Review fixes: offline-drop durability, gateway handoff retry, restore-log freshness, coalesced persist

Adversarial review of the storm/deadlock PR surfaced four issues: - Offline blackhole (must-fix): a deposit made while relays were down persisted its dedup key even though the drop only sat in the in-memory pending queue — app killed before reconnect meant the relaunch lost the drop but the persisted key blocked every re-deposit for 24h. The persisted snapshot now excludes keys still pending; they become durable only when flushPendingDrops actually publishes them. - Gateway handoff: seen-event IDs were consumed before the deliverToPeer handoff; a failed handoff (peer walked away) permanently dropped the event for a single-gateway island. deliverToPeer now reports whether the handoff was attempted, and a failure releases the seen slot so a relaunch or backlog redelivery retries. - Restore-path logs: central/peripheral-restore captures logged the init sentinel bgRemaining=∞. The cache is now seeded in init's main-thread branch and restore captures route through the sampler, which refreshes the cached budget before logging. - Persist cost: the dedup record was a full JSON encode + atomic write on the main actor per mutation (once per event during a backlog re-fetch). Writes now coalesce behind a 1s window, flushed immediately on background/terminate; panic wipe stays immediate. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-24 23:45:18 +00:00 · 2026-07-09 18:11:49 +02:00
parent d905452df2
commit 07b5ac31d8
5 changed files with 163 additions and 9 deletions
@@ -281,12 +281,18 @@ final class BLEService: NSObject {
        
        // Set up application state tracking (iOS only)
        #if os(iOS)
-        // Check initial state on main thread
+        // Check initial state on main thread. The background-budget cache is
+        // seeded here too: a background-restore launch captures Bluetooth
+        // status before any lifecycle notification fires, and the init-time
+        // sentinel would log a meaningless bgRemaining=∞ for exactly the
+        // wake window that matters.
        if Thread.isMainThread {
            isAppActive = UIApplication.shared.applicationState == .active
+            refreshCachedBackgroundTimeRemaining()
        } else {
            DispatchQueue.main.sync {
                isAppActive = UIApplication.shared.applicationState == .active
+                refreshCachedBackgroundTimeRemaining()
            }
        }
        
@@ -1729,7 +1735,10 @@ extension BLEService: CBCentralManagerDelegate {
            recentPeripheralCache.record(peripheral, peripheralID: identifier, at: Date())
        }

-        captureBluetoothStatus(context: "central-restore")
+        // Via the sampler (not a direct capture): it refreshes the cached
+        // background budget on main first, so the restore log shows the real
+        // wake window instead of the init sentinel.
+        logBluetoothStatus("central-restore")

        if central.state == .poweredOn {
            startScanning()
@@ -2488,7 +2497,8 @@ extension BLEService: CBPeripheralManagerDelegate {
            }
        }

-        captureBluetoothStatus(context: "peripheral-restore")
+        // Via the sampler for a fresh background budget (see central-restore).
+        logBluetoothStatus("peripheral-restore")

        if peripheral.state == .poweredOn && !peripheral.isAdvertising {
            peripheral.startAdvertising(buildAdvertisementData())
@@ -9,6 +9,11 @@
 import BitFoundation
 import BitLogger
 import Foundation
+#if os(iOS)
+import UIKit
+#elseif os(macOS)
+import AppKit
+#endif

 /// Courier delivery over the internet bridge: sealed courier envelopes are
 /// parked on relays as kind-1401 "drops" tagged with their rotating
@@ -48,6 +53,10 @@ final class BridgeCourierService: ObservableObject {
        /// Encoded envelope cap for a drop (16 KiB ciphertext + TLV slack).
        static let maxDropEnvelopeBytes = 20 * 1024
        static let maxTrackedIDs = 512
+        /// Coalescing window for dedup-record writes: a backlog re-fetch
+        /// mutates the seen set once per event, and each snapshot save is a
+        /// full JSON encode + atomic write on the main actor.
+        static let dedupPersistCoalesceSeconds: TimeInterval = 1.0
    }

