockVoid {
+            self._logCaptures = []
+        }
+
         for node in self._nodes {
             node.log.warning("======================== TEST TEAR DOWN: SHUTDOWN ========================")
             try! await node.shutdown().wait()
@@ -173,7 +177,6 @@ open class ClusteredActorSystemsXCTestCase: XCTestCase {
         self.lock.withLockVoid {
             self._nodes = []
             self._testKits = []
-            self._logCaptures = []
         }
 
         if self.inspectDetectActorLeaks {
@@ -187,7 +190,10 @@ open class ClusteredActorSystemsXCTestCase: XCTestCase {
     }
 
     public func testKit(_ system: ClusterSystem) -> ActorTestKit {
-        guard let idx = self._nodes.firstIndex(where: { s in s.cluster.node == system.cluster.node }) else {
+        let idx = self.lock.withLock {
+            self._nodes.firstIndex(where: { s in s.cluster.node == system.cluster.node })
+        }
+        guard let idx else {
             fatalError("Must only call with system that was spawned using `setUpNode()`, was: \(system)")
         }
 

Sources/DistributedActorsTestKit/LogCapture.swift

@@ -35,7 +35,10 @@ public final class LogCapture {
 
     public func logger(label: String) -> Logger {
         self.captureLabel = label
-        return Logger(label: "LogCapture(\(label))", LogCaptureLogHandler(label: label, self))
+        let handler = LogCaptureLogHandler(label: label, settings: self.settings) { [weak self] log in
+            self?.append(log)
+        }
+        return Logger(label: "LogCapture(\(label))", handler)
     }
 
     func append(_ log: CapturedLogMessage) {
@@ -197,17 +200,19 @@ public struct CapturedLogMessage {
 
 struct LogCaptureLogHandler: LogHandler {
     let label: String
-    let capture: LogCapture
+    let settings: LogCapture.Settings
+    let append: (CapturedLogMessage) -> Void
 
-    init(label: String, _ capture: LogCapture) {
+    init(label: String, settings: LogCapture.Settings, append: @escaping (CapturedLogMessage) -> Void) {
         self.label = label
-        self.capture = capture
+        self.settings = settings
+        self.append = append
     }
 
     public func log(level: Logger.Level, message: Logger.Message, metadata: Logger.Metadata?, file: String, function: String, line: UInt) {
         let actorPath = self.metadata["actor/path"].map { "\($0)" } ?? ""
 
-        guard self.capture.settings.filterActorPaths.contains(where: { path in
+        guard self.settings.filterActorPaths.contains(where: { path in
             if (path == "") { // TODO(swift): rdar://98691039 String.starts(with:) has a bug when given an empty string, so we have to avoid it
                 return true
             }
@@ -216,13 +221,13 @@ struct LogCaptureLogHandler: LogHandler {
         }) else {
             return // ignore this actor's logs, it was filtered out
         }
-        guard !self.capture.settings.excludeActorPaths.contains(actorPath) else {
+        guard !self.settings.excludeActorPaths.contains(actorPath) else {
             return // actor was excluded explicitly
         }
-        guard self.capture.settings.grep.isEmpty || self.capture.settings.grep.contains(where: { "\(message)".contains($0) }) else {
+        guard self.settings.grep.isEmpty || self.settings.grep.contains(where: { "\(message)".contains($0) }) else {
             return // log was included explicitly
         }
-        guard !self.capture.settings.excludeGrep.contains(where: { "\(message)".contains($0) }) else {
+        guard !self.settings.excludeGrep.contains(where: { "\(message)".contains($0) }) else {
             return // log was excluded explicitly
         }
 
@@ -231,7 +236,7 @@ struct LogCaptureLogHandler: LogHandler {
         _metadata.merge(metadata ?? [:], uniquingKeysWith: { _, r in r })
         _metadata["label"] = "\(self.label)"
 
-        self.capture.append(CapturedLogMessage(date: date, level: level, message: message, metadata: _metadata, file: file, function: function, line: line))
+        self.append(CapturedLogMessage(date: date, level: level, message: message, metadata: _metadata, file: file, function: function, line: line))
     }
 
     public subscript(metadataKey metadataKey: String) -> Logger.Metadata.Value? {
@@ -247,7 +252,7 @@ struct LogCaptureLogHandler: LogHandler {
 
     public var logLevel: Logger.Level {
         get {
-            self.capture.settings.minimumLogLevel
+            self.settings.minimumLogLevel
         }
         set {
             // ignore, we always collect all logs

Sources/DistributedActorsTestKit/TestProbes.swift

@@ -853,8 +853,8 @@ private distributed actor _TestProbeInternal: LifecycleWatch {
         private var onEnqueue: ((Item) -> Void)?
 
         init() {
-            self.items = AsyncStream { continuation in
-                self.onEnqueue = { item in
+            self.items = AsyncStream { [weak self] continuation in
+                self?.onEnqueue = { item in
                     continuation.yield(item)
                 }
             }

Sources/DistributedCluster/ActorRefProvider.swift

@@ -50,13 +50,13 @@ internal struct RemoteActorRefProvider: _ActorRefProvider {
     private let localNode: Cluster.Node
     private let localProvider: LocalActorRefProvider
 
-    let cluster: ClusterShell
+    weak var cluster: ClusterShell?
     // TODO: should cache perhaps also associations to inject them eagerly to actor refs?
 
