I am using Game Center turn-based matches for my card game. It feels like a good fit because people want to check their email or write a text sometimes between turns. Some users have said that they do really want to be able to have a very asynchronous playing experience.
For the people that are keeping the game open between their turns, I want to update the screen to reflect things that the other players are doing on their turns. I have set up a listener on GKLocalPlayer that responds to player:receivedTurnEventForMatch:didBecomeActive. The documentation says that this will get called when match data is saved by another player even if it doesn't become the player's turn (the player on the current device). That doesn't appear to be true 100% of the time. In fact, it appears that it only gets called about 1 in 3 times that match data is saved by other players. It seems more reliable when it becomes the player's turn, but even that isn't 100% reliable.
I am using saveCurrentTurnWithMatchData:completionHandler: on GKTurnBasedMatch to save match data that doesn't end the current player's turn and I'm calling endTurnWithNextParticipants:turnTimeout:matchData:completionHandler: on GKTurnBasedMatch to save the data when it does end the current player's turn. There are a few scenarios where I want to call saveCurrentTurn… with updated matchData. In my game, you can have computers playing in your multiplayer game as well. So, a human player may play a card and then a computer may play a card before that GKPlayer's turn ends. There are also scenarios where an individual player may play twice. (eg. A player plays the last card on a trick. That player takes the trick and gets to lead the next trick.)
I have set up a ton of logging around this and I can see clear scenarios where one device calls saveCurrentTurn… and the completionHandler is called without an error and the other device doesn't get notified with a call to player:receivedTurnEvent… I have also added logging to verify that each time I'm calling saveCurrentTurn… that I'm calling it with new matchData. I'm not making redundant calls.
If I go to the device that didn't get the updated matchData and force it to load the matchData for the match again, it gets the updated data. So, it's definitely getting saved.
I have tried throttling the calls to saveCurrentTurn… so that they don't happen in immediate succession and that didn't help.
Both devices in my testing are running iOS 8.4. There appears to have been an issue in iOS 8.3 that is fixed now (see this question). This Apple forum post also reports this issue 2 years ago and it appears that bug reports were filed and marked fixed.
Has anyone else seen this? I would love to know that I'm doing something wrong. Any ideas are very welcome.
I do something similar. In my game, each player has multiple pieces, saving the match when each piece moves so that other players--if they're in the game--can watch what's happening in real time. Like you describe, the Game Center messaging is almost completely useless.
As you referenced, in 8.3, the "end of turn" messages were completely broken. As of 8.4, they happen most, but not all, of the time. As you're seeing, the "match has been saved" notifications are also erratic. Here are some tips I've used to increase the success rate:
Slow down the saves. If you save too fast, only the last one arrives at the recipient. I set up an NSArray queue, and each time I want to save the match, I add the new matchData to that queue. I have a timer loop running that does the actual saveCurrentTurnWithMatchData, pops the item off the stack if the save was successful, and then sets up a new timer to call itself again a little later. I'm using 2 second intervals which seems to be working well.
Append each new piece of data, don't overwrite. Put a sequence number on each piece of data. So, if you save seq numbers 1, 2, 3 and 4, but the recipient only receives a notice for #4, the records for 1, 2 and 3 are there in the match object. The recipient needs to track the last record it read, and then iterate through any new records from that point when it receives an updated matchData.
I also use the queue's NSArray writeToFile: function to maintain a list of the pending saves. If the user exits the game before the queue is flushed, I reload the queue NSArray from disk at the next startup
Note that even with this mechanism, the notifications to the recipient are erratic. Generally speaking, they arrive in batches of 4+. Then nothing happens until 3 or 4 more saves happen, which again all show up together. Making 1 save and letting the game sit for 10 minutes will probably never generate a notice on the recipient's machine. But, if you save 4 or 6 times in a row, all of them tend to show up in a burst.
Sometimes, the notifications just stop for a few hours. Not sure if this a sandbox flaw or a game-center-in-general flaw. There are no failures of any kind, the messages just stop working for a while. Sometimes, the next morning, they show up in a burst. Sometimes not. In the end, I've stopped relying on the notifications. I set up another timer loop to continuously download the match. It checks if it has become my turn or not, it checks if new updates have been added to the matchData. Then calls player:receivedTurnEventForMatch:didBecomeActive. As far a receivedTurnEventForMatch: knows, it was launched because of an event and it merrily goes on about its business.
It does seem that saving the match is pretty prompt. If you don't get an error, it seems pretty certain that the updated match is immediately available for other players to consume... they just need to know to consume it. The messaging framework, though, has to be viewed as completely unreliable and non-guaranteed. Hence, the timer loop to continuously poll the match.
Edit: arguably, once I implemented #2, #1 shouldn't really matter. Any notification received by the recipient will trigger reading all new records in the data. But, this "hardening" has evolved over the past few months as I wrestle with Game Center's shortcomings. I just haven't gotten around to removing #1.