1
0
Fork 0
mirror of https://github.com/denoland/deno.git synced 2024-11-24 15:19:26 -05:00
denoland-deno/ext/http/service.rs
Nathan Whitaker 930ccf928a
perf(ext/http): Reduce size of ResponseBytesInner (#24840)
I noticed
[`set_response_body`](ce42f82b5a/ext/http/service.rs (L439-L443))
was unexpectedly hot in profiles, with most of the time being spent in
`memmove`.

It turns out that `ResponseBytesInner` was _massive_ (5624 bytes), so
every time we moved a `ResponseBytesInner` (for instance in
`set_response_body`) we were doing a >5kb memmove, which adds up pretty
quickly.

This PR boxes the two larger variants (the compression streams),
shrinking `ResponseBytesInner` to a reasonable 48 bytes.

---
  Benchmarked with a simple hello world server:
```ts
// hello-server.ts
Deno.serve((_req) => {
  return new Response("Hello world");
});
// run with `deno run -A hello-server.ts`
// in separate terminal `wrk -d 10s http://127.0.0.1:8000`
```

Main:
```
Running 10s test @ http://127.0.0.1:8000/
  2 threads and 10 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    53.39us    9.53us   0.98ms   92.78%
    Req/Sec    86.57k     3.56k   91.58k    91.09%
  1739319 requests in 10.10s, 248.81MB read
Requests/sec: 172220.92
Transfer/sec:     24.64MB
```

This PR:
```
Running 10s test @ http://127.0.0.1:8000/
  2 threads and 10 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    45.44us    8.49us   0.91ms   90.04%
    Req/Sec   100.65k     2.26k  102.65k    96.53%
  2022296 requests in 10.10s, 289.29MB read
Requests/sec: 200226.20
Transfer/sec:     28.64MB
```

So a nice ~15% bump. (With response body compression, the gain is ~10%
for gzip and neutral for brotli)
2024-08-02 00:30:26 +00:00

692 lines
20 KiB
Rust

// Copyright 2018-2024 the Deno authors. All rights reserved. MIT license.
use crate::request_properties::HttpConnectionProperties;
use crate::response_body::ResponseBytesInner;
use crate::response_body::ResponseStreamResult;
use deno_core::error::AnyError;
use deno_core::futures::ready;
use deno_core::BufView;
use deno_core::OpState;
use deno_core::ResourceId;
use http::request::Parts;
use hyper::body::Body;
use hyper::body::Frame;
use hyper::body::Incoming;
use hyper::body::SizeHint;
use hyper::header::HeaderMap;
use hyper::upgrade::OnUpgrade;
use scopeguard::guard;
use scopeguard::ScopeGuard;
use std::cell::Cell;
use std::cell::Ref;
use std::cell::RefCell;
use std::cell::RefMut;
use std::future::Future;
use std::mem::ManuallyDrop;
use std::pin::Pin;
use std::rc::Rc;
use std::task::Context;
use std::task::Poll;
use std::task::Waker;
pub type Request = hyper::Request<Incoming>;
pub type Response = hyper::Response<HttpRecordResponse>;
#[cfg(feature = "__http_tracing")]
pub static RECORD_COUNT: std::sync::atomic::AtomicUsize =
std::sync::atomic::AtomicUsize::new(0);
macro_rules! http_general_trace {
($($args:expr),*) => {
#[cfg(feature = "__http_tracing")]
{
let count = $crate::service::RECORD_COUNT
.load(std::sync::atomic::Ordering::SeqCst);
println!(
"HTTP [+{count}]: {}",
format!($($args),*),
);
}
};
}
macro_rules! http_trace {
($record:expr $(, $args:expr)*) => {
#[cfg(feature = "__http_tracing")]
{
let count = $crate::service::RECORD_COUNT
.load(std::sync::atomic::Ordering::SeqCst);
println!(
"HTTP [+{count}] id={:p} strong={}: {}",
$record,
std::rc::Rc::strong_count(&$record),
format!($($args),*),
);
}
};
}
pub(crate) use http_general_trace;
pub(crate) use http_trace;
pub(crate) struct HttpServerStateInner {
pool: Vec<(Rc<HttpRecord>, HeaderMap)>,
}
/// A signalling version of `Rc` that allows one to poll for when all other references
/// to the `Rc` have been dropped.
#[repr(transparent)]
pub(crate) struct SignallingRc<T>(Rc<(T, Cell<Option<Waker>>)>);
impl<T> SignallingRc<T> {
#[inline]
pub fn new(t: T) -> Self {
Self(Rc::new((t, Default::default())))
}
#[inline]
pub fn strong_count(&self) -> usize {
Rc::strong_count(&self.0)
}
/// Resolves when this is the only remaining reference.