     // TODO: restructure it somehow, perhaps we dont need the full abstraction like this
     init(
         settings: ClusterSystemSettings,
-        cluster: ClusterShell,
+        cluster: ClusterShell?,
         localProvider: LocalActorRefProvider
     ) {
         precondition(settings.enabled, "Remote actor provider should only be used when clustering is enabled")

Sources/DistributedCluster/Cluster/ClusterControl.swift

@@ -29,7 +29,11 @@ public struct ClusterControl {
     ///
     /// This sequence begins with a snapshot of the current cluster state and continues with events representing changes
     /// since the snapshot.
-    public let events: ClusterEventStream
+    public var events: ClusterEventStream {
+        self._events
+    }
+
+    internal var _events: ClusterEventStream
 
     /// Offers a snapshot of membership, which may be used to perform ad-hoc tests against the membership.
     /// Note that this view may be immediately outdated after checking if, if e.g. a membership change is just being processed.
@@ -67,14 +71,14 @@ public struct ClusterControl {
         }
     }
 
-    private let cluster: ClusterShell?
+    private weak var cluster: ClusterShell?
     internal let ref: ClusterShell.Ref
 
     init(_ settings: ClusterSystemSettings, cluster: ClusterShell?, clusterRef: ClusterShell.Ref, eventStream: ClusterEventStream) {
         self.settings = settings
         self.cluster = cluster
         self.ref = clusterRef
-        self.events = eventStream
+        self._events = eventStream
 
         var initialMembership: Cluster.Membership = .empty
         _ = initialMembership.join(settings.bindNode)

Sources/DistributedCluster/Cluster/ClusterEventStream.swift

@@ -23,7 +23,7 @@ import Logging
 public struct ClusterEventStream: AsyncSequence {
     public typealias Element = Cluster.Event
 
-    private let actor: ClusterEventStreamActor?
+    private var actor: ClusterEventStreamActor?
 
     internal init(_ system: ClusterSystem, customName: String? = nil) {
         let props = ClusterEventStreamActor.props(customName: customName)
@@ -37,6 +37,10 @@ public struct ClusterEventStream: AsyncSequence {
         self.actor = nil
     }
 
+    internal mutating func clean() {
+        self.actor = nil
+    }
+
     nonisolated func subscribe(_ ref: _ActorRef<Cluster.Event>, file: String = #filePath, line: UInt = #line) {
         Task {
             await self._subscribe(ref, file: file, line: line)

Sources/DistributedCluster/Cluster/ClusterShell.swift

@@ -120,7 +120,7 @@ internal class ClusterShell {
 
         // 2) Ensure the failure detector knows about this node
         Task {
-            await self._swimShell.whenLocal { __secretlyKnownToBeLocal in // TODO(distributed): rename once https://github.com/apple/swift/pull/42098 is implemented
+            await self._swimShell?.whenLocal { __secretlyKnownToBeLocal in // TODO(distributed): rename once https://github.com/apple/swift/pull/42098 is implemented
                 __secretlyKnownToBeLocal.monitor(node: associated.handshake.remoteNode)
             }
         }
@@ -154,7 +154,7 @@ internal class ClusterShell {
         // it's gossip will also propagate the information through the cluster
         traceLog_Remote(system.cluster.node, "Finish terminate association [\(remoteNode)]: Notifying SWIM, .confirmDead")
         system.log.warning("Confirm .dead to underlying SWIM, node: \(reflecting: remoteNode)")
-        self._swimShell.confirmDead(node: remoteNode)
+        self._swimShell?.confirmDead(node: remoteNode)
 
         // it is important that we first check the contains; as otherwise we'd re-add a .down member for what was already removed (!)
         if state.membership.contains(remoteNode) {
@@ -244,9 +244,9 @@ internal class ClusterShell {
         return pool
     }
 
-    internal private(set) var _swimShell: SWIMActor!
+    internal private(set) var _swimShell: SWIMActor?
 
-    private var clusterEvents: ClusterEventStream!
+    private var clusterEvents: ClusterEventStream?
 
     // ==== ------------------------------------------------------------------------------------------------------------
     // MARK: Cluster Shell, reference used for issuing commands to the cluster
@@ -364,6 +364,11 @@ internal class ClusterShell {
         .init()
             .supervision(strategy: .escalate) // always escalate failures, if this actor fails we're in big trouble -> terminate the system
             ._asWellKnown
+
+    deinit {
+        self.clusterEvents?.clean()
+        self._swimShell?.clean()
+    }
 }
 
 // ==== ----------------------------------------------------------------------------------------------------------------
@@ -429,10 +434,14 @@ extension ClusterShell {
                     }
                 )
 
