This is simply programming error, but it is interesting that using
libstdc++ does not reveal this error. With clang++-libc++, we got
std::system_error: mutex lock faild: Invalid argument. This is
because we did not give a name to lock object, so it is immediately
destructed. I think this will fix the reported crash on Mac OSX.
We may run into race condition if execve is called at the same time
when fcntl is called. But we just does this for now to compile
nghttp2 applications under older kernel.
This also means that at least one stream whose dpri is
NGHTTP2_STREAM_DPRI_TOP exists, its siblings descendants have no
chance to send streams, even if their parent stream has
NGHTTP2_STREAM_DPRI_NODATA.
tiny-nghttpd is HTTP/2 server and its purpose is measure the
performance of nghttp2 library code. Currently it only accepts direct
HTTP/2 connection only.
Previously when nghttp2_stream_resume_deferred_data() is called,
deferred flags in stream->flags are all cleared. This is not ideal
because if application returned NGHTTP2_ERR_DEFERRED, and also that
stream is deferred by flow control, then all flags are cleared and
read callback will be invoked again. This commit fixes this issue.
This changes error condition of nghttp2_session_resume_data().
Previously we return error if stream was deferred by flow control.
Now we don't return error in this case. We just clear
NGHTTP2_FLAG_DEFERRED_USER and if still
NGHTTP2_FLAG_DEFERRED_FLOW_CONTROL is set, just return 0.
Previously when connection level remote flow control window gets 0, we
mark the stream having DATA frame with
NGHTTP2_STREAM_FLAG_DEFERRED_FLOW_CONTROL. When connection level
WINDOW_UPDATE is received, we checks all existing streams, including
closed ones, and call nghttp2_stream_resume_deferred_data(). The
profiler shows this is expensive.
Now we prepare dedicated priority queue for DATA frames. And we don't
mark stream with NGHTTP2_STREAM_FLAG_DEFERRED_FLOW_CONTROL when DATA
cannot be sent solely due to connection level flow control. Instead,
we just queue DATA item to queue. We won't pop DATA item from queue
when connection level remote window size is 0. This way, we avoid the
expensive operation for all streams when WINDOW_UPDATE is arrived.
Tatsuhiro Tsujikawa
Svante Signell