I want to call a completion handler synchronously inside a critical section (using #synchronized block). I am trying to wait for completion handler using semaphore, but the semaphore signal is never called.
Here is what I am doing:
NSNumber *lock = 0;
#synchronized(lock) {
// critical section code begins
dispatch_semaphore_t sema = dispatch_semaphore_create(0);
[self someMethodWithCompletionHandler:^(BOOL result) {
// execute completion handler
dispatch_semaphore_signal(sema);
}];
// wait for completion block to finish
dispatch_semaphore_wait(sema, DISPATCH_TIME_FOREVER);
// critical section code ends
}
I believe, due to #synchronized block the completion handler is called on the same thread as the caller which results in a deadlock. Is that right? If yes, how else can this be achieved?
Thanks!
I would avoid #synchronized and use a serial dispatch queue to serialize the work in the critical section. You can still use a semaphore to block the serial dispatch queue until the asynchronous operation is complete.
The serial dispatch queue guarantees that only one block of code can enter the critical section at a time and there is no risk of blocking the main queue.
You will need to create the serial queue during your class initialisation
#property (strong,nonatomic) dispatch_queue_t serialQueue;
- (id)init {
if (self = [super init]) {
self.serialQueue = dispatch_queue_create("com.example.CriticalTaskQueue", NULL);
}
return self;
}
- submitSomeCriticalWork() {
dispatch_async(self.serialQueue, ^{
dispatch_semaphore_t sema = dispatch_semaphore_create(0);
[self someMethodWithCompletionHandler:^(BOOL result) {
// execute completion handler
dispatch_semaphore_signal(sema);
}];
// wait for completion block to finish
dispatch_semaphore_wait(sema, DISPATCH_TIME_FOREVER);
// critical section code ends
}];
}
Related
I have a method that is being called several times. But I need to act only in the very last method call. I tried dispatch_async but didn't work because still is been queue the calls:
-(void)doingSomething:(NSString*)someValue
{
dispatch_async(dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0), ^{
[NSThread sleepForTimeInterval:1.0f];
dispatch_async(dispatch_get_main_queue(), ^{
// do something with the last call
});
});
}
Any of you knows a way to queue the calls and only use the very last call?
I'll really appreciate your help
My suggestion would be to use a dispatch_group. Call dispatch_group_enter before you call dispatch_async, and call dispatch_group_leave and the end of the block that's executed by dispatch_async. Then, after you've enqueued all the blocks, use dispatch_group_notify to schedule the completion block, which will run after all the other dispatch_async blocks have finished.
dispatch_group_t group = dispatch_group_create();
for (...) {
dispatch_group_enter(group);
dispatch_async(dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0), ^{
...
dispatch_group_leave(group);
});
}
dispatch_group_notify(group, dispatch_get_main_queue(), ^{
// This gets called when all the other blocks have finished
});
Alternatively, you could use an NSOperationQueue instead of libdispatch, and make a completion operation which lists every other operation as a dependency. This does have the disadvantage that the completion operation won't be executed on the main queue, though.
NSOperation *completionOperation = [NSBlockOperation blockOperationWithBlock:^{
// This gets called when all the other operations have finished
}];
for (...) {
NSOperation *operation = [NSBlockOperation blockOperationWithBlock:^{
...
}];
[completionOperation addDependency:operation];
[operationQueue addOperation:operation];
}
[operationQueue addOperation:completionOperation];
I am creating a serial queue in which i add two task as shown below
dispatch_queue_t serial = dispatch_queue_create("com.apple.serial", DISPATCH_QUEUE_SERIAL);
**//Task 1**
dispatch_async(serial, ^{
[NMUserAPIManager getUserProfileData:^(NMUser *objUser) {
NSLog(#"Get User Profile .....");
_objUser = objUser;
}];
});
**//Task 2**
dispatch_async(serial, ^{
[NMUserAPIManager getUserRecentTransactionData:^(NSDictionary *responseDictionary) {
_accountTableView.hidden = NO;
[self recentTransactionSetup:responseDictionary];
NSLog(#"Get User Recent transaction");
dispatch_async(dispatch_get_main_queue(), ^{
[self reloadTableData];
});
}];
});
Inside that two task i am calling web service with NSURLSession. Problem is that before my Task 1 completion handle Task2 completion handle get called. According to theory by using serial queue each task waits for the previous task to finish before being executed. It my understanding is correct.
NSURLSession's already run on a background thread, so the issue you are seeing here is that as far as your serial queue is concerned once you call 'getUserProfileData:' technically the work for that block in your queue is finished because the NSURLSession is running on a different thread. If your main goal here is to simply call your second task after your first one completes I don't think you need your own queue you would probably be better off simply doing something like:
[NMUserAPIManager getUserProfileData:^(NMUser *objUser) {
NSLog(#"Get User Profile .....");
_objUser = objUser;
[self getUserTransactions];
}];
-(void)getUserTransactions
{
[NMUserAPIManager getUserRecentTransactionData:^(NSDictionary *responseDictionary) {
_accountTableView.hidden = NO;
[self recentTransactionSetup:responseDictionary];
NSLog(#"Get User Recent transaction");
dispatch_async(dispatch_get_main_queue(), ^{
[self reloadTableData];
});
}];
}
EDIT:
If you are looking for something a little more robust I would check out this post for how you can subclass NSOperation to make your own Asynchronous Operation which you can then use with an NSOperationQueue.
