581 lines
20 KiB
Swift
581 lines
20 KiB
Swift
//
|
|
// Copyright 2018 Signal Messenger, LLC
|
|
// SPDX-License-Identifier: AGPL-3.0-only
|
|
//
|
|
|
|
import Foundation
|
|
import GRDB
|
|
import SignalCoreKit
|
|
|
|
/// JobQueue - A durable work queue
|
|
///
|
|
/// When work needs to be done, add it to the JobQueue.
|
|
/// The JobQueue will persist a JobRecord to be sure that work can be restarted if the app is killed.
|
|
///
|
|
/// The actual work, is carried out in a DurableOperation which the JobQueue spins off, based on the contents
|
|
/// of a JobRecord.
|
|
///
|
|
/// For a concrete example, take message sending.
|
|
/// Add an outgoing message to the MessageSenderJobQueue, which first records a SSKMessageSenderJobRecord.
|
|
/// The MessageSenderJobQueue then uses that SSKMessageSenderJobRecord to create a MessageSenderOperation which
|
|
/// takes care of the actual business of communicating with the service.
|
|
///
|
|
/// DurableOperations are retryable - via their `remainingRetries` logic. However, if the operation encounters
|
|
/// an error where `error.isRetryable == false`, the operation will fail, regardless of available retries.
|
|
|
|
public enum JobError: Error {
|
|
case permanentFailure(description: String)
|
|
case obsolete(description: String)
|
|
}
|
|
|
|
public protocol DurableOperation: AnyObject, Equatable {
|
|
associatedtype JobRecordType: JobRecord
|
|
associatedtype DurableOperationDelegateType: DurableOperationDelegate
|
|
|
|
var jobRecord: JobRecordType { get }
|
|
var durableOperationDelegate: DurableOperationDelegateType? { get set }
|
|
var operation: OWSOperation { get }
|
|
|
|
var maxRetries: UInt { get }
|
|
|
|
/// This could be computed using ``maxRetries`` and ``JobRecord/failureCount``,
|
|
/// but because in practice these are ``OWSOperation``s we make it a mutable
|
|
/// stored property instead for compatibility.
|
|
var remainingRetries: UInt { get set }
|
|
}
|
|
|
|
public protocol DurableOperationDelegate: AnyObject {
|
|
associatedtype DurableOperationType: DurableOperation
|
|
|
|
func durableOperationDidSucceed(_ operation: DurableOperationType, transaction: SDSAnyWriteTransaction)
|
|
func durableOperation(_ operation: DurableOperationType, didReportError: Error, transaction: SDSAnyWriteTransaction)
|
|
func durableOperation(_ operation: DurableOperationType, didFailWithError error: Error, transaction: SDSAnyWriteTransaction)
|
|
}
|
|
|
|
public protocol JobQueue: DurableOperationDelegate, Dependencies {
|
|
typealias JobRecordType = DurableOperationType.JobRecordType
|
|
|
|
var runningOperations: AtomicArray<DurableOperationType> { get set }
|
|
|
|
var isSetup: AtomicBool { get }
|
|
func setup()
|
|
func didMarkAsReady(oldJobRecord: JobRecordType, transaction: SDSAnyWriteTransaction)
|
|
|
|
func operationQueue(jobRecord: JobRecordType) -> OperationQueue
|
|
func buildOperation(jobRecord: JobRecordType, transaction: SDSAnyReadTransaction) throws -> DurableOperationType
|
|
|
|
/// When `requiresInternet` is true, we immediately run any jobs which are waiting for retry upon detecting Reachability.
|
|
///
|
|
/// Because `Reachability` isn't 100% reliable, the jobs will be attempted regardless of what we think our current Reachability is.
|
|
/// However, because these jobs will likely fail many times in succession, their `retryInterval` could be quite long by the time we
|
|
/// are back online.