#[inline]
pub fn poll_complete(&self, cx: &mut Context<'_>) -> Poll<()> {
if Rc::strong_count(&self.0) == 1 {
Poll::Ready(())
} else {
self.0 .1.set(Some(cx.waker().clone()));
Poll::Pending
}
}
}
impl<T> Clone for SignallingRc<T> {
#[inline]
fn clone(&self) -> Self {
Self(self.0.clone())
}
}
impl<T> Drop for SignallingRc<T> {
#[inline]
fn drop(&mut self) {
// Trigger the waker iff the refcount is about to become 1.
if Rc::strong_count(&self.0) == 2 {
if let Some(waker) = self.0 .1.take() {
waker.wake();
}
}
}
}
impl<T> std::ops::Deref for SignallingRc<T> {
type Target = T;
#[inline]
fn deref(&self) -> &Self::Target {
&self.0 .0
}
}
pub(crate) struct HttpServerState(RefCell<HttpServerStateInner>);
impl HttpServerState {
pub fn new() -> SignallingRc<Self> {
SignallingRc::new(Self(RefCell::new(HttpServerStateInner {
pool: Vec::new(),
})))
}
}
impl std::ops::Deref for HttpServerState {
type Target = RefCell<HttpServerStateInner>;
fn deref(&self) -> &Self::Target {
&self.0
}
}
enum RequestBodyState {
Incoming(Incoming),
Resource(#[allow(dead_code)] HttpRequestBodyAutocloser),
}
impl From<Incoming> for RequestBodyState {
fn from(value: Incoming) -> Self {
RequestBodyState::Incoming(value)
}
}
/// Ensures that the request body closes itself when no longer needed.
pub struct HttpRequestBodyAutocloser(ResourceId, Rc<RefCell<OpState>>);
impl HttpRequestBodyAutocloser {
pub fn new(res: ResourceId, op_state: Rc<RefCell<OpState>>) -> Self {
Self(res, op_state)
}
}
impl Drop for HttpRequestBodyAutocloser {
fn drop(&mut self) {
if let Ok(res) = self.1.borrow_mut().resource_table.take_any(self.0) {
res.close();
}
}
}
pub(crate) async fn handle_request(
request: Request,
request_info: HttpConnectionProperties,
server_state: SignallingRc<HttpServerState>, // Keep server alive for duration of this future.
tx: tokio::sync::mpsc::Sender<Rc<HttpRecord>>,
) -> Result<Response, hyper_v014::Error> {
// If the underlying TCP connection is closed, this future will be dropped
// and execution could stop at any await point.
// The HttpRecord must live until JavaScript is done processing so is wrapped
// in an Rc. The guard ensures unneeded resources are freed at cancellation.
let guarded_record = guard(
HttpRecord::new(request, request_info, server_state),
HttpRecord::cancel,
);
// Clone HttpRecord and send to JavaScript for processing.
// Safe to unwrap as channel receiver is never closed.
tx.send(guarded_record.clone()).await.unwrap();
// Wait for JavaScript handler to return request.
http_trace!(*guarded_record, "handle_request response_ready.await");
guarded_record.response_ready().await;
// Defuse the guard. Must not await after this point.
