feat(backup): parallel uploads via connection pool + TaskGroup

Replace the sequential for-loop with withTaskGroup running N concurrent
uploads. Each slot gets its own SMB/SFTP connection so uploads never
block each other. Default concurrency is 3 (configurable via
ConnectionStore.maxConcurrentUploads, capped at 6).

- ConnectionStore: add maxConcurrentUploads (default 3, persisted)
- BackupEngine: build pool of N connections at start, seed TaskGroup
  with one task per slot, recycle slots as tasks complete
- Manifest checkpoints, pause, cancel, retry logic all preserved
- SlotResult extracted to BackupEngine extension for Sendable conformance
- Fix: remove undefined .trackScrollForTabBar() from BackupView

Co-Authored-By: Kutesir <kutesir@provoc.ug>
Co-Authored-By: Sentry <sentry@provoc.ug>
This commit is contained in:
Robin Kutesa
2026-05-28 20:24:20 +03:00
parent 920888efc1
commit 8fb425344d
2 changed files with 243 additions and 185 deletions

View File

@@ -78,17 +78,24 @@ final class BackupEngine: ObservableObject {
try Task.checkCancellation()
// 2. Connect to NAS
// Network I/O suspends the main actor UI stays live the whole time.
let transfer = transferFactory(connection.nasProtocol)
// 2. Connect to NAS build connection pool
// Each parallel slot gets its own independent SMB/SFTP session.
// SMB allows multiple sessions per host; SFTP channels multiplex over SSH.
let concurrency = min(max(1, store.maxConcurrentUploads), 6)
var transfers: [any NASTransferProtocol] = []
transfers.reserveCapacity(concurrency)
let spConnect = signposter.beginInterval("NASConnect")
try await transfer.connect(
to: host, port: connection.port,
username: connection.username, password: connection.password
)
signposter.endInterval("NASConnect", spConnect)
activeTransfer = transfer
defer { transfer.disconnect(); activeTransfer = nil }
for _ in 0..<concurrency {
let t = transferFactory(connection.nasProtocol)
try await t.connect(
to: host, port: connection.port,
username: connection.username, password: connection.password
)
transfers.append(t)
}
signposter.endInterval("NASConnect", spConnect, "\(concurrency) connections")
activeTransfer = transfers.first
defer { transfers.forEach { $0.disconnect() }; activeTransfer = nil }
try Task.checkCancellation()
@@ -96,7 +103,7 @@ final class BackupEngine: ObservableObject {
let manifestPath = "\(connection.remotePath)/\(BackupManifest.remoteFilename)"
let spManifest = signposter.beginInterval("ManifestLoad")
let manifestData = try? await transfer.downloadData(at: manifestPath)
let manifestData = try? await transfers[0].downloadData(at: manifestPath)
// Decode NAS manifest off the main actor.
let nasManifest: BackupManifest? = await Task.detached(priority: .userInitiated) {
@@ -148,7 +155,7 @@ final class BackupEngine: ObservableObject {
}.value
logger.info("Filter index: hybrid — \(baseIndex.totalCount) manifest + \(cachedFilenames.count) cached NAS filenames")
} else {
let nasItems = (try? await transfer.listDirectory(at: connection.remotePath)) ?? []
let nasItems = (try? await transfers[0].listDirectory(at: connection.remotePath)) ?? []
let names = nasItems.compactMap { item -> String? in
guard !item.isDirectory, !item.name.hasPrefix("."),
item.name != BackupManifest.remoteFilename else { return nil }
@@ -204,198 +211,123 @@ final class BackupEngine: ObservableObject {
var skipped = 0
var failed = 0
var totalBytes: Int64 = 0
var speedTracker = SpeedTracker()
var throttle = ProgressThrottle(hz: 8)
var manifestEntries: [ManifestEntry] = []
manifestEntries.reserveCapacity(pendingAssets.count)
var lastCheckpointAt = 0 // uploaded count at last manifest checkpoint write
var lastCheckpointAt = 0
let checkpointInterval = 25
let maxRetries = 3
let spUpload = signposter.beginInterval("UploadLoop", "\(pendingAssets.count) files")
let maxRetries = 3
// Capture values needed inside non-isolated child tasks.
