Summary:
This diff is part of larger change to switch to using `folly::ObserverContainer` (introduced in D27062840).
This diff:
- Adds methods to `QuicSocket` to support adding and removing `ObserverContainer` observers. By default, `QuicSocket` implementations will not support observers as `QuicSocket::getSocketObserverContainer` returns `nullptr`, and this will cause `QuicSocket::addObserver` and `QuicSocket::removeObserver` to fail gracefully.
- Adds support in `QuicClientTransport` and `QuicServerTransport` for `ObserverContainer` observers by overriding `getSocketObserverContainer`. Each implementation creates an `ObserverContainer` on construction that is destroyed on destruction.
- Adds the `close` event to the new observer interface and adds corresponding support in `QuicTransportBase`. This allows both the old and new observers to receive the close event. The next diff will do a hard cutover of the rest of the events.
Differential Revision: D35000960
fbshipit-source-id: e75807c97f4385532bf36244591a259aa0a2f4cc
Summary:
This implements a basic keepalive timer at the transport layer.
The basic idea is to leverage the same conditions as the idle timer. The keepalive timer will fire ~15% before the idle timer and trigger a PING frame. This has the effect of keeping the connection open if the PING is acknowledged.
Reviewed By: jbeshay
Differential Revision: D34256174
fbshipit-source-id: a8cee579f5b09bdfa23bcdc7cf4c70de299de3bd
Summary: The two lists of transport knobs in two different locations keep getting out of sync. This consolidates them into one location which is QuicConstants.
Reviewed By: mjoras
Differential Revision: D34227259
fbshipit-source-id: 26b20620f4ef85f8887e7294512155060a057839
Summary: Add new connection end cb option to client/server transport
Reviewed By: mjoras
Differential Revision: D34190871
fbshipit-source-id: 375cf95bc8dbe4ab759140cfb4b33b732869953f
Summary:
Add an experimental setting in the TokenlessPacer to use an adjustable burstSize in the to compensate for delays in pacer loop timer with a limit of 5 loop intervals.
This practically restores some concept of tokens in the pacer. This should help with pacer timer delays observed in relatively high BDP links.
To help quantify how often the delays happen, the QuicStats pacer_lagged counter is updated every time the pacer is delayed by more than 10% when the connection is not application limited.
In a test configuration across a 150 Mbps link with 150 ms round-trip delay, the experimental setting achieves ~25% higher bandwidth.
Reviewed By: bschlinker, mjoras
Differential Revision: D33035430
fbshipit-source-id: 4aa02821d609facf6c830647459319d44c6fc3bd
Summary:
Maine change is `MockConnectionCallback` -> `MockConnectionSetupCallback` + `MockConnectionCallbackNew`.
Everything else is changing tests to use the two new classes.
Differential Revision: D33076321
fbshipit-source-id: a938b63ce59f07f549b3e725caad8785348db7ed
Summary:
Experiment with using adaptive thresholds for detecting packet losses through reordering or timeouts. When transportSettings.useAdaptiveLossThresholds is set to true, every time a spurious loss is detected, the thresholds will be updated to avoid marking similar packets as losses. In cases with a high degree of packet reordering, this improves bandwidth utilization significantly.
The useAdaptiveLossThresholds setting can be set using a new transport knob.
This was verified by transferring a 5 MB object across 50 Mbps link with high random packet reordering (emulated using netem)
```
+ tc -s qdisc replace dev veth7 root handle 1:0 netem rate 50Mbit limit 1250
+ tc -s qdisc add dev veth7 parent 1:0 handle 2: netem delay 15ms 5ms distribution normal limit 1250
```
Without adaptive thresholds, ~43% of sent bytes are retransmissions. Goodput is 2 Mbps.
With adaptive thresholds, ~3% of sent bytes are retransmissions. Goodput is ~17.5 Mbps.
Reviewed By: mjoras
Differential Revision: D33354584
fbshipit-source-id: 390bfb3398d499fcdf97a5cfd5654da3eeabed17
Summary:
Adds paddingModulo to QuicPacketBuilder, which pads up short header packets so that the length remaining of a packet is a multiple of paddingModulo
ie a message of length 25 and max packet size of 40 with a paddingModulo of 8 will get 7 padding frames to a total length of 32. (and a message of length 26 will get 6 padding frames to also go up to a total length of 32)
Reviewed By: mjoras
Differential Revision: D32858254
fbshipit-source-id: 99ada01108429db9f9bba1e342daf99e80385179
Summary: - logging number of bytes the server sent during the handshake for insight.
Reviewed By: mjoras
Differential Revision: D33069800
fbshipit-source-id: e7e8f25183ee30de99e2971bae3c6b93882f6e63
Summary:
Add experimental CUBIC's hystart delay bounds that ensure CUBIC does exit Hystart too early. Current limits of 2,8 us are too short.