let record = ScopeGuard::into_inner(guarded_record);
http_trace!(record, "handle_request complete");
let response = record.into_response();
Ok(response)
}
struct HttpRecordInner {
server_state: SignallingRc<HttpServerState>,
request_info: HttpConnectionProperties,
request_parts: http::request::Parts,
request_body: Option<RequestBodyState>,
response_parts: Option<http::response::Parts>,
response_ready: bool,
response_waker: Option<Waker>,
response_body: ResponseBytesInner,
response_body_finished: bool,
response_body_waker: Option<Waker>,
trailers: Option<HeaderMap>,
been_dropped: bool,
finished: bool,
needs_close_after_finish: bool,
}
pub struct HttpRecord(RefCell<Option<HttpRecordInner>>);
#[cfg(feature = "__http_tracing")]
impl Drop for HttpRecord {
fn drop(&mut self) {
RECORD_COUNT
.fetch_sub(1, std::sync::atomic::Ordering::SeqCst)
.checked_sub(1)
.expect("Count went below zero");
http_general_trace!("HttpRecord::drop");
}
}
impl HttpRecord {
fn new(
request: Request,
request_info: HttpConnectionProperties,
server_state: SignallingRc<HttpServerState>,
) -> Rc<Self> {
let (request_parts, request_body) = request.into_parts();
let request_body = Some(request_body.into());
let (mut response_parts, _) = http::Response::new(()).into_parts();
let record =
if let Some((record, headers)) = server_state.borrow_mut().pool.pop() {
response_parts.headers = headers;
http_trace!(record, "HttpRecord::reuse");
record
} else {
#[cfg(feature = "__http_tracing")]
{
RECORD_COUNT.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
}
#[allow(clippy::let_and_return)]
let record = Rc::new(Self(RefCell::new(None)));
http_trace!(record, "HttpRecord::new");
record
};
*record.0.borrow_mut() = Some(HttpRecordInner {
server_state,
request_info,
request_parts,
request_body,
response_parts: Some(response_parts),
response_ready: false,
response_waker: None,
response_body: ResponseBytesInner::Empty,
response_body_finished: false,
response_body_waker: None,
trailers: None,
been_dropped: false,
finished: false,
needs_close_after_finish: false,
});
record
}
fn finish(self: Rc<Self>) {
http_trace!(self, "HttpRecord::finish");
let mut inner = self.self_mut();
inner.response_body_finished = true;
let response_body_waker = inner.response_body_waker.take();
let needs_close_after_finish = inner.needs_close_after_finish;
drop(inner);
if let Some(waker) = response_body_waker {
waker.wake();
}
if !needs_close_after_finish {
self.recycle();
}
}
pub fn close_after_finish(self: Rc<Self>) {
debug_assert!(self.self_ref().needs_close_after_finish);
let mut inner = self.self_mut();
inner.needs_close_after_finish = false;
if !inner.finished {
drop(inner);
self.recycle();
}
}
pub fn needs_close_after_finish(&self) -> RefMut<'_, bool> {
RefMut::map(self.self_mut(), |inner| &mut inner.needs_close_after_finish)
}
fn recycle(self: Rc<Self>) {
assert!(
Rc::strong_count(&self) == 1,
"HTTP state error: Expected to be last strong reference"
);
let HttpRecordInner {
server_state,
request_parts: Parts { mut headers, .. },
..
} = self.0.borrow_mut().take().unwrap();
let inflight = server_state.strong_count();
http_trace!(self, "HttpRecord::recycle inflight={}", inflight);
// Keep a buffer of allocations on hand to be reused by incoming requests.
// Estimated target size is 16 + 1/8 the number of inflight requests.
let target = 16 + (inflight >> 3);
let pool = &mut server_state.borrow_mut().pool;
if target > pool.len() {
headers.clear();
pool.push((self, headers));
} else if target < pool.len() - 8 {
pool.truncate(target);
}
}
fn self_ref(&self) -> Ref<'_, HttpRecordInner> {
Ref::map(self.0.borrow(), |option| option.as_ref().unwrap())
}
fn self_mut(&self) -> RefMut<'_, HttpRecordInner> {
RefMut::map(self.0.borrow_mut(), |option| option.as_mut().unwrap())
}
/// Perform the Hyper upgrade on this record.
pub fn upgrade(&self) -> Result<OnUpgrade, AnyError> {
// Manually perform the upgrade. We're peeking into hyper's underlying machinery here a bit
self
.self_mut()
.request_parts
.extensions
.remove::<OnUpgrade>()
.ok_or_else(|| AnyError::msg("upgrade unavailable"))
}
/// Take the Hyper body from this record.
pub fn take_request_body(&self) -> Option<Incoming> {
let body_holder = &mut self.self_mut().request_body;
let body = body_holder.take();
match body {
Some(RequestBodyState::Incoming(body)) => Some(body),
x => {
*body_holder = x;
None
}
}
}
/// Replace the request body with a resource ID and the OpState we'll need to shut it down.
/// We cannot keep just the resource itself, as JS code might be reading from the resource ID
/// to generate the response data (requiring us to keep it in the resource table).
pub fn put_resource(&self, res: HttpRequestBodyAutocloser) {
self.self_mut().request_body = Some(RequestBodyState::Resource(res));
}
/// Cleanup resources not needed after the future is dropped.
fn cancel(self: Rc<Self>) {
http_trace!(self, "HttpRecord::cancel");
let mut inner = self.self_mut();
if inner.response_ready {
// Future dropped between wake() and async fn resuming.
drop(inner);
self.finish();
return;
}
inner.been_dropped = true;
// The request body might include actual resources.