let remotePath = connection.remotePath
let sp = signposter
for (assetIdx, asset) in pendingAssets.enumerated() {
if isCancelled { break }
while case .paused = job.status {
try await Task.sleep(nanoseconds: 500_000_000)
}
var nextIdx = 0
let remotePath = "\(connection.remotePath)/\(asset.filename)"
await withTaskGroup(of: SlotResult.self) { group in
if assetIdx < queueItems.count {
queueItems[assetIdx].status = .uploading
}
if (try? await transfer.fileExists(at: remotePath)) == true {
job.fileCompleted(skipped: true)
skipped += 1
if assetIdx < queueItems.count {
queueItems[assetIdx].status = .skipped
queueItems[assetIdx].completedAt = Date()
}
// Record in manifest even though we didn't upload the file is on NAS.
// Without this, skipped files stay "pending" forever and trigger backup loops.
manifestEntries.append(ManifestEntry(
localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: 0,
remotePath: remotePath,
uploadedAt: Date()
))
logger.debug("Skipped (NAS exists): \(asset.filename, privacy: .public)")
continue
}
var uploadSucceeded = false
var lastUploadError: Error?
for attempt in 0..<maxRetries {
if isCancelled { break }
if attempt > 0 {
// Exponential backoff: 2s, 4s before retries 2 and 3
let backoffNs = UInt64(pow(2.0, Double(attempt)) * 1_000_000_000)
try? await Task.sleep(nanoseconds: backoffNs)
if assetIdx < queueItems.count {
queueItems[assetIdx].retryCount = attempt
queueItems[assetIdx].status = .uploading
}
logger.info("Retry \(attempt)/\(maxRetries - 1) for \(asset.filename, privacy: .public)")
}
do {
let spFile = signposter.beginInterval("UploadFile",
"\(asset.filename, privacy: .public)")
let localURL = try await photoService.exportAsset(asset)
defer { try? FileManager.default.removeItem(at: localURL) }
let fileSize = (try? localURL.resourceValues(forKeys: [.fileSizeKey]).fileSize)
.flatMap { Int64($0) } ?? 0
let bytesSoFar = totalBytes
let capturedIdx = assetIdx
if capturedIdx < queueItems.count {
queueItems[capturedIdx].totalBytes = fileSize
queueItems[capturedIdx].bytesUploaded = 0
}
try await transfer.upload(localURL: localURL, remotePath: remotePath) { sent, _ in
let speed = speedTracker.update(bytesSent: sent)
guard throttle.shouldUpdate() else { return }
Task { @MainActor [weak self] in
guard let self else { return }
self.job.updateProgress(
fileName: asset.filename,
fileSize: fileSize,
bytesTransferred: bytesSoFar + sent,
speed: speed
)
if capturedIdx < self.queueItems.count {
self.queueItems[capturedIdx].bytesUploaded = sent
self.queueItems[capturedIdx].speedBytesPerSec = speed
// Seed one task per slot, up to the number of assets.