|
|
var requiresInternet: Bool { get }
|
|
|
|
var isEnabled: Bool { get }
|
|
}
|
|
|
|
// MARK: -
|
|
|
|
public extension JobQueue {
|
|
|
|
func add(
|
|
jobRecord: JobRecordType,
|
|
transaction: SDSAnyWriteTransaction
|
|
) {
|
|
owsAssertDebug(jobRecord.status == .ready)
|
|
|
|
jobRecord.anyInsert(transaction: transaction)
|
|
|
|
transaction.addTransactionFinalizationBlock(
|
|
forKey: "jobQueue.\(JobRecordType.jobRecordType.jobRecordLabel).startWorkImmediatelyIfAppIsReady"
|
|
) { transaction in
|
|
self.startWorkImmediatelyIfAppIsReady(transaction: transaction)
|
|
}
|
|
|
|
transaction.addAsyncCompletion(queue: .global()) {
|
|
self.startWorkWhenAppIsReady()
|
|
}
|
|
}
|
|
|
|
func startWorkImmediatelyIfAppIsReady(transaction: SDSAnyWriteTransaction) {
|
|
guard isEnabled else { return }
|
|
guard !CurrentAppContext().isRunningTests else { return }
|
|
guard AppReadiness.isAppReady else { return }
|
|
guard isSetup.get() else { return }
|
|
workStep(transaction: transaction)
|
|
}
|
|
|
|
func startWorkWhenAppIsReady() {
|
|
guard isEnabled else { return }
|
|
|
|
guard !CurrentAppContext().isRunningTests else {
|
|
DispatchQueue.global().async {
|
|
self.workStep()
|
|
}
|
|
return
|
|
}
|
|
|
|
AppReadiness.runNowOrWhenAppDidBecomeReadyAsync {
|
|
guard self.isSetup.get() else {
|
|
return
|
|
}
|
|
DispatchQueue.global().async {
|
|
self.workStep()
|
|
}
|
|
}
|
|
}
|
|
|
|
func workStep() {
|
|
guard isEnabled else { return }
|
|
|
|
guard isSetup.get() else {
|
|
if !CurrentAppContext().isRunningTests {
|
|
owsFailDebug("not setup")
|
|
}
|
|
|
|
return
|
|
}
|
|
|
|
databaseStorage.write { self.workStep(transaction: $0) }
|
|
}
|
|
|
|
func workStep(transaction: SDSAnyWriteTransaction) {
|
|
let nextJob: JobRecordType?
|
|
|
|
do {
|
|
nextJob = try JobRecordFinderImpl(db: DependenciesBridge.shared.db).getNextReady(transaction: transaction.asV2Write)
|
|
} catch let error {
|
|
Logger.error("Couldn't start next job: \(error)")
|
|
return
|
|
}
|
|
|
|
guard let nextJob else {
|
|
return
|
|
}
|
|
|
|
do {
|
|
try nextJob.saveReadyAsRunning(transaction: transaction)
|
|
|
|
let operationQueue = operationQueue(jobRecord: nextJob)
|
|
let durableOperation = try buildOperation(jobRecord: nextJob, transaction: transaction)
|
|
|
|
durableOperation.durableOperationDelegate = self as? Self.DurableOperationType.DurableOperationDelegateType
|
|
owsAssertDebug(durableOperation.durableOperationDelegate != nil)
|
|
|
|
let remainingRetries = remainingRetries(durableOperation: durableOperation)
|
|
durableOperation.remainingRetries = remainingRetries
|
|
|
|
transaction.addSyncCompletion {
|
|
self.runningOperations.append(durableOperation)
|
|
operationQueue.addOperation(durableOperation.operation)
|
|
}
|
|
} catch JobError.permanentFailure(let description) {
|
|
owsFailDebug("permanent failure: \(description)")
|
|
nextJob.saveAsPermanentlyFailed(transaction: transaction)
|
|
} catch JobError.obsolete(let description) {
|
|
// TODO is this even worthwhile to have obsolete state? Should we just delete the task outright?
