DATA_BLOCKED frames should be sent in the same as the STREAM frames that
resulted in the connection becoming blocked on connection flow control.
If there's not enough space left in that packet, the DATA_BLOCKED frame
is sent in the next packet.
* refactor the framer to pack both control and STREAM frames
* refactor framer STREAM frame packing logic
* pack STREAM_DATA_BLOCKED in the same packet as the STREAM frame
This makes debugging easier (and is slightly more efficient). In the
pathological case where there is not enough space remaning in the packet
to pack the STREAM_DATA_BLOCKED frame, it is queued for the next packet.
* add an integration test
Instead of transferring data in one direction, we should echo the data
sent. This makes sure that a roughly equal number of packets is sent in
each direction.
The Go timer resolution on Windows leaves a lot to be desired. By
bumping the timeout duration the test will take longer, but at least it
won't be flaky.
* use require in benchmark tests
* translate the QLOGDIR test
* translate handshake tests
* translate the handshake RTT tests
* translate the early data test
* translate the MTU tests
* translate the key update test
* translate the stateless reset tests
* translate the packetization test
* translate the close test
* translate the resumption test
* translate the tracer test
* translate the connection ID length test
* translate the RTT tests
* translate the multiplexing tests
* translate the drop tests
* translate the handshake drop tests
* translate the 0-RTT tests
* translate the hotswap test
* translate the stream test
* translate the unidirectional stream test
* translate the timeout tests
* translate the MITM test
* rewrite the datagram tests
* translate the cancellation tests
* translate the deadline tests
* translate the test helpers
* enable DPLPMTUD on macOS dual-stack sockets
https://datatracker.ietf.org/doc/draft-seemann-tsvwg-udp-fragmentation/
contains details on how IP fragmentation is handled on different
platforms.
* only enable DF on macOS Sequoia (and newer) dual-stack sockets
* fix macOS version numbers
* fix comment in MTU integration test
* skip dual-stack test on old macOS versions
* http3: rename RoundTripper to Transport
* http3: rename SingleDestinationRoundTripper to ClientConn
* http3: construct the ClientConn via Transport.NewClientConn
* send goaway when server is shutting down
* http3: track next stream ID instead of last stream ID for GOAWAYs
* refactor the graceful shutdown integration tests
* http3: improve GOAWAY frame parsing tests
* http3: simplify server graceful shutdown logic
* http3: simplify parsing of GOAWAY frames
* http3: clean up initialization of server contexts
* http3: fix race condition in graceful shutdown logic
---------
Co-authored-by: WeidiDeng <weidi_deng@icloud.com>
* http3: implement server idle timeout support
This update introduces the ability for an HTTP/3 server to enforce an idle
timeout on connections. This timeout will trigger when no new requests are
received on a connection, irrespective of any PING frames received at the
QUIC level.
* fix deadlock when http3 idle timeout is not enabled
* fix typo
* Switch to a more efficient implementation
* Avoid a goroutine
* Avoid constent re-adjusting of a timer
* Works with hijacked streams
* Generalize the idle timeout description
* Add an integration test for http server idle timeout
* Attempt to fix other tests impacted by the new idle timeout test
* pass a context to Transport.ConnContext
This context is cancelled when the QUIC connection is closed, or when
the QUIC handshake fails. This allows the application to easily build
and garbage collect a map of active connections.
* correctly handle fresh contexts returned from ConnContext