For example I have a method with three async blocks. Each block result is needed to perform next block to achieve final methods result. So, what I'm looking for is a nice GCD strategy to make'em perform in a strict order and without dead locks
__block id task1Result;
__block id task2Result;
__block id finalResult;
[self startTask1:^(id result) { task1Result = result }]
[self startTask2:task1Result block:^(id result) { task2Result = result }]
[self startTask3:task2Result block:^(id result) { finalResult = result }]
UPD. I have found a solution:
dispatch_semaphore_t sem = dispatch_semaphore_create(0);
__block id task1Result;
__block id task2Result;
__block id finalResult;
[self startTask1:^(id result) {
task1Result = result;
dispatch_semaphore_signal(sem);
}];
dispatch_semaphore_wait(sem, DISPATCH_TIME_FOREVER);
[self startTask2:task1Result block:^(id result) {
task2Result = result;
dispatch_semaphore_signal(sem);
}];
dispatch_semaphore_wait(sem, DISPATCH_TIME_FOREVER);
[self startTask3:task2Result block:^(id result) { finalResult = result }];
But in my case I faced a problem with some library method which brings app to deadlock. ><
Create a serial dispatch queue like described here:
https://developer.apple.com/library/ios/documentation/General/Conceptual/ConcurrencyProgrammingGuide/OperationQueues/OperationQueues.html
In a nutshell:
dispatch_queue_t queue;
queue = dispatch_queue_create("com.example.MyQueue", NULL);
dispatch_async(queue, ^{
printf("Do some work here.\n");
});
dispatch_async(queue, ^{
printf("When finished do next task.\n");
});
Be aware that you have to handle the queue yourself.
If each task directly consumes the result of the previous task, can’t you start each one from the completion callback of its predecessor? You’ll still need a dispatch group to wait for the last task to complete, though.
dispatch_group_t group = dispatch_group_create();
__block id result;
dispatch_group_enter(group);
[self startTask1:^(id task1Result) {
[self startTask2:task1Result block:^(id task2Result) {
[self startTask3:task2Result block:^(id finalResult) {
result = finalResult;
dispatch_group_leave(group);
}];
}];
}];
dispatch_group_wait(group, DISPATCH_TIME_FOREVER);
One complication you may run into is whether attempting to enqueue tasks from a completion handler runs the risk of deadlock, i.e. if your completion handlers are invoked on the same serial queue that handles enqueueing tasks.
I have developed the following method, which checks the app's ability to communicate with the server.
The method performs a simple query and knows that if it gets a result, the app should be connected (basic ping mechanism).
- (BOOL)isAppConnected
{
__block BOOL isConnected = NO;
dispatch_semaphore_t semaphore = dispatch_semaphore_create(0);
[[SFRestAPI sharedInstance] performSOQLQuery:#"SELECT id FROM Account LIMIT 1"
failBlock:^(NSError *e) {
isConnected = NO;
NSLog(#"NOT CONNECTED %#", e);
NSLog(#"fail block ON THE MAIN THREAD? %hhd", [NSThread isMainThread]);
dispatch_semaphore_signal(semaphore);
} completeBlock:^(NSDictionary *dict) {
isConnected = YES;
NSLog(#"%#", dict);
NSLog(#"complete block ON THE MAIN THREAD? %hhd", [NSThread isMainThread]);
dispatch_semaphore_signal(semaphore);
}];
// if the wait times-out we will receive a non-zero result and can assume no connection to SF
//When using: DISPATCH_TIME_FOREVER the app hangs forever!!
int waitResult = dispatch_semaphore_wait(semaphore, 30 * NSEC_PER_SEC);
NSLog(#"waitResult: %d", waitResult);
return isConnected;
}
I am using the 'dispatch_semaphore_wait' as suggested in the Apple documentation
My goal is to wait on the response or a short timeout to figure out if we really have a valid connection.
With the code above, 'dispatch_semaphore_wait' never actually waits, i.e. execution does not stop at that line but it continues immediately (always returning 49 as the result to the dispatch_semaphore_wait call). That is unless I use DISPATCH_TIME_FOREVER in which case the app hangs forever...
At the moment I am calling this method from the main thread. I am aware that this is a bad idea, but I wanted to see this working as expected before refactoring.
What could be causing this behaviour?
Thanks.
The parameter for dispatch_semaphore_wait is not a delay, but the time when the semaphore should wake up. Yours will wake up 30 seconds after Midnight, Jan 1st. 1970 (or 2001, not sure). Use the dispatch_time function.
