This wraps epoll and io_uring the other way: previously, we had the
io_uring file descriptor registered in epoll and when it became ready,
we could query its completion ring. The problem is that the poll
wakeups cause considerable overhead in the Linux kernel, see
https://lore.kernel.org/io-uring/20250128133927.3989681-9-max.kellermann@ionos.com/
By wrapping epoll inside io_uring using
io_uring_prep_poll_multishot(), the "outer" loop is io_uring and
inside it, we have epoll. This adds another system call for epoll,
but that will go away as soon as most operations are going through
io_uring. Previously, io_uring had this extra system call.
This is important for operations that have complex cancellation
procedures (e.g. if they need to free buffers). They might leave an
Operation instance in the queue.
The output gets the standard MPD log format with domain curl and a timestamp.
Using CURLOPT_DEBUGFUNCTION that is only called when CURLOPT_VERBOSE is in effect
when MPD log level is verbose.
The BlockingCall() in InitUringInputPlugin() did not work because the
EventThread was not yet started. This was never noticed until commit
e309941646 which enabled `IORING_SETUP_SINGLE_ISSUER`, and suddenly
the kernel refused to accept io_uring_submit() calls from the
io_thread because io_uring_setup() had been called from the main
thread.
Once a NFS request fails with NFS4ERR_EXPIRED, the whole connection is
broken and we need to open a new one. I wish libnfs would report this
to us as a connection-level error, but instead only the one request
fails; therefore this workaround is an ugly kludge.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/2175
The resume/seek was received asynchronously and meanwhile an error
might have occurred that needs to be handled.
This fixes another NFS-related crash bug.
libFLAC tries to keep on seeking a stream even after a (fatal) read
error has occurred. Let the DecoderClient decide whether proceeding
is possible.
This fixes a crash bug when playing a file over NFS and the NFS
connection fails.
When running with stdout output to debug the server
or misbehaving clients, it is useful to have the timestamp
to detect timing issues and response times.
Especially when opening and playing online sources that
block the connection thread sometime for a significantly
long time that makes the client-server unresponsive
and cause timeouts in clients.