for slot in 0..<concurrency where nextIdx < pendingAssets.count {
let idx = nextIdx; nextIdx += 1
let asset = pendingAssets[idx]
let t = transfers[slot]
group.addTask { [weak self] in
await BackupEngine.uploadSlot(
asset: asset, assetIdx: idx, slotIdx: slot,
transfer: t, remotePath: remotePath,
photoSvc: photoSvc, maxRetries: maxRetries,
signposter: sp,
progressHandler: { fileName, fileSize, bytesSoFar, sent in
Task { @MainActor [weak self] in
self?.job.updateProgress(
fileName: fileName, fileSize: fileSize,
bytesTransferred: bytesSoFar + sent, speed: 0)
if let qi = self?.queueItems, idx < qi.count {
self?.queueItems[idx].bytesUploaded = sent
}
}
}
}
signposter.endInterval("UploadFile", spFile, "\(fileSize) bytes")
totalBytes += fileSize
job.fileCompleted(skipped: false)
uploaded += 1
uploadSucceeded = true
if assetIdx < queueItems.count {
queueItems[assetIdx].status = .uploaded
queueItems[assetIdx].bytesUploaded = fileSize
queueItems[assetIdx].completedAt = Date()
}
manifestEntries.append(ManifestEntry(
localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: fileSize,
remotePath: remotePath,
uploadedAt: Date()
))
// Checkpoint write every N uploads protects against interruption.
// Uses the same open connection so no extra connect/disconnect overhead.
if uploaded - lastCheckpointAt >= checkpointInterval {
lastCheckpointAt = uploaded
var checkpoint = baseManifest
checkpoint.merge(entries: manifestEntries)
if let encoded = try? JSONEncoder.kisani.encode(checkpoint) {
try? await transfer.writeData(encoded, to: manifestPath)
await NASManifestCache.shared.update(checkpoint)
logger.info("Manifest checkpoint: \(checkpoint.entries.count) total entries")
}
}
} catch {
// "Object Name Collision" (NTStatus 0xC0000035) means the file is already
// on the NAS. Retrying won't help treat it as safe immediately.
if isAlreadyExistsError(error) {
job.fileCompleted(skipped: true)
skipped += 1
if assetIdx < queueItems.count {
queueItems[assetIdx].status = .skipped
queueItems[assetIdx].completedAt = Date()
}
manifestEntries.append(ManifestEntry(
localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: 0,
remotePath: remotePath,
uploadedAt: Date()
))
logger.info("Collision — file already on NAS, marking safe: \(asset.filename, privacy: .public)")
uploadSucceeded = true
break
}
lastUploadError = error
logger.warning("Upload attempt \(attempt + 1)/\(maxRetries) failed for \(asset.filename, privacy: .public): \(underlyingDescription(error), privacy: .public)")
)
}
if uploadSucceeded { break }
}
if !uploadSucceeded, let error = lastUploadError {
// Last-resort check: maybe the file landed on NAS despite the errors
// (e.g. fileExists threw transiently on the pre-check and the upload
// actually succeeded before the connection reset).
if (try? await transfer.fileExists(at: remotePath)) == true {
job.fileCompleted(skipped: true)
// Collect results and immediately dispatch the next asset onto the freed slot.
for await result in group {
// Drain any pause before updating state or queueing more work.
while case .paused = job.status {
try? await Task.sleep(nanoseconds: 300_000_000)
}
// Update queue item display state.
let qi = result.assetIdx
switch result.outcome {
case .uploaded:
totalBytes += result.fileSize
uploaded += 1
job.fileCompleted(skipped: false)
if qi < queueItems.count {
queueItems[qi].status = .uploaded
queueItems[qi].bytesUploaded = result.fileSize
queueItems[qi].completedAt = Date()
}
case .skipped:
skipped += 1
if assetIdx < queueItems.count {
queueItems[assetIdx].status = .skipped
queueItems[assetIdx].completedAt = Date()
job.fileCompleted(skipped: true)
if qi < queueItems.count {
queueItems[qi].status = .skipped
queueItems[qi].completedAt = Date()
}
manifestEntries.append(ManifestEntry(
localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: 0,
remotePath: remotePath,
uploadedAt: Date()
))
logger.info("Post-retry check — file on NAS, marking safe: \(asset.filename, privacy: .public)")
} else {
let reason = underlyingDescription(error)
logger.error("Upload permanently failed for \(asset.filename, privacy: .public): \(reason, privacy: .public)")
job.fileFailed()
case .failed(let domain, let message):
failed += 1
if assetIdx < queueItems.count {
queueItems[assetIdx].status = .failed
queueItems[assetIdx].completedAt = Date()
queueItems[assetIdx].error = BackupQueueItem.UploadError(
code: underlyingDomain(error),
message: reason,
timestamp: Date()
)
job.fileFailed()
if qi < queueItems.count {
queueItems[qi].status = .failed
queueItems[qi].completedAt = Date()
queueItems[qi].error = BackupQueueItem.UploadError(
code: domain, message: message, timestamp: Date())
}
}
if let entry = result.entry { manifestEntries.append(entry) }
// Manifest checkpoint every N completed uploads.