On a 100 Mbps link with 150 ms round-trip delay, this improved CUBIC's throughput from ~14 Mbps to ~80 Mbps.
Reviewed By: mjoras
Differential Revision: D33035403
fbshipit-source-id: bdeb081111f8053bc7fb01002316d991d8bac2e0
Summary:
This implements a global (per process) limit on unfinished handshakes from unverified source addresses.
This limits the ability of an attacker to create connection state without also allocating connection state themselves. By default the limit is 1024.
Reviewed By: kvtsoy
Differential Revision: D32772165
fbshipit-source-id: 6c195169377a9f687c54bc9782cc58fe085e1275
Summary: Otherwise we can end up in a situation where the non-DSR scheduler was limited by the loop time while the DSR scheduler is not, leading to an inconsistency.
Reviewed By: lnicco
Differential Revision: D32570027
fbshipit-source-id: f43d2517589c22bac0f2bb76626cc55c2a21fa5d
Summary:
- Change the knob param handler to receive a Transport rather than a ConnectionState.
- Add a new transport knob that can set the priority threshold and target utilization factor for automatically controlling background mode based upon active streams.
Reviewed By: mjoras
Differential Revision: D31847477
fbshipit-source-id: 4cfeff04b9fe60eb25ab484f460e252ef6970646
Summary:
Experiment with a less aggressive variant of BBR that could be used for background requests in order to improve responsiveness of foreground requests.
BBR's agressiveness can be controlled through a new transport knob CC_AGRESSIVENESS_KNOB(0xccab) which accepts values between 25-100. The value indicates the percentage of the available BW the congestion controller should try to use, leaving headroom for other flows.
Aggressiveness of BBR is reduced by adjusting BBR's startup gain during the STARTUP phase. During the ProbeBW phase, BBR uses a longer pacing gain progression with values that intentionally underutilize the available BW.
Reviewed By: bschlinker, mjoras
Differential Revision: D31219364
fbshipit-source-id: c67ffcf03d27fd4d41be2d58f00f4f960c7646c7
Summary:
This diff introduces split callbacks to QUIC.
Old: callback:
* `ConnectionCallback`
Two new callbacks:
* `ConnectionSetupCallback`
* `ConnectionStreamsCallback`
For now the idea is that both old and new callbacks will live side by side and mvfst will decide which callbacks scheme to use via the `useSplitConnectionCallbacks` flag supplied on both QUIC client and server creation.
Reviewed By: mjoras
Differential Revision: D30407129
fbshipit-source-id: 7b726c2652ebe00b47f44bf291cffb9d9f6c884b
Summary: As in title. This isn't actually very risky to have on, and is actually confusing if you try to use the APIs without the flag since nothing will happen.
Reviewed By: shodoco
Differential Revision: D30585491
fbshipit-source-id: d79598ab5c723675d63cb71f011b9107acda6d7d
Summary:
This change just adds a (currently no-op) flag that will be used in diffs up the stack.
The idea here is that I'll add split QUIC connection callback interfaces that will live side by side with existing single monolithic callback for now. We will experiment with split callbacks on small scale to see that there is no regressions and then will phase out the old callback gradually.
This flag is to control which callback(s) to use.
Reviewed By: mjoras
Differential Revision: D30399667
fbshipit-source-id: 8fc4e4a005e93cf6d48a987f49edee33b90dbbf1
Summary:
These are either no longer relevant, are unlikely to be done, or are spculative enough that they don't deserve code space.
Hope here is to make our search for TODOs higher signal.
Reviewed By: lnicco
Differential Revision: D29769792
fbshipit-source-id: 7cfa62cdc15e72d8b7b0cd5dbb5913ea3ca3dc5a
Summary: Add a new transport knob (MAX_PACING_RATE_KNOB) to allow the client to suggest a limit on the pacing rate used by the server, if it has pacing enabled.
Reviewed By: mjoras
Differential Revision: D29335469
fbshipit-source-id: 8f0a881e8235e0d787ea86ba1e268a36eecf6151
Summary:
This function actually assumes both are there. This mostly happens in unit tests anyway.
(Note: this ignores all push blocking failures!)
Reviewed By: yangchi
Differential Revision: D29187689
fbshipit-source-id: a7ca516c18de4f0261ffec985f1d4e9f63b071d9
Summary:
Add a new transport knob `NOTSENT_BUFFER_SIZE_KNOB` that will allow us to run quick experiments for different buffer sizes on the server side.
## Background
In TCP, the send buffer accounts for both the unacked and unsent data. SO_SNDBUF is the sum of both and is statically configured for the socket. TCP_NOTSENT_LOWAT only limits the unsent data which still allows the SO_SNDBUF to still accommodate a big BDP.