|
|
nextJob.saveAsObsolete(transaction: transaction)
|
|
} catch {
|
|
owsFailDebug("unexpected error")
|
|
}
|
|
|
|
transaction.addAsyncCompletionOffMain { self.workStep() }
|
|
}
|
|
|
|
func restartOldJobs() {
|
|
guard CurrentAppContext().isMainApp else { return }
|
|
guard isEnabled else { return }
|
|
|
|
databaseStorage.write { transaction in
|
|
let runningRecords: [JobRecordType]
|
|
do {
|
|
runningRecords = try JobRecordFinderImpl(db: DependenciesBridge.shared.db).allRecords(
|
|
status: JobRecord.Status.running,
|
|
transaction: transaction.asV2Write
|
|
)
|
|
} catch {
|
|
Logger.error("Couldn't restart old jobs: \(error)")
|
|
return
|
|
}
|
|
Logger.info("marking old `running` \(JobRecordType.jobRecordType.jobRecordLabel) JobRecords as ready: \(runningRecords.count)")
|
|
for jobRecord in runningRecords {
|
|
do {
|
|
try jobRecord.saveRunningAsReady(transaction: transaction)
|
|
self.didMarkAsReady(oldJobRecord: jobRecord, transaction: transaction)
|
|
} catch {
|
|
owsFailDebug("failed to mark old running records as ready error: \(error)")
|
|
jobRecord.saveAsPermanentlyFailed(transaction: transaction)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func pruneStaleJobs() {
|
|
guard CurrentAppContext().isMainApp else { return }
|
|
guard isEnabled else { return }
|
|
|
|
databaseStorage.write { transaction in
|
|
let staleRecords: [JobRecordType]
|
|
do {
|
|
staleRecords = try JobRecordFinderImpl(db: DependenciesBridge.shared.db).staleRecords(transaction: transaction.asV2Write)
|
|
} catch {
|
|
Logger.error("Failed to prune stale jobs! \(error)")
|
|
return
|
|
}
|
|
|
|
Logger.info("Pruning stale \(JobRecordType.jobRecordType.jobRecordLabel) job records: \(staleRecords.count).")
|
|
|
|
for jobRecord in staleRecords {
|
|
jobRecord.anyRemove(transaction: transaction)
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Unless you need special handling, your setup method can be as simple as
|
|
///
|
|
/// func setup() {
|
|
/// defaultSetup()
|
|
/// }
|
|
///
|
|
/// So you might ask, why not just rename this method to `setup`? Because
|
|
/// `setup` is called from objc, and default implementations from a protocol
|
|
/// cannot be marked as @objc.
|
|
func defaultSetup() {
|
|
guard isEnabled else { return }
|
|
|
|
guard !isSetup.get() else {
|
|
owsFailDebug("already ready already")
|
|
return
|
|
}
|
|
|
|
DispatchQueue.global().async(.promise) {
|
|
self.restartOldJobs()
|
|
self.pruneStaleJobs()
|
|
}.done { [weak self] in
|
|
guard let self = self else {
|
|
return
|
|
}
|
|
if self.requiresInternet {
|
|
// FIXME: The returned observer token is never unregistered.
|
|
// In practice all our JobQueues live forever, so this isn't a problem.