    static let shared = BridgeCourierService()
@@ -70,7 +79,9 @@ final class BridgeCourierService: ObservableObject {
    /// Opens a drop addressed to us (tag verified inside).
    var openEnvelope: (@MainActor (CourierEnvelope) -> Void)?
    /// Hands a drop to a matching local peer as a directed courier packet.
-    var deliverToPeer: (@MainActor (CourierEnvelope, PeerID) -> Void)?
+    /// Returns false when the handoff could not even be attempted (peer no
+    /// longer reachable), so the drop event stays retryable.
+    var deliverToPeer: (@MainActor (CourierEnvelope, PeerID) -> Bool)?
    /// Held envelopes eligible for (re)publish, honoring the cooldown.
    var heldEnvelopes: (@MainActor (TimeInterval) -> [CourierEnvelope])?
    /// Timer injection for tests; nil arms a real `Task`.
@@ -98,6 +109,8 @@ final class BridgeCourierService: ObservableObject {
    private let now: () -> Date
    private let dedupStore: BridgeDropDedupStore

+    private var dedupPersistScheduled = false
+
    init(now: @escaping () -> Date = Date.init, dedupStore: BridgeDropDedupStore? = nil) {
        self.now = now
        self.dedupStore = dedupStore ?? BridgeDropDedupStore(persistsToDisk: !TestEnvironment.isRunningTests)
@@ -115,11 +128,44 @@ final class BridgeCourierService: ObservableObject {
            entries: snapshot.seenDropEventIDs,
            now: date
        )
+        // A coalesced dedup write scheduled just before a background kill
+        // would be lost; flush when the app backgrounds or terminates.
+        #if os(iOS)
+        let flushNotifications = [UIApplication.didEnterBackgroundNotification, UIApplication.willTerminateNotification]
+        #else
+        let flushNotifications = [NSApplication.willTerminateNotification]
+        #endif
+        for name in flushNotifications {
+            NotificationCenter.default.addObserver(forName: name, object: nil, queue: .main) { [weak self] _ in
+                MainActor.assumeIsolated { self?.flushDedupSnapshot() }
+            }
+        }
    }

+    /// Schedules a coalesced write of the dedup record (see
+    /// `Limits.dedupPersistCoalesceSeconds`); lifecycle notifications flush
+    /// any scheduled write before a background kill could drop it.
    private func persistDedup() {
+        guard !dedupPersistScheduled else { return }
+        dedupPersistScheduled = true
+        Task { @MainActor [weak self] in
+            try? await Task.sleep(nanoseconds: UInt64(Limits.dedupPersistCoalesceSeconds * 1_000_000_000))
+            guard let self else { return }
+            self.dedupPersistScheduled = false
+            self.flushDedupSnapshot()
+        }
+    }
+
+    /// Writes the dedup record now. Sender keys still sitting in the
+    /// in-memory `pendingDrops` queue are excluded: their drop is not durable
+    /// until it actually reaches a relay, and persisting the key early would
+    /// turn "app killed before relays connected" into a silent 24h blackhole
+    /// (the relaunch loses the queued drop but the persisted key blocks every
+    /// re-deposit). `flushPendingDrops` re-persists once they publish.
+    func flushDedupSnapshot() {
+        let pendingKeys = Set(pendingDrops.compactMap(\.dedupKey))
        dedupStore.save(BridgeDropDedupStore.Snapshot(
-            publishedDropKeys: publishedDropKeys.entries,
+            publishedDropKeys: publishedDropKeys.entries.filter { !pendingKeys.contains($0.key) },
            seenDropEventIDs: seenDropEventIDs.entries
        ))
    }
@@ -221,9 +267,11 @@ final class BridgeCourierService: ObservableObject {
            // retry sweep can attempt a fresh deposit.
            if let key = item.dedupKey {
                publishedDropKeys.remove(key)
-                persistDedup()
            }
        }
+        // Flushed keys just became durable (published, so no longer excluded
+        // as pending) or were released above; either way the record changed.
+        persistDedup()
    }