- (BOOL)isAppConnected
{
__block BOOL isConnected = NO;
dispatch_semaphore_t semaphore = dispatch_semaphore_create(0);
// Add this code...
dispatch_async(dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT, 0), ^{
[[SFRestAPI sharedInstance] performSOQLQuery:#"SELECT id FROM Account LIMIT 1"
failBlock:^(NSError *e) {
isConnected = NO;
NSLog(#"NOT CONNECTED %#", e);
NSLog(#"fail block ON THE MAIN THREAD? %hhd", [NSThread isMainThread]);
dispatch_semaphore_signal(semaphore);
} completeBlock:^(NSDictionary *dict) {
isConnected = YES;
NSLog(#"%#", dict);
NSLog(#"complete block ON THE MAIN THREAD? %hhd", [NSThread isMainThread]);
dispatch_semaphore_signal(semaphore);
}];
});
int waitResult = dispatch_semaphore_wait(semaphore, DISPATCH_TIME_FOREVER);
NSLog(#"waitResult: %d", waitResult);
return isConnected;
}
First of all if your fail and complete block are invoked on the main thread then you are going to wait forever, or until the timeout you specify 'times out'
The reason is because the main thread starts waiting after you call dispatch_semaphore_wait. And then if your performSOQLQuery calls the blocks on the main thread nothing will happen until the time out 'times out'.
Now of coarse if you specify a time out of forever the semaphore will never signal or let go, which means your main thread will wait forever, for itself.
Change the waiting code to this: never make the main thread wait
dispatch_async(dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_HIGH, 0), ^{
int waitResult = dispatch_semaphore_wait(semaphore, DISPATCH_TIME_FOREVER);
NSLog(#"waitResult: %d", waitResult);
});
And do not return a Bool as you are doing, since this is a lengthy operation, you want to use a block that has a result.
Also make sure that your performSOQLQuery() method is not on the main thread please.
This is essentially what I'm doing to run an asynchronous method synchronously:
This essentially works when called once, but when called multiple times, it will eventually stay inside the while loop and never get signaled. Any ideas on how to set a timer to eventually time out after sometime?
__block SomeClass *result = nil;
dispatch_semaphore_t semaphore = dispatch_semaphore_create(0);
dispatch_queue_t queue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_LOW, 0UL);
dispatch_async(queue, ^{
[[SomeManager sharedInstance] someMethodWithCallback:^(id responseObject, NSError *error) {
if (!error) {
result = (SomeClass *)ResponseObject;
}
dispatch_semaphore_signal(semaphore);
}];
});
// wait with a time limit
while (dispatch_semaphore_wait(semaphore, DISPATCH_TIME_NOW)) {
[[NSRunLoop currentRunLoop] runMode:NSDefaultRunLoopMode beforeDate:[NSDate dateWithTimeIntervalSinceNow:0]];
}
dispatch_release(semaphore);
Thanks
That looks kind of like GCD abuse to me. ;) Are you running the run loop because this is executing on the main thread? Why not just use a dispatch_async() from your completion handler to invoke a handler on the main thread? eg:
- (void)handleDataReady: (id) results error: (NSError *) error {
// update your app
}
- (void)performAsyncUpdate {
dispatch_queue_t queue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_LOW, 0UL);
dispatch_async(queue, ^{
[[SomeManager sharedInstance] someMethodWithCallback:^(id responseObject, NSError *error) {
dispatch_async(dispatch_get_main_queue(), ^{
[self handleDataReady:responseObject error:error];
}];
});
}
If you really want to make it synchronous, i.e. blocking the calling thread until the operation completes then use the following pattern (of course you want to avoid blocking threads if possible)
NSCondition *waitCondtion = [NSCondition new];
__block BOOL completed = NO;
dispatch_queue_t queue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_LOW, 0UL);
dispatch_async(queue, ^{
[[SomeManager sharedInstance] someMethodWithCallback:^(id responseObject, NSError *error) {
if (!error) {
result = (SomeClass *)ResponseObject;
}
[waitCondtion lock];
completed = YES;
[waitCondition signal];
[waitCondition unlock];
}];
});
[waitCondtion lock];
if (!completed)
[waitCondtion wait];
[waitCondition unlock];
You can also use "waitUntilDate:" to timeout the wait after a period.
However, this pattern only works as long as the "someMethodWithCallback does not call its callback block on the same thread that is being blocked. I have copied your code because it is not obvious how "someMethodWithCallback" is implemented. Since this method is using an asynchronous pattern, then it must be doing something asynchronously therefore why are you calling it inside a dispatch_async? What thread will it call its callback block on?
You should "fill" the completion handler with whatever code you require to process the result when the completion handler finished (and also completely removing that run loop).
In order to "abort" an asynchronous operation, you should provide a cancel message which you send the asynchronous result provider.
In your case, since you have a singleton, the cancel message would have to be send like this:
[[SomeManager sharedInstance] cancel];
When the operation receives the cancel message, it should as soon as possible abort its task and call the completion handler with an appropriate NSError object indicating that it has been cancelled.
Note, that cancel messages may be asynchronous - that means, when it returns, the receiver may still execute the task.
You may achieve a "timeout" with setting up a timer, which sends the cancel message the operation, unless it has been invalidated when the operation finished.