🐌 [Performance] Don't spawn new threads in every game loop, just offload it into a ThreadPool

[0blu]

🐌 Performance

Currently a new thread is created when MapManager::Update invokes CreateNewInstancesForPlayers.

core/src/game/Maps/MapManager.cpp

Line 339 in 645a677

std::thread instanceCreationThread = std::thread(&MapManager::CreateNewInstancesForPlayers, this);

Thread creation should be avoided. Especially that often and for such a small workload.
Its better to off load it into a ThreadPool which accepts small tasks like this and processes them with a pre-existing thread.

[Wall-core]

good change, as I've had (very rare) cases where TBB malloc fails when creating a new thread for this purpose and crashes (without crashing mangos)

[0blu]

Interesting that no one catched this earlier.
Creating native system thread is super workload heavy.

I think this if one of the smaller things that I will tackle after native branch.
Right now there is no ThreadPool for small tasks in mangos.
The class that is called ThreadPool is a more like a BroadcastThreadPool which executes the same workload on different threads.

Edit: Okay tests have shown, its not that slow (~0.07ms creation time, ~0.20ms until thread actually starts) but we should keep this in mind for the future. (every microsecond counts when dealing with the game loop)

[Stoabrogga]

I run Vmangos in GDB and it spammed my log file with thousands of "New Thread" messages:

log entries

Restoring deleted items to players ...

>> Restored 0 previously deleted items.
[New Thread 0x7fffc7ee4640 (LWP 4643)]
[New Thread 0x7fffc76e3640 (LWP 4644)]

==========================================================
Current content is set to Patch 1.12: Drums of War.
Supported client build is set to 5875.
==========================================================

World initialized.
SERVER STARTUP TIME: 0 minutes 4 seconds
[New Thread 0x7fffc6ee2640 (LWP 4645)]
[New Thread 0x7fffc66e1640 (LWP 4646)]

>> Only expired mails (need to be return or delete) or DB table `mail` is empty.
[New Thread 0x7fffc5ee0640 (LWP 4647)]
[New Thread 0x7fffc56df640 (LWP 4648)]

mangos>[New Thread 0x7fffc4ede640 (LWP 4649)]
Max allowed socket connections 1048576
[New Thread 0x7fffbcffd640 (LWP 4650)]
[New Thread 0x7fffabfff640 (LWP 4651)]
[New Thread 0x7fffab7fe640 (LWP 4652)]
[New Thread 0x7fffaaffd640 (LWP 4653)]
[Thread 0x7fffabfff640 (LWP 4651) exited]
[New Thread 0x7fffabfff640 (LWP 4654)]
[Thread 0x7fffabfff640 (LWP 4654) exited]
[New Thread 0x7fffabfff640 (LWP 4655)]
[Thread 0x7fffabfff640 (LWP 4655) exited]
[New Thread 0x7fffabfff640 (LWP 4656)]
[Thread 0x7fffabfff640 (LWP 4656) exited]
[New Thread 0x7fffabfff640 (LWP 4657)]
[Thread 0x7fffabfff640 (LWP 4657) exited]
[New Thread 0x7fffabfff640 (LWP 4658)]
[Thread 0x7fffabfff640 (LWP 4658) exited]
[New Thread 0x7fffabfff640 (LWP 4659)]
[Thread 0x7fffabfff640 (LWP 4659) exited]
[New Thread 0x7fffabfff640 (LWP 4660)]
[Thread 0x7fffabfff640 (LWP 4660) exited]
[New Thread 0x7fffabfff640 (LWP 4661)]
[Thread 0x7fffabfff640 (LWP 4661) exited]
[New Thread 0x7fffabfff640 (LWP 4662)]
[Thread 0x7fffabfff640 (LWP 4662) exited]
[New Thread 0x7fffabfff640 (LWP 4663)]
[Thread 0x7fffabfff640 (LWP 4663) exited]

Is this the same issue? I had to disable thread-event logging by adding this GDB parameter:

-ex "set print thread-events off"

[0blu]

Is this the same issue? I had to disable thread-event logging by adding this GDB parameter:

Yes, this is also the way how I discovered it.
I was working with thread_static and realized that the constructor is called for every game tick.