    // MARK: - Subscription (recipient + gateway watch)
@@ -328,7 +376,14 @@ final class BridgeCourierService: ObservableObject {
        }
        if let match = watchedPeerTags.first(where: { $0.tagsHex.contains(tagHex) }) {
            SecureLogger.info("📦🌉 Courier drop fetched for local peer \(match.peerID.id.prefix(8))…", category: .session)
-            deliverToPeer?(envelope, match.peerID)
+            if deliverToPeer?(envelope, match.peerID) != true {
+                // The best-effort handoff never left this device (the peer
+                // walked away between the relay fetch and the mesh send).
+                // Release the seen slot so a relaunch or backlog redelivery
+                // retries — a single-gateway island has no other carrier.
+                seenDropEventIDs.remove(event.id)
+                persistDedup()
+            }
        }
    }

@@ -97,6 +97,10 @@ final class BridgeDropDedupStore {

    func save(_ snapshot: Snapshot) {
        guard let fileURL else { return }
+        guard !(snapshot.publishedDropKeys.isEmpty && snapshot.seenDropEventIDs.isEmpty) else {
+            try? FileManager.default.removeItem(at: fileURL)
+            return
+        }
        do {
            try FileManager.default.createDirectory(
                at: fileURL.deletingLastPathComponent(),
@@ -567,7 +567,7 @@ private extension ChatViewModelBootstrapper {
            bleService?.openBridgedCourierEnvelope(envelope)
        }
        courier.deliverToPeer = { [weak bleService] envelope, peerID in
-            bleService?.deliverBridgedEnvelope(envelope, to: peerID)
+            bleService?.deliverBridgedEnvelope(envelope, to: peerID) ?? false
        }
        courier.heldEnvelopes = { cooldown in
            CourierStore.shared.envelopesForBridgePublish(cooldown: cooldown)
@@ -24,6 +24,7 @@ struct BridgeCourierServiceTests {
        var localPeers: [(peerID: PeerID, noiseKey: Data)] = []
        var held: [CourierEnvelope] = []
        var sealResult: CourierEnvelope?
+        var deliverResult = true

        private(set) var publishedEvents: [NostrEvent] = []
        private(set) var openedSubscriptions: [[String]] = []
@@ -49,7 +50,10 @@ struct BridgeCourierServiceTests {
                return self?.sealResult
            }
            service.openEnvelope = { [weak self] envelope in self?.openedEnvelopes.append(envelope) }
-            service.deliverToPeer = { [weak self] envelope, peer in self?.delivered.append((envelope, peer)) }
+            service.deliverToPeer = { [weak self] envelope, peer in
+                self?.delivered.append((envelope, peer))
+                return self?.deliverResult ?? false
+            }
            service.heldEnvelopes = { [weak self] cooldown in
                self?.heldCooldowns.append(cooldown)
                return self?.held ?? []
@@ -190,6 +194,9 @@ struct BridgeCourierServiceTests {
        fixture.sealResult = makeEnvelope(recipientKey: recipientKey)
        #expect(fixture.service.depositDrop(content: "hello", messageID: messageID, recipientNoiseKey: recipientKey))
        #expect(fixture.publishedEvents.count == 1)
+        // Persistence is coalesced; a real launch flushes within a second or
+        // on backgrounding — tests flush explicitly.
+        fixture.service.flushDedupSnapshot()

        // "Relaunch": a fresh service over the same store must refuse to
        // publish the same message ID again (before even re-sealing it).
@@ -214,6 +221,7 @@ struct BridgeCourierServiceTests {
        let event = try makeDropEvent(for: makeEnvelope(recipientKey: myKey))
        fixture.service.handleDropEvent(event)
        #expect(fixture.openedEnvelopes.count == 1)
+        fixture.service.flushDedupSnapshot()

        let relaunched = Fixture(dedupStore: BridgeDropDedupStore(fileURL: fileURL))
        relaunched.myKey = myKey
@@ -222,6 +230,83 @@ struct BridgeCourierServiceTests {
        #expect(relaunched.openedEnvelopes.isEmpty)
    }