inner.request_body.take();
}
/// Complete this record, potentially expunging it if it is fully complete (ie: cancelled as well).
pub fn complete(self: Rc<Self>) {
http_trace!(self, "HttpRecord::complete");
let mut inner = self.self_mut();
assert!(
!inner.response_ready,
"HTTP state error: Entry has already been completed"
);
if inner.been_dropped {
drop(inner);
self.finish();
return;
}
inner.response_ready = true;
if let Some(waker) = inner.response_waker.take() {
drop(inner);
waker.wake();
}
}
fn take_response_body(&self) -> ResponseBytesInner {
let mut inner = self.self_mut();
debug_assert!(
!matches!(inner.response_body, ResponseBytesInner::Done),
"HTTP state error: response body already complete"
);
std::mem::replace(&mut inner.response_body, ResponseBytesInner::Done)
}
/// Has the future for this record been dropped? ie, has the underlying TCP connection
/// been closed?
pub fn cancelled(&self) -> bool {
self.self_ref().been_dropped
}
/// Get a mutable reference to the response status and headers.
pub fn response_parts(&self) -> RefMut<'_, http::response::Parts> {
RefMut::map(self.self_mut(), |inner| {
inner.response_parts.as_mut().unwrap()
})
}
/// Get a mutable reference to the trailers.
pub fn trailers(&self) -> RefMut<'_, Option<HeaderMap>> {
RefMut::map(self.self_mut(), |inner| &mut inner.trailers)
}
pub fn set_response_body(&self, response_body: ResponseBytesInner) {
let mut inner = self.self_mut();
debug_assert!(matches!(inner.response_body, ResponseBytesInner::Empty));
inner.response_body = response_body;
}
/// Take the response.
fn into_response(self: Rc<Self>) -> Response {
let parts = self.self_mut().response_parts.take().unwrap();
let body = HttpRecordResponse(ManuallyDrop::new(self));
Response::from_parts(parts, body)
}
/// Get a reference to the connection properties.
pub fn request_info(&self) -> Ref<'_, HttpConnectionProperties> {
Ref::map(self.self_ref(), |inner| &inner.request_info)
}
/// Get a reference to the request parts.
pub fn request_parts(&self) -> Ref<'_, Parts> {
Ref::map(self.self_ref(), |inner| &inner.request_parts)
}
/// Resolves when response head is ready.
fn response_ready(&self) -> impl Future<Output = ()> + '_ {
struct HttpRecordReady<'a>(&'a HttpRecord);
impl<'a> Future for HttpRecordReady<'a> {
type Output = ();
fn poll(
self: Pin<&mut Self>,
cx: &mut Context<'_>,
) -> Poll<Self::Output> {
let mut mut_self = self.0.self_mut();
if mut_self.response_ready {
return Poll::Ready(());
}
mut_self.response_waker = Some(cx.waker().clone());
Poll::Pending
}
}
HttpRecordReady(self)
}
/// Resolves when response body has finished streaming. Returns true if the
/// response completed.
pub fn response_body_finished(&self) -> impl Future<Output = bool> + '_ {
struct HttpRecordFinished<'a>(&'a HttpRecord);
impl<'a> Future for HttpRecordFinished<'a> {
type Output = bool;
fn poll(
self: Pin<&mut Self>,
cx: &mut Context<'_>,
) -> Poll<Self::Output> {
let mut mut_self = self.0.self_mut();
if mut_self.response_body_finished {
// If we sent the response body and the trailers, this body completed successfully
return Poll::Ready(
mut_self.response_body.is_complete() && mut_self.trailers.is_none(),
);
}
mut_self.response_body_waker = Some(cx.waker().clone());
Poll::Pending
}
}
HttpRecordFinished(self)
}
}
#[repr(transparent)]
pub struct HttpRecordResponse(ManuallyDrop<Rc<HttpRecord>>);
impl Body for HttpRecordResponse {
type Data = BufView;
type Error = AnyError;
fn poll_frame(
self: Pin<&mut Self>,
cx: &mut Context<'_>,
) -> Poll<Option<Result<Frame<Self::Data>, Self::Error>>> {
use crate::response_body::PollFrame;
let record = &self.0;
let res = loop {
let mut inner = record.self_mut();
let res = match &mut inner.response_body {
ResponseBytesInner::Done | ResponseBytesInner::Empty => {
if let Some(trailers) = inner.trailers.take() {
return Poll::Ready(Some(Ok(Frame::trailers(trailers))));
}
unreachable!()
}
ResponseBytesInner::Bytes(..) => {
drop(inner);
let ResponseBytesInner::Bytes(data) = record.take_response_body()
else {
unreachable!();
};
return Poll::Ready(Some(Ok(Frame::data(data))));
}
ResponseBytesInner::UncompressedStream(stm) => {
ready!(Pin::new(stm).poll_frame(cx))
}
ResponseBytesInner::GZipStream(stm) => {
ready!(Pin::new(stm.as_mut()).poll_frame(cx))
}
ResponseBytesInner::BrotliStream(stm) => {
ready!(Pin::new(stm.as_mut()).poll_frame(cx))
}
};
// This is where we retry the NoData response
if matches!(res, ResponseStreamResult::NoData) {
continue;
}
break res;
};
if matches!(res, ResponseStreamResult::EndOfStream) {
if let Some(trailers) = record.self_mut().trailers.take() {
return Poll::Ready(Some(Ok(Frame::trailers(trailers))));
}
record.take_response_body();
}
Poll::Ready(res.into())
}
fn is_end_stream(&self) -> bool {
let inner = self.0.self_ref();
matches!(
inner.response_body,
ResponseBytesInner::Done | ResponseBytesInner::Empty
) && inner.trailers.is_none()
}
fn size_hint(&self) -> SizeHint {
// The size hint currently only used in the case where it is exact bounds in hyper, but we'll pass it through
// anyways just in case hyper needs it.