In mvfst, the buffer usage we currently track (sumCurStreamBufferLen) is only for unsent data. Its size is controlled by the TransportSetting totalBufferSpaceAvailable. Unacked packets are tracked as outstanding packets and do not count toward that buffer space. Effectively, a limit on this buffer achieves the same effect as TCP's NOTSENT_LOWAT socket option.
Reviewed By: mjoras, yangchi
Differential Revision: D28843244
fbshipit-source-id: 3b61619038f3b41508fe76c2873e41486ae97739
Summary:
add 3 new transport knobs:
1. set congestion control algorithm (string algorithm name)
2. set pacing rtt factor to be used during "startup" phase of congestion control, right now only used by bbr and copa
3. set pacing rtt factor to be used by default during congestion control, right now only used by bbr and copa
Reviewed By: yangchi
Differential Revision: D28462540
fbshipit-source-id: 2c861885e472cb2bce0b51daeef2b143e9628080
Summary: make space for backend subdir
Reviewed By: mjoras
Differential Revision: D27999777
fbshipit-source-id: 1d270b8ce3349980a1e4c68035a1f39c095237b5
Summary:
This diff hooks the DSR write function into QuicServerTransport's
write path.
Reviewed By: mjoras
Differential Revision: D27890711
fbshipit-source-id: ac4452373c0baafe091f93fb54fccf87be604a9c
Summary: as title, this is a testing only API
Reviewed By: mjoras
Differential Revision: D27746039
fbshipit-source-id: cf87836810a4579f622152ccb17aca49a0d605e3
Summary:
as title. This also moves a FizzCryptoTestFactory from FizzCryptoFactoryTest to TestUtils so that it can be used in other test code
This change has an unfortunate side-effect that cryptoFactory_ in both client and server will be moved from stack to heap.
Reviewed By: mjoras
Differential Revision: D27264488
fbshipit-source-id: febc307fb02cb136d58fe70bee648d35431acff0
Summary: This seems to cause some issues with CGNAT type networks where the client is actually using v6 or v4.
Reviewed By: yangchi
Differential Revision: D27772485
fbshipit-source-id: ac118441caad38301f2a22e657cefb398a5210da
Summary:
As before we will now aggressively send probes on all spaces with probes available when the PTO timer fires.
This time with more unit tests and some bug fixes.
Reviewed By: yangchi
Differential Revision: D27338523
fbshipit-source-id: 8a9ccb90ed691e996fab4afa2f132c0f99044fbc
Summary: As in title. There's a bug here somewhere with empty write loops we need to find.
Reviewed By: yangchi
Differential Revision: D27279100
fbshipit-source-id: e1d26fbf8d6df1590d464a6504a8b940b46794e0
Summary:
Previously we would only send probes for the first space which had one available, i.e. Initial before Handshake before AppData. Since we only have one PTO timer this can lead to situations where we perpetually probe with only Initials, which can significantly delay the handshake if we should have probed with Handshakes.
With this diff we will keep the single PTO timer but aggressively write more probes from all spaces if they are available.
Additionally this refactors some counters into EnumArrays
Reviewed By: yangchi
Differential Revision: D27235199
fbshipit-source-id: ef3614a833bf0f02f5806846a1335fa7ac2a4dc8
Summary:
Before the change, there's no good way to recreate Cubic CC instance with custom CC factory, because Cubic is created by default.
On client side this requires calling setCongestionControl() or setTransportSettings() after calling setCongestionControllerFactory(), which is normally the case.
Reviewed By: yangchi
Differential Revision: D26401996
fbshipit-source-id: dfda39be835c67b9db42f726b3ac64c7b3d37c2f
Summary: Adding QuiConnectionStats to client transport. Moving getConnectionsStats() logic from server worker into transport base class.
Reviewed By: lnicco
Differential Revision: D26316635
fbshipit-source-id: a384eee5d1bc7b23d908e0b03fafcc4ee962b0b9
Summary:
Given the large number of callbacks that are being triggered from the Observer
this change makes it possible to enable through a simple config, just the
subset of callbacks that a consumer is interested in receiving.
Observer and Socket Lifecycle callbacks are enabled by default, they are not
configurable.
Reviewed By: bschlinker
Differential Revision: D25879382
fbshipit-source-id: abe79ed92e958ddc96475c347f8ec7204400cdfa
Summary:
We were using the LifecycleObserver and InstrumentationObserver classes
separately, to generate and receive callbacks.
This change migrates both these to use the unified Observer callback class and
adjusts the unit tests.
Reviewed By: bschlinker
Differential Revision: D25845845
fbshipit-source-id: c489400f5d70bccadbcc1d957136c5ade36b65ff