+    @Test func offlineQueuedDropStaysRedepositableAfterRelaunch() throws {
+        // A deposit made while relays are down only joins the in-memory
+        // pending queue. Its dedup key must NOT be durable yet: if the app is
+        // killed before relays connect, the relaunch loses the queued drop —
+        // a persisted key would then block every 120s re-deposit for 24h and
+        // the message would silently never reach a relay.
+        let fileURL = FileManager.default.temporaryDirectory
+            .appendingPathComponent("bridge-dedup-\(UUID().uuidString).json")
+        defer { try? FileManager.default.removeItem(at: fileURL) }
+        let recipientKey = Fixture.randomKey()
+        let messageID = UUID().uuidString
+
+        let fixture = Fixture(dedupStore: BridgeDropDedupStore(fileURL: fileURL))
+        fixture.relaysConnected = false
+        fixture.sealResult = makeEnvelope(recipientKey: recipientKey)
+        #expect(fixture.service.depositDrop(content: "later", messageID: messageID, recipientNoiseKey: recipientKey))
+        #expect(fixture.publishedEvents.isEmpty)
+        // Even a flush while the drop is still pending must exclude its key.
+        fixture.service.flushDedupSnapshot()
+
+        // "App killed before relays connected": pendingDrops were memory-only.
+        let relaunched = Fixture(dedupStore: BridgeDropDedupStore(fileURL: fileURL))
+        relaunched.sealResult = makeEnvelope(recipientKey: recipientKey)
+        #expect(relaunched.service.depositDrop(content: "later", messageID: messageID, recipientNoiseKey: recipientKey))
+        #expect(relaunched.publishedEvents.count == 1)
+    }
+
+    @Test func publishedPendingDropBecomesDurableAfterFlush() throws {
+        // Counterpart: once the queued drop actually publishes on reconnect,
+        // its key becomes durable and a relaunch must not republish.
+        let fileURL = FileManager.default.temporaryDirectory
+            .appendingPathComponent("bridge-dedup-\(UUID().uuidString).json")
+        defer { try? FileManager.default.removeItem(at: fileURL) }
+        let recipientKey = Fixture.randomKey()
+        let messageID = UUID().uuidString
+
+        let fixture = Fixture(dedupStore: BridgeDropDedupStore(fileURL: fileURL))
+        fixture.relaysConnected = false
+        fixture.sealResult = makeEnvelope(recipientKey: recipientKey)
+        #expect(fixture.service.depositDrop(content: "later", messageID: messageID, recipientNoiseKey: recipientKey))
+        fixture.relaysConnected = true
+        fixture.service.flushPendingDrops()
+        #expect(fixture.publishedEvents.count == 1)
+        fixture.service.flushDedupSnapshot()
+
+        let relaunched = Fixture(dedupStore: BridgeDropDedupStore(fileURL: fileURL))
+        relaunched.sealResult = makeEnvelope(recipientKey: recipientKey)
+        #expect(!relaunched.service.depositDrop(content: "later", messageID: messageID, recipientNoiseKey: recipientKey))
+        #expect(relaunched.publishedEvents.isEmpty)
+    }
+
+    @Test func failedGatewayHandoffReleasesSeenSlot() throws {
+        // A gateway's deliverToPeer handoff is best-effort: when it fails
+        // (the peer walked away between relay fetch and mesh send), the drop
+        // event must stay retryable — for a single-gateway mesh island this
+        // gateway is the recipient's only carrier.
+        let fixture = Fixture()
+        let peerKey = Fixture.randomKey()
+        let peer = PeerID(str: "aabbccdd00112233")
+        fixture.localPeers = [(peer, peerKey)]
+        fixture.service.refresh()
+        let event = try makeDropEvent(for: makeEnvelope(recipientKey: peerKey))
+
+        fixture.deliverResult = false
+        fixture.service.handleDropEvent(event)
+        #expect(fixture.delivered.count == 1)
+
+        // Redelivery (relaunch/backlog re-fetch) retries the handoff …
+        fixture.deliverResult = true
+        fixture.service.handleDropEvent(event)
+        #expect(fixture.delivered.count == 2)
+
+        // … and a successful handoff consumes the event for good.
+        fixture.service.handleDropEvent(event)
+        #expect(fixture.delivered.count == 2)
+    }
+
    @Test func distinctDropsUseDistinctThrowawayKeys() {
        let fixture = Fixture()
        let keyA = Fixture.randomKey()