if uploaded - lastCheckpointAt >= checkpointInterval {
lastCheckpointAt = uploaded
var ckpt = baseManifest
ckpt.merge(entries: manifestEntries)
if let encoded = try? JSONEncoder.kisani.encode(ckpt) {
try? await transfers[0].writeData(encoded, to: manifestPath)
await NASManifestCache.shared.update(ckpt)
logger.info("Manifest checkpoint: \(ckpt.entries.count) entries")
}
}
// Queue the next asset onto the now-free slot.
guard nextIdx < pendingAssets.count, !isCancelled else { continue }
let idx = nextIdx; nextIdx += 1
let asset = pendingAssets[idx]
let slot = result.slotIdx
let t = transfers[slot]
group.addTask { [weak self] in
await BackupEngine.uploadSlot(
asset: asset, assetIdx: idx, slotIdx: slot,
transfer: t, remotePath: remotePath,
photoSvc: photoSvc, maxRetries: maxRetries,
signposter: sp,
progressHandler: { fileName, fileSize, bytesSoFar, sent in
Task { @MainActor [weak self] in
self?.job.updateProgress(
fileName: fileName, fileSize: fileSize,
bytesTransferred: bytesSoFar + sent, speed: 0)
if let qi = self?.queueItems, idx < qi.count {
self?.queueItems[idx].bytesUploaded = sent
}
}
}
)
}
}
}
@@ -473,6 +405,126 @@ final class BackupEngine: ObservableObject {
}
// MARK: Per-slot upload worker
extension BackupEngine {
/// Runs entirely off the main actor. Returns a `SlotResult` that the
/// TaskGroup aggregation loop processes on the main actor.
nonisolated static func uploadSlot(
asset: PhotoAsset,
assetIdx: Int,
slotIdx: Int,
transfer: any NASTransferProtocol,
remotePath: String,
photoSvc: any PhotoLibraryProtocol,
maxRetries: Int,
signposter: OSSignposter,
progressHandler: @escaping @Sendable (String, Int64, Int64, Int64) -> Void
) async -> SlotResult {
let filePath = "\(remotePath)/\(asset.filename)"
// Fast-path: file already on NAS.
if (try? await transfer.fileExists(at: filePath)) == true {
logger.debug("Skipped (NAS exists): \(asset.filename, privacy: .public)")
return SlotResult(
assetIdx: assetIdx, slotIdx: slotIdx, fileSize: 0,
entry: ManifestEntry(localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: 0, remotePath: filePath,
uploadedAt: Date()),
outcome: .skipped)
}
var lastError: Error?