|
|
NotificationCenter.default.addObserver(
|
|
forName: SSKReachability.owsReachabilityDidChange,
|
|
object: nil,
|
|
queue: nil
|
|
) { _ in
|
|
if self.reachabilityManager.isReachable {
|
|
self.becameReachable()
|
|
}
|
|
}
|
|
}
|
|
|
|
self.isSetup.set(true)
|
|
self.startWorkWhenAppIsReady()
|
|
}
|
|
}
|
|
|
|
func remainingRetries(durableOperation: DurableOperationType) -> UInt {
|
|
let maxRetries = durableOperation.maxRetries
|
|
let failureCount = durableOperation.jobRecord.failureCount
|
|
|
|
guard maxRetries > failureCount else {
|
|
return 0
|
|
}
|
|
|
|
return maxRetries - failureCount
|
|
}
|
|
|
|
func becameReachable() {
|
|
guard requiresInternet else {
|
|
owsFailDebug("should only be called if `requiresInternet` is true")
|
|
return
|
|
}
|
|
|
|
_ = self.runAnyQueuedRetry()
|
|
}
|
|
|
|
func runAnyQueuedRetry() -> DurableOperationType? {
|
|
guard let runningDurableOperation = self.runningOperations.first else {
|
|
return nil
|
|
}
|
|
runningDurableOperation.operation.runAnyQueuedRetry()
|
|
|
|
return runningDurableOperation
|
|
}
|
|
|
|
// MARK: DurableOperationDelegate
|
|
|
|
func durableOperationDidSucceed(_ operation: DurableOperationType, transaction: SDSAnyWriteTransaction) {
|
|
runningOperations.remove(operation)
|
|
operation.jobRecord.anyRemove(transaction: transaction)
|
|
}
|
|
|
|
func durableOperation(_ operation: DurableOperationType, didReportError: Error, transaction: SDSAnyWriteTransaction) {
|
|
do {
|
|
try operation.jobRecord.addFailure(transaction: transaction)
|
|
} catch {
|
|
owsFailDebug("error while addingFailure: \(error)")
|
|
operation.jobRecord.saveAsPermanentlyFailed(transaction: transaction)
|
|
}
|
|
}
|
|
|
|
func durableOperation(_ operation: DurableOperationType, didFailWithError error: Error, transaction: SDSAnyWriteTransaction) {
|
|
runningOperations.remove(operation)
|
|
operation.jobRecord.saveAsPermanentlyFailed(transaction: transaction)
|
|
}
|
|
}
|
|
|
|
public protocol JobRecordFinder<JobRecordType> {
|
|
associatedtype JobRecordType: JobRecord
|
|
|
|
/// Fetches a single JobRecord from the database.
|
|
///
|
|
/// Returns `nil` a JobRecord doesn't exist for `rowId`.
|
|
func fetchJob(rowId: JobRecord.RowId, tx: DBReadTransaction) throws -> JobRecordType?
|
|
|
|
/// Removes a single JobRecord from the database.
|
|
func removeJob(_ jobRecord: JobRecordType, tx: DBWriteTransaction)
|
|
|
|
/// Fetches all runnable jobs.
|
|
///
|
|
/// This method may use multiple transactions, may use write transactions,
|
|
/// may delete jobs that can't ever be run, etc.
|
|
///
|
|
/// It returns all jobs that can be run.
|
|
///
|
|
/// Conforming types should avoid long-running write transactions.
|
|
func loadRunnableJobs() async throws -> [JobRecordType]
|
|
|
|
func enumerateJobRecords(
|
|
transaction: DBReadTransaction,
|
|
block: (JobRecordType, inout Bool) -> Void
|
|
) throws
|
|
|
|
func enumerateJobRecords(
|
|
status: JobRecord.Status,
|
|
transaction: DBReadTransaction,
|
|
block: (JobRecordType, inout Bool) -> Void
|
|
) throws
|
|
}
|
|
|
|
public extension JobRecordFinder {
|
|
func getNextReady(transaction: DBReadTransaction) throws -> JobRecordType? {
|
|
var result: JobRecordType?
|
|
try enumerateJobRecords(status: .ready, transaction: transaction) { jobRecord, stop in
|
|
// Skip job records that aren't for the current process, we can't run these.
