Hello, Maybe there is a race condition between suspending/resuming connection. This happens only if I use "poll()" - not with "epoll()". Requests were created using "wget" (http 1.0) - one after the other.
What I did: The MHD-Thread calls "MHD_suspend_connection()" inside "MHD_ContentReaderCallback()". The worker thread calls "MHD_resume_connection()" if the connection is suspended. Sometimes the MHD thread sleeps forever - it does not continue to work on resume connections. The polling-variable "num_connections" is zero. The "connection_suspended"-flag and calls to "MHD_resume/suspend_connection()" are protected by a mutex. The callstack from the core-dump: #0 0x00007f38889c512d in poll () at ../sysdeps/unix/syscall-template.S:81 #1 0x00000000006786c6 in MHD_poll_all (daemon=0x359a930, may_block=1) at /home/markus/Website/gloriatv/dev/src/contrib/microhttpd/daemon.c:2485 #2 0x0000000000678ba9 in MHD_poll (daemon=0x359a930, may_block=1) at /home/markus/Website/gloriatv/dev/src/contrib/microhttpd/daemon.c:2642 #3 0x0000000000679557 in MHD_select_thread (cls=0x359a930) at /home/markus/Website/gloriatv/dev/src/contrib/microhttpd/daemon.c:2928 #4 0x00007f388a758182 in start_thread (arg=0x7f38846fb700) at pthread_create.c:312 #5 0x00007f38889d247d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111 Locals on frame #1: i = 0 timeout = -1 poll_server = 2 poll_listen = 0 tmp = 114 'r' ltimeout = 301000 poll_pipe = 1 p = 0x7f387c0008c0 num_connections = 0 pos = 0x0 next = 0x0 Thanks a lot Markus
