Add support for QUIC upstream #14829

alyssawilk · 2021-01-26T16:11:47Z

As we close in on #2557 we want to start thinking about upstream support for QUIC.

The main use case may be for mobile (envoyproxy/envoy-mobile#1033) but it could be used in general for forward proxying and cross-data center hops.

v1 would be hard-coded QUIC with no failover, but for use on the internet it would eventually need to include things like QUIC vs TCP racing, 0-rtt credential caching, alt-scv cache, etc.

Only adding explicit (hard-configured, or downstream-initiated) HTTP/3. Getting Auto for UDP/TCP is going to take substantially more work. HTTP/3 config will be rejected initially to keep this PR simple as possible. Risk Level: Low (unused, hidden) Testing: new unit tests Docs Changes: n/a Release Notes: n/a Part of #14829 Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

davinci26 · 2021-02-19T02:18:39Z

Feel free to tag me for any windows specific (test or otherwise) issue

This is working end to end for a number of the protocol integration tests, but still needs a bunch of work before it's production ready (I triaged some of the excludes but not all, watermarks not working etc). I think given this works for the happy path it's worth checking in, and I can iterate on getting individual tests working and doing the TODOs. Risk Level: low (minor core refactors, major changes are all behind hidden config) Testing: integration tests, some unit tests Docs Changes: n/a Release Notes: n/a #14829 Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

Enabling (most of the) the downstream tests for quic upstream for more coverage. Enabling cookie tests given Dan's fix Fixing a couple of tests with filters assuming headers-come-with-fin Defaulting header size to Envoy defaults. Risk Level: low (quic / test code) Testing: yes Docs Changes: n/a Release Notes: n/a part of #14829 Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

As quic connections are created early, we can't set header limits in waitForHttpConnection Part of #14829 Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

Alas I didn't end up moving all of the protocol options over, as the cluster has its own timeout params so most of the QuicProtocolOptions are listener-specific. Commit Message: adding quic protocol config to quic upstream Risk Level: n/a (quic only) Testing: unit tests Docs Changes: n/a Release Notes: n/a part of #14829 Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

Tagging a bunch of quic test failures with more details. Adapting the status code tests to QUIC and fixing QUIC code to validate. minor refactors to core code. Risk Level: low (minor core refactors) Testing: new integration tests Docs Changes: n/a Release Notes: n/a part of #14829 Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

Respecting upstream and downstream caps on number of headers for HTTP/3 Risk Level: n/a (http/3) Testing: turned up integration tests Docs Changes: n/a Release Notes: n/a #14829 among others. Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

Commit Message: Add initial stream and connection level flow control windows to envoy.config.core.v3.QuicProtocolOptions which is be used in QUIC listener config and Http3 upstream cluster config. Risk Level: low Testing: re-enable more Http3 downstream protocol test. Part of #2557 #12930 #14829 Signed-off-by: Dan Zhang <danzh@google.com> Co-authored-by: Dan Zhang <danzh@google.com>

Risk Level: n/a Testing: yes! Docs Changes: n/a Release Notes: n/a #14829 among others Signed-off-by: Alyssa Wilk <alyssar@chromium.org>

Respecting upstream and downstream caps on number of headers for HTTP/3 Risk Level: n/a (http/3) Testing: turned up integration tests Docs Changes: n/a Release Notes: n/a envoyproxy#14829 among others. Signed-off-by: Alyssa Wilk <alyssar@chromium.org> Signed-off-by: Gokul Nair <gnair@twitter.com>

Commit Message: Add initial stream and connection level flow control windows to envoy.config.core.v3.QuicProtocolOptions which is be used in QUIC listener config and Http3 upstream cluster config. Risk Level: low Testing: re-enable more Http3 downstream protocol test. Part of envoyproxy#2557 envoyproxy#12930 envoyproxy#14829 Signed-off-by: Dan Zhang <danzh@google.com> Co-authored-by: Dan Zhang <danzh@google.com> Signed-off-by: Gokul Nair <gnair@twitter.com>

Risk Level: n/a Testing: yes! Docs Changes: n/a Release Notes: n/a envoyproxy#14829 among others Signed-off-by: Alyssa Wilk <alyssar@chromium.org> Signed-off-by: Gokul Nair <gnair@twitter.com>

Risk Level: n/a (hidden by default) Testing: e2e tests Docs Changes: n/a Release Notes: n/a Part of envoyproxy#14829 Signed-off-by: Gokul Nair <gnair@twitter.com>

Commit Message: use max header size in HCM config to initialize quic session. Additional Description: change QuicSession::Initialize() and filter chain creation order to set max header list size for each session before Initialize(). Move Initialize() of Http3 connection from CodecClient to CodecClientProd. Added CodecClient::connect() interface. Change EnvoyQuicConnection not to directly inherit from QuicConnection. And renamed it. Fix a use-after-free bug in FakeUpstream which causes ASAN failure. Risk Level: low Testing: more protocol H3 tests Part of envoyproxy#2557 envoyproxy#12930 envoyproxy#14829 Signed-off-by: Dan Zhang <danzh@google.com> Signed-off-by: Gokul Nair <gnair@twitter.com>

envoyproxy#16128) Commit Message: use posted callback to block/unblock quic stream in EnvoyQuicStream::readDisable() with weak_ptr to stream to handle stream life time issue. Additional Description: currently if readDisabled() is called in decodeHeaders|Trailers(), the stream will be blocked right away which breaks the assumption QUICHE has that a stream shouldn't be blocked during OnInitialHeadersComplete() and OnTrailingHeadersComplete(). This change makes the stream state change completely outside of QUICHE call stack. (The unblocking is already outside of QUICHE call stack in existing implementation.) Also simplify the blockage state change logic. Risk Level: low Testing: added more unit tests and enabled Http2UpstreamIntegrationTest::ManyLargeSimultaneousRequestWithBufferLimits which was flaky. Part of envoyproxy#2557 envoyproxy#14829 Signed-off-by: Dan Zhang <danzh@google.com> Co-authored-by: Dan Zhang <danzh@google.com> Signed-off-by: Gokul Nair <gnair@twitter.com>