self.0.self_ref().response_body.size_hint()
}
}
impl Drop for HttpRecordResponse {
fn drop(&mut self) {
// SAFETY: this ManuallyDrop is not used again.
let record = unsafe { ManuallyDrop::take(&mut self.0) };
http_trace!(record, "HttpRecordResponse::drop");
record.finish();
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::response_body::Compression;
use crate::response_body::ResponseBytesInner;
use bytes::Buf;
use deno_net::raw::NetworkStreamType;
use hyper::body::Body;
use hyper::service::service_fn;
use hyper::service::HttpService;
use hyper_util::rt::TokioIo;
use std::error::Error as StdError;
/// Execute client request on service and concurrently map the response.
async fn serve_request<B, S, T, F>(
req: http::Request<B>,
service: S,
map_response: impl FnOnce(hyper::Response<Incoming>) -> F,
) -> hyper::Result<T>
where
B: Body + Send + 'static, // Send bound due to DuplexStream
B::Data: Send,
B::Error: Into<Box<dyn StdError + Send + Sync>>,
S: HttpService<Incoming>,
S::Error: Into<Box<dyn StdError + Send + Sync>>,
S::ResBody: 'static,
<S::ResBody as Body>::Error: Into<Box<dyn StdError + Send + Sync>>,
F: std::future::Future<Output = hyper::Result<T>>,
{
use hyper::client::conn::http1::handshake;
use hyper::server::conn::http1::Builder;
let (stream_client, stream_server) = tokio::io::duplex(16 * 1024);
let conn_server =
Builder::new().serve_connection(TokioIo::new(stream_server), service);
let (mut sender, conn_client) =
handshake(TokioIo::new(stream_client)).await?;
let (res, _, _) = tokio::try_join!(
async move {
let res = sender.send_request(req).await?;
map_response(res).await
},
conn_server,
conn_client,
)?;
Ok(res)
}
#[tokio::test]
async fn test_handle_request() -> Result<(), AnyError> {
let (tx, mut rx) = tokio::sync::mpsc::channel(10);
let server_state = HttpServerState::new();
let server_state_check = server_state.clone();
let request_info = HttpConnectionProperties {
peer_address: "".into(),
peer_port: None,
local_port: None,
stream_type: NetworkStreamType::Tcp,
};
let svc = service_fn(move |req: hyper::Request<Incoming>| {
handle_request(
req,
request_info.clone(),
server_state.clone(),
tx.clone(),
)
});
let client_req = http::Request::builder().uri("/").body("".to_string())?;
// Response produced by concurrent tasks
tokio::try_join!(
async move {
// JavaScript handler produces response
let record = rx.recv().await.unwrap();
record.set_response_body(ResponseBytesInner::from_vec(
Compression::None,
b"hello world".to_vec(),
));
record.complete();
Ok(())
},
// Server connection executes service
async move {
serve_request(client_req, svc, |res| async {
// Client reads the response
use http_body_util::BodyExt;
assert_eq!(res.status(), 200);
let body = res.collect().await?.to_bytes();
assert_eq!(body.chunk(), b"hello world");
Ok(())
})
.await
},
)?;
assert_eq!(server_state_check.strong_count(), 1);
Ok(())
}
}