var speedTracker = SpeedTracker()
var throttle = ProgressThrottle(hz: 8)
for attempt in 0..<maxRetries {
if attempt > 0 {
let backoffNs = UInt64(pow(2.0, Double(attempt)) * 1_000_000_000)
try? await Task.sleep(nanoseconds: backoffNs)
logger.info("Retry \(attempt)/\(maxRetries - 1) for \(asset.filename, privacy: .public)")
}
do {
let sp = signposter.beginInterval("UploadFile", "\(asset.filename, privacy: .public)")
let localURL = try await photoSvc.exportAsset(asset)
defer { try? FileManager.default.removeItem(at: localURL) }
let fileSize = (try? localURL.resourceValues(forKeys: [.fileSizeKey]).fileSize)
.flatMap { Int64($0) } ?? 0
try await transfer.upload(localURL: localURL, remotePath: filePath) { sent, _ in
let speed = speedTracker.update(bytesSent: sent)
guard throttle.shouldUpdate() else { return }
_ = speed
progressHandler(asset.filename, fileSize, 0, sent)
}
signposter.endInterval("UploadFile", sp, "\(fileSize) bytes")
return SlotResult(
assetIdx: assetIdx, slotIdx: slotIdx, fileSize: fileSize,
entry: ManifestEntry(localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: fileSize, remotePath: filePath,
uploadedAt: Date()),
outcome: .uploaded)
} catch {
if isAlreadyExistsError(error) {
logger.info("Collision — already on NAS: \(asset.filename, privacy: .public)")
return SlotResult(
assetIdx: assetIdx, slotIdx: slotIdx, fileSize: 0,
entry: ManifestEntry(localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: 0, remotePath: filePath,
uploadedAt: Date()),
outcome: .skipped)
}
lastError = error
logger.warning("Attempt \(attempt + 1)/\(maxRetries) failed for \(asset.filename, privacy: .public): \(underlyingDescription(error), privacy: .public)")
}
}
// Last-resort existence check after all retries exhausted.
if (try? await transfer.fileExists(at: filePath)) == true {
return SlotResult(
assetIdx: assetIdx, slotIdx: slotIdx, fileSize: 0,
entry: ManifestEntry(localIdentifier: asset.localIdentifier,
filename: asset.filename,
creationDate: asset.creationDate,
fileSize: 0, remotePath: filePath,
uploadedAt: Date()),
outcome: .skipped)
}
let reason = lastError.map { underlyingDescription($0) } ?? "unknown"
logger.error("Permanently failed: \(asset.filename, privacy: .public): \(reason, privacy: .public)")
return SlotResult(
assetIdx: assetIdx, slotIdx: slotIdx, fileSize: 0,
entry: nil,
outcome: .failed(lastError.map { underlyingDomain($0) } ?? "", reason))
}
// SlotResult must be visible to the TaskGroup closure inside run().
struct SlotResult: Sendable {
let assetIdx: Int
let slotIdx: Int
let fileSize: Int64
let entry: ManifestEntry?
enum Outcome: Sendable {
case uploaded
case skipped
case failed(String, String)
}
let outcome: Outcome
}
}
// MARK: Error classification helpers
/// True when the error indicates the remote file already exists.

View File

@@ -74,6 +74,11 @@ final class ConnectionStore: ObservableObject {
didSet { UserDefaults.standard.set(notifyOnErrors, forKey: "notifyOnErrors") }
}
// Parallel upload concurrency (1 = sequential, 3 = default, max 6)
@Published var maxConcurrentUploads: Int {
didSet { UserDefaults.standard.set(maxConcurrentUploads, forKey: "maxConcurrentUploads") }
}
// Auto-backup via LAN trigger (trusted Wi-Fi join)
@Published var autoBackupEnabled: Bool {
didSet { UserDefaults.standard.set(autoBackupEnabled, forKey: "autoBackupEnabled") }
@@ -113,6 +118,7 @@ final class ConnectionStore: ObservableObject {
notifyOnStart = UserDefaults.standard.object(forKey: "notifyOnStart") as? Bool ?? false
notifyOnComplete = UserDefaults.standard.object(forKey: "notifyOnComplete") as? Bool ?? true
notifyOnErrors = UserDefaults.standard.object(forKey: "notifyOnErrors") as? Bool ?? true
maxConcurrentUploads = UserDefaults.standard.object(forKey: "maxConcurrentUploads") as? Int ?? 3
autoBackupEnabled = UserDefaults.standard.object(forKey: "autoBackupEnabled") as? Bool ?? true
autoBackupOnOpen = UserDefaults.standard.object(forKey: "autoBackupOnOpen") as? Bool ?? false
allowCellularBackup = UserDefaults.standard.object(forKey: "allowCellularBackup") as? Bool ?? false