|
|
guard jobRecord.canBeRunByCurrentProcess else {
|
|
return
|
|
}
|
|
result = jobRecord
|
|
stop = true
|
|
}
|
|
return result
|
|
}
|
|
|
|
func allRecords(status: JobRecord.Status, transaction: DBReadTransaction) throws -> [JobRecordType] {
|
|
var result: [JobRecordType] = []
|
|
try enumerateJobRecords(status: status, transaction: transaction) { jobRecord, _ in
|
|
result.append(jobRecord)
|
|
}
|
|
return result
|
|
}
|
|
|
|
func staleRecords(transaction: DBReadTransaction) throws -> [JobRecordType] {
|
|
var result: [JobRecordType] = []
|
|
|
|
try enumerateJobRecords(transaction: transaction) { jobRecord, _ in
|
|
let isStale: Bool = {
|
|
switch jobRecord.status {
|
|
case .running:
|
|
return false
|
|
case .ready:
|
|
return !jobRecord.canBeRunByCurrentProcess
|
|
case
|
|
.obsolete,
|
|
.permanentlyFailed,
|
|
.unknown:
|
|
return true
|
|
}
|
|
}()
|
|
|
|
if isStale {
|
|
result.append(jobRecord)
|
|
}
|
|
}
|
|
|
|
return result
|
|
}
|
|
}
|
|
|
|
private enum Constants {
|
|
/// The number of JobRecords to fetch in a batch.
|
|
///
|
|
/// Most job queues won't ever have more than a few records at the same
|
|
/// time. Other times, a job queue may build up a huge backlog, and this
|
|
/// value can help prune it efficiently.
|
|
static let batchSize = 400
|
|
}
|
|
|
|
public class JobRecordFinderImpl<JobRecordType>: JobRecordFinder where JobRecordType: JobRecord {
|
|
private let db: DB
|
|
|
|
public init(db: DB) {
|
|
self.db = db
|
|
}
|
|
|
|
private func iterateJobsWith(
|
|
sql: String,
|
|
arguments: StatementArguments,
|
|
database: Database,
|
|
block: (JobRecordType, inout Bool) -> Void
|
|
) throws {
|
|
let cursor = try JobRecordType.fetchCursor(
|
|
database,
|
|
sql: sql,
|
|
arguments: arguments
|
|
)
|
|
|
|
var stop = false
|
|
while let nextJobRecord = try cursor.next() {
|
|
block(nextJobRecord, &stop)
|
|
|
|
if stop {
|
|
return
|
|
}
|
|
}
|
|
}
|
|
|
|
public func enumerateJobRecords(
|
|
transaction: DBReadTransaction,
|
|
block: (JobRecordType, inout Bool) -> Void
|
|
) throws {
|
|
let transaction = SDSDB.shimOnlyBridge(transaction)
|
|
|
|
let sql = """
|
|
SELECT * FROM \(JobRecord.databaseTableName)
|
|
WHERE \(JobRecord.columnName(.label)) = ?
|
|
ORDER BY \(JobRecord.columnName(.id))
|
|
"""
|
|
|
|
try iterateJobsWith(
|
|
sql: sql,
|
|
arguments: [JobRecordType.jobRecordType.jobRecordLabel],
|
|
database: transaction.unwrapGrdbRead.database,
|
|
block: block
|
|
)
|
|
}
|
|
|
|
public func enumerateJobRecords(
|
|
status: JobRecord.Status,
|
|
transaction: DBReadTransaction,
|
|
block: (JobRecordType, inout Bool) -> Void
|
|
) throws {
|
|
let transaction = SDSDB.shimOnlyBridge(transaction)
|
|
|
|
let sql = """
|
|
SELECT * FROM \(JobRecord.databaseTableName)
|
|
WHERE \(JobRecord.columnName(.status)) = ?
|
|
AND \(JobRecord.columnName(.label)) = ?
