FastCGI socket and multiple threads about cutelyst

[dantti]

interesting, I'll try to reproduce, does it also fails on client side or is it just this message?

[dantti]

Ok, managed to reproduce, and as a workaround you can use TCP sockets which worked fine. Not sure yet why it happens, but once that message is out it stops working 😭

[buschmann23]

Yes. Same for me. After that message, nothing happens any more. Apache than returns a 502 error. Currently I am using the FastCGI socket with multiple processes, that works fine. :)

[dantti]

Ok, this seems to be a complex issue, I dunno why in earlier versions it worked, but be it threads or process it seems that when a thread/processes gets the connection and the other one get's an invalid value which end up on that message, which I'm not sure yet how makes QLocalServer delete it's socket file. Odd that I recall this working when I first added the UNIX socket feature... ATM I'd recommend using TCP ...

[buschmann23]

Hm, when I try it with multiple processes I get no error, but after a while only one process of six seems to be active and reacts to requests. (tested with wieghttp and siege)

[dantti]

Ok, it's a Qt bug, and indeed it can happen with process too.
Basically if accept fails due the other thread/process accepting it then it will close the server and delete the socket file, then that message is emitted because the FD QSocketNotifier was watching is now invalid, since I run Cutelyst on AWS with a single core CPU and using a single process this issue didn't hit me.

I can implement the experimental-thread-balance for local sockets which will fix the issue for threaded apps...

I'll be opening a bug report and eventually I'll fix it myself since my QtNetwork bugs usually stay there for a while.