+                guard let events = self.clusterEvents else {
+                    throw ClusterSystemError(.shuttingDown(""))
+                }
+
                 var state = ClusterShellState(
                     settings: self.settings,
                     channel: chan,
-                    events: self.clusterEvents,
+                    events: events,
                     gossiperControl: gossiperControl,
                     log: context.log
                 )
@@ -452,7 +461,7 @@ extension ClusterShell {
     }
 
     private func publish(_ event: Cluster.Event) {
-        self.publish(event, to: self.clusterEvents)
+        self.clusterEvents.map { self.publish(event, to: $0) }
     }
 
     private func publish(_ event: Cluster.Event, to eventStream: ClusterEventStream) {
@@ -633,7 +642,7 @@ extension ClusterShell {
 
             Task { [eventsToPublish] in
                 for event in eventsToPublish {
-                    await self.clusterEvents.publish(event)
+                    await self.clusterEvents?.publish(event)
                 }
             }
 
@@ -663,7 +672,7 @@ extension ClusterShell {
 
     func tryConfirmDeadToSWIM(_ context: _ActorContext<Message>, _ state: ClusterShellState, change: Cluster.MembershipChange) {
         if change.status.isAtLeast(.down) {
-            self._swimShell.confirmDead(node: change.member.node)
+            self._swimShell?.confirmDead(node: change.member.node)
         }
     }
 
@@ -1271,7 +1280,7 @@ extension ClusterShell {
         }
 
         // whenever we down a node we must ensure to confirm it to swim, so it won't keep monitoring it forever needlessly
-        self._swimShell.confirmDead(node: memberToDown.node)
+        self._swimShell?.confirmDead(node: memberToDown.node)
 
         if memberToDown.node == state.selfNode {
             // ==== ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Sources/DistributedCluster/Cluster/DistributedNodeDeathWatcher.swift

@@ -54,20 +54,20 @@ internal actor DistributedNodeDeathWatcher {
         // initialized
 
         let events = actorSystem.cluster.events
-        self.eventListenerTask = Task {
+        self.eventListenerTask = Task { [weak self] in
             for try await event in events {
                 switch event {
                 case .membershipChange(let change):
-                    self.membershipChanged(change)
+                    await self?.membershipChanged(change)
                 case .snapshot(let membership):
                     let diff = Cluster.Membership._diff(from: .empty, to: membership)
                     for change in diff.changes {
-                        self.membershipChanged(change)
+                        await self?.membershipChanged(change)
                     }
                 case .leadershipChange, .reachabilityChange:
                     break // ignore those, they don't affect downing
                 case ._PLEASE_DO_NOT_EXHAUSTIVELY_MATCH_THIS_ENUM_NEW_CASES_MIGHT_BE_ADDED_IN_THE_FUTURE:
-                    self.log.error("Received Cluster.Event [\(event)]. This should not happen, please file an issue.")
+                    self?.log.error("Received Cluster.Event [\(event)]. This should not happen, please file an issue.")
                 }
             }
         }

Sources/DistributedCluster/Cluster/Downing/DowningStrategy.swift

@@ -92,17 +92,17 @@ internal distributed actor DowningStrategyShell {
     init(_ strategy: DowningStrategy, system: ActorSystem) async {
         self.strategy = strategy
         self.actorSystem = system
+        let events = system.cluster.events
         self.eventsListeningTask = Task { [weak self] in
-            try await self?.whenLocal { myself in
-                for await event in system.cluster.events {
+            for await event in events {
+                try await self?.whenLocal { myself in
                     try myself.receiveClusterEvent(event)
                 }
             }
         }
     }
 
     deinit {
-        print("\(Self.self) DEINIT")
         self.eventsListeningTask?.cancel()
 //        self.eventsListeningTask = nil
     }

Sources/DistributedCluster/Cluster/Reception/OperationLogDistributedReceptionist.swift

@@ -259,20 +259,22 @@ public distributed actor OpLogDistributedReceptionist: DistributedReceptionist,
             "\(Self.self) expects to be on well known path: /system/receptionist, but was: \(self.id.fullDescription)"
         ) // TODO(distributed): remove when we remove paths entirely
 
-        self.eventsListeningTask = Task { [weak self, system] in
-            try await self?.whenLocal { __secretlyKnownToBeLocal in
-                for try await event in system.cluster.events {
-                    __secretlyKnownToBeLocal.onClusterEvent(event: event)
+        let events = system.cluster.events
+        self.eventsListeningTask = Task { [weak self] in
+            for try await event in events {
+                await self?.whenLocal { myself in
+                    myself.onClusterEvent(event: event)
                 }
             }
         }
 
+        let interval = system.settings.receptionist.ackPullReplicationIntervalSlow
         // === timers ------------------
         // periodically gossip to other receptionists with the last seqNr we've seen,
         // and if it happens to be outdated by then this will cause a push from that node.
         self.slowACKReplicationTimerTask = Task { [weak self] in
-            await self?.whenLocal { myself in
-                for await _ in AsyncTimerSequence.repeating(every: system.settings.receptionist.ackPullReplicationIntervalSlow, clock: .continuous) {
+            for await _ in AsyncTimerSequence.repeating(every: interval, clock: .continuous) {
+                await self?.whenLocal { myself in
                     myself.periodicAckTick()
                 }
             }