Previously, in server side, we used closed streams to detect the error
that the misbehaving client sends a frame on the incoming stream it
explicitly closed. With this commit, we make a further step, and
detect one more error case. Since we retain closed streams as long as
the sum of its size and the number of opened streams are equal or less
than max concurrent streams, we can safely say that if we get a frame
which is sent on the stream that is not found in either closed or
opened stream, it is already closed or has not existed. Then we can
send GOAWAY.
The previous code shrinks closed streams when we closed another
stream, but now it is removed. It is enough to adjust closed streams
when new incoming stream is created.
While creating this commit, we noticed that
NGHTTP2_INITIAL_MAX_CONCURRENT_STREAMS is defined as INT32_MAX. But
since SETTINGS can contain value up to UINT32_MAX, it is not enough.
However, since the stream ID space is limited to INT32_MAX, it is high
enough. We could keep this value, but this time we deprecate
NGHTTP2_INITIAL_MAX_CONCURRENT_STREAMS macro. While it is in public
header, the effect of deprecating it is negligible because of the
reason we wrote above, and usually application sets much smaller value
(say, 100) as SETTINGS_MAX_CONCURRENT_STREAMS.
https://tools.ietf.org/html/rfc7540#section-5.3.3 explains how to
transform dependency tree to avoid circular dependency. Previously,
we wrongly always moved the dependent stream under the root stream.
The correct destination is the parent stream of the stream to
reprioritize. This commit fixes this bug.
We have a code to call error callback when invalid header is received
and it is treated as stream error. But we didn't if the incoming
header is invalid, but just ignored. This generosity is required to
handle public Internet connections especially when nghttp2 is used as
forward proxy.
This commit also fixes the bug that old configuration is still used
for worker process. The another bug fix is that inherited, but not
used fd is not closed in worker process. That makes reloading next
configuration fail if it contains the address which are leaked into
worker process.
This commit implements configuration reloading with SIGHUP.
There are rough edges left:
* Rename SignalServer with more meaningful name, say, WorkerProcess.
* We should introduce global configuration object which is not
affected by configuration reloading. It should hold cmdcfgs, argc,
argv, and last worker PID.
* We should close the listener file descriptor when some operation was
failed after that.
The N-th -p options sets N-th URI's weight. If the number of -p is
less than the number of URIs, the last -p value is repeated. If no -p
is set, use default weight, 16.
clang + libc++ does not work, it still requires libc++_shared.so
runtime even if -lstdc++ is used, which supposed to link with static
version of libc++.
Previously, we didn't retry request on connection failure. Sometimes
we hit the edge case where connection is about to lost just when we
write request. To avoid this situation, we now retry request to
failed attempt. We also add ConnectBlocker to MemcachedConnection not
to attempt to connect to memcached if connection could not be made
previously.
Wenfeng Liu
Wenfeng Liu
Tatsuhiro Tsujikawa