1. Recently we have updated our receive window size to a larger value.
Unfortunately, this also results in auto-test pumping through
more data, which probably takes more time on CI.
At the moment we do not have any public API on QNAM's level to
customize HTTP/2 parameters (aka 5.10/FF and so on). So we use the fact
that QNAM is QObject and we can set a property on it. This property
is our Http2::ProtocolParameters object that allows us to configure:
- HPACK parameters (in 5.10 - noop)
- session receive window size
- different SETTINGS as described by RFC 7540, 6.5.2.
2. Undocumented environment variable to set ENABLE_PUSH is not needed
anymore.
3. In 5.11 Http2::ProtocolParameter will become a public API
and we'll introduce a new setter in QNAM.
Change-Id: If08fd5e09e7c0b61cf9700b426b60b5837b6b2e6
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
We were using the default ones, provided by RFC7540. It appears they are way
too restrictive and conservative: when downloading something relatively big,
a stream keeps spending the whole session/its own 'recv' windows and thus
we have to constantly send WINDOW_UPDATE frames. This significantly slows
down our HTTP/2 implementation, making it orders of magnitude slower than
HTTP/1.1. To fix this:
- We send SETTINGS_INITIAL_WINDOW_SIZE in the first SETTINGS frame
to inform our peer that per-stream WINDOW is bigger than 64Kb
- We increase the session's receive window size.
Task-number: QTBUG-63722
Change-Id: I31312fcfd5f0fc0aee6aaa5d3562cc7d1b931adc
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Without TLS (and thus ALPN/NPN negotiation) HTTP/2 requires
a protocol upgrade procedure, as described in RFC 7540, 3.2.
We start as HTTP/1.1 (and thus we create QHttpProtocolHandler first),
augmenting the headers we send with 'Upgrade: h2c'. In case
we receive HTTP/1.1 response with status code 101 ('Switching
Protocols'), we continue as HTTP/2 session, creating QHttp2ProtocolHandler
and pretending the first request we sent was HTTP/2 request
on a real HTTP/2 stream. If the first response is something different
from 101, we continue as HTTP/1.1. This change also required
auto-test update: our toy-server now has to respond to
the initial HTTP/1.1 request on a platform without ALPN/NPN.
As a bonus a subtle flakyness in 'goaway' auto-test went
away (well, it was fixed).
[ChangeLog][QtNetwork][HTTP/2] In case of clear text HTTP/2 we
now initiate a required protocol upgrade procedure instead of
'H2Direct' connection.
Task-number: QTBUG-61397
Change-Id: I573fa304fdaf661490159037dc47775d97c8ea5b
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Previously we were always setting SpdyWasUsedAttribute for SPDY/HTTP/2/HTTP/1.1
(true/false) which is confusing. Now if HTTP2AllowedAttribute was set to true on
a request, we set HTTP2WasUsedAttribute. Otherwise, as we did before, we're setting
SpdyWasUsedAttribute.
Change-Id: I0c44cfb5469fef0c12719baa951197ee2accee4a
Reviewed-by: Markus Goetz (Woboq GmbH) <markus@woboq.com>
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
After it started to fail (somehow it's only OpenSUSE 42.1) again and again
and after a quick re-evaluation it appears the logic testing SETTINGS|ACK
is incorrect. We (client side) start by sending the preface and then
continue to send our request(s). The other side (server) starts from sending its
SETTINGS frame. These settings must be ACKed, but apparently it can happen,
that server receives a requests and sends a reply before it receives SETTINGS|ACK,
resulting in replyFinished (replyFinishedWithError) signal and event loop stopping.
As a result - QVERIFY(serverGotSettingsACK) fails.
Task-number: QTBUG-58758
Change-Id: I8184cf459b2b88f70c646171e0115c184237fad1
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
- Fix the case when we erroneously handled stream ID == 0 in a GOAWAY frame as
an invalid stream ID.
- _q_receivedReply: convert do{}while() loop into to while(){} to prevent
it from handling any frames after GOAWAY frame received and all active frame
finished.
- sendRequest - if we received GOAWAY, also clear spdyRequests in the connection
channel, otherwise it keeps re-trying to send requests!
- Http network connection channel never resets a protocolHandler in _q_encrypted/
_q_connected, which is BAD for HTTP/2, since HTTP/2 has unique per-connection
compression context and must be reset - now we recreate the protocol handler in
_q_encrypted or _q_connected (https/http).
- Update autotest.
Task-number: QTBUG-57600
Change-Id: Ib864ce52287bab23334ff43a83ba4b0b7cb52c60
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
HTTP/2 allows a server to pre-emptively send (or "push") responses (along
with corresponding "promised" requests) to a client in association with a
previous client-initiated request. This can be useful when the server
knows the client will need to have those responses available in order
to fully process the response to the original request.
Server push is semantically equivalent to a server responding to a request;
however, in this case, that request is also sent by the server, as a
PUSH_PROMISE frame.
The PUSH_PROMISE frame includes a header block that contains a complete set
of request header fields that the server attributes to the request.
After sending the PUSH_PROMISE frame, the server can begin delivering the
pushed response as a response on a server-initiated stream that uses the
promised stream identifier.
This patch:
- fixes the HPACK decompression of PUSH_PROMISE frames;
- allows a user to enable PUSH_PROMISE;
- processes and caches pushed data for promised streams;
- updates auto-test - emulates a simple PUSH_PROMISE
scenario.
Change-Id: Ic4850863a5e3895320baac3871a723fc091b4aca
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
HTTP/2 does not require TLS connection, it can work in a cleartext mode.
Plus at the moment only OpenSSL backend allows HTTP/2 negotiation
via ALPN/NPN (and none of our CI configurations with OpenSSL supports
these extensions, rendering HTTP/2 auto-test useless). This patch
implements cleartext HTTP/2 ('h2c') in 'direct' mode - this is
allowed if a client has a prior knowledge that HTTP/2 is supported by
a server.
Change-Id: I4978775e9732c40bc77f549b83bb4a5d1761887e
Reviewed-by: Alex Trotsenko <alex1973tr@gmail.com>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Add autotest for QHttp2ProtocolHandler. This patch contains a very simplistic
"in-process HTTP2 server" for testing the protocol's basic logic/flow control/error
handling and emulating possible scenarios.
Task-number: QTBUG-50956
Change-Id: Ie02d3329c5182277a3c7c84f1bae8d02308e945d
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