|
|
ORDER BY \(JobRecord.columnName(.id))
|
|
"""
|
|
|
|
try iterateJobsWith(
|
|
sql: sql,
|
|
arguments: [status.rawValue, JobRecordType.jobRecordType.jobRecordLabel],
|
|
database: transaction.unwrapGrdbRead.database,
|
|
block: block
|
|
)
|
|
}
|
|
|
|
public func fetchJob(rowId: JobRecord.RowId, tx: DBReadTransaction) throws -> JobRecordType? {
|
|
do {
|
|
let db = SDSDB.shimOnlyBridge(tx).unwrapGrdbRead.database
|
|
return try JobRecordType.fetchOne(db, key: rowId)
|
|
} catch {
|
|
throw error.grdbErrorForLogging
|
|
}
|
|
}
|
|
|
|
public func removeJob(_ jobRecord: JobRecordType, tx: DBWriteTransaction) {
|
|
jobRecord.anyRemove(transaction: SDSDB.shimOnlyBridge(tx))
|
|
}
|
|
|
|
public func loadRunnableJobs() async throws -> [JobRecordType] {
|
|
var allRunnableJobs = [JobRecordType]()
|
|
var afterRowId: JobRecord.RowId?
|
|
while true {
|
|
let (runnableJobs, hasMoreAfterRowId) = try await db.awaitableWrite { tx in
|
|
try self.fetchAndPruneSomePersistedJobs(afterRowId: afterRowId, tx: tx)
|
|
}
|
|
allRunnableJobs.append(contentsOf: runnableJobs)
|
|
guard let hasMoreAfterRowId else {
|
|
break
|
|
}
|
|
afterRowId = hasMoreAfterRowId
|
|
}
|
|
return allRunnableJobs
|
|
}
|
|
|
|
private func fetchAndPruneSomePersistedJobs(
|
|
afterRowId: JobRecord.RowId?,
|
|
tx: DBWriteTransaction
|
|
) throws -> ([JobRecordType], hasMoreAfterRowId: JobRecord.RowId?) {
|
|
let (jobs, hasMore) = try fetchSomeJobs(afterRowId: afterRowId, tx: tx)
|
|
var runnableJobs = [JobRecordType]()
|
|
for job in jobs {
|
|
let canRunJob: Bool = {
|
|
// This property is deprecated. If it's set, it means the job was created
|
|
// for a prior version of the application, and that version definitely
|
|
// can't be the current process.
|
|
guard job.exclusiveProcessIdentifier == nil else {
|
|
return false
|
|
}
|
|
// If a job has failed or is obsolete, we can remove it. We previously
|
|
// distinguished `.ready` from `.running`, but now they're treated exactly
|
|
// the same when we restart existing jobs.
|
|
switch job.status {
|
|
case .unknown, .permanentlyFailed, .obsolete:
|
|
return false
|
|
case .ready, .running:
|
|
break
|
|
}
|
|
return true
|
|
}()
|
|
if canRunJob {
|
|
runnableJobs.append(job)
|
|
} else {
|
|
removeJob(job, tx: tx)
|
|
}
|
|
}
|
|
return (runnableJobs, hasMore ? jobs.last!.id! : nil)
|
|
}
|
|
|
|
private func fetchSomeJobs(
|
|
afterRowId: JobRecord.RowId?,
|
|
tx: DBReadTransaction
|
|
) throws -> ([JobRecordType], hasMore: Bool) {
|
|
var sql = """
|
|
SELECT * FROM \(JobRecordType.databaseTableName)
|
|
WHERE "\(JobRecordType.columnName(.label))" = ?
|
|
"""
|
|
var arguments: StatementArguments = [JobRecordType.jobRecordType.jobRecordLabel]
|
|
if let afterRowId {
|
|
sql += """
|
|
AND \(JobRecordType.columnName(.id)) > ?
|
|
"""
|
|
arguments += [afterRowId]
|
|
}
|
|
sql += """
|
|
ORDER BY "\(JobRecordType.columnName(.id))"
|
|
LIMIT \(Constants.batchSize)
|
|
"""
|
|
do {
|
|
let db = SDSDB.shimOnlyBridge(tx).unwrapGrdbRead.database
|
|
let results = try JobRecordType.fetchAll(db, sql: sql, arguments: arguments)
|
|
return (results, results.count == Constants.batchSize)
|
|
} catch {
|
|
throw error.grdbErrorForLogging
|
|
}
|
|
}
|
|
}
|