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)
- upgrade to v8 12.8
- optimizes DataView bigint methods
- fixes global interceptors
- includes CPED methods for ALS
- fix global resolution
- makes global resolution consistent using host_defined_options.
originally a separate patch but due to the global interceptor bug it
needs to be included in this pr for all tests to pass.
Fixes https://github.com/denoland/deno/issues/24756. Fixes
https://github.com/denoland/deno/issues/24796.
This also gets vitest working when using
[`--pool=forks`](https://vitest.dev/guide/improving-performance#pool)
(which is the default as of vitest 2.0). Ref
https://github.com/denoland/deno/issues/23882.
---
This PR resolves a handful of issues with child_process IPC. In
particular:
- We didn't support sending typed array views over IPC
- Opening an IPC channel resulted in the event loop never exiting
- Sending a `null` over IPC would terminate the channel
- There was some UB in the read implementation (transmuting an `&[u8]`
to `&mut [u8]`)
- The `send` method wasn't returning anything, so there was no way to
signal backpressure (this also resulted in the benchmark
`child_process_ipc.mjs` being misleading, as it tried to respect
backpressure. That gave node much worse results at larger message sizes,
and gave us much worse results at smaller message sizes).
- We weren't setting up the `channel` property on the `process` global
(or on the `ChildProcess` object), and also didn't have a way to
ref/unref the channel
- Calling `kill` multiple times (or disconnecting the channel, then
calling kill) would throw an error
- Node couldn't spawn a deno subprocess and communicate with it over IPC
This commit duplicates ops from "ext/fetch" to "ext/node" to
kick off a bigger rewrite of "node:http".
Most of duplication is temporary and will be removed as these
ops evolve.
Support `MessagePort.once` in Node mode and enable relevant
`worker_threads` test. Noticed that another Node test was passing as
well, so I enabled that too.
This change aims to replace all relative import specifiers targeted at
`tests/util/std` with mapped ones (using a `deno.json` file). Towards
updating the `std` git submodule.
Sending ALPN to a proxy, and then when tunneling, requires better
juggling of TLS configs. This improves the choice of TLS config in the
proxy connector, based on what reqwest does. It also includes some
`ext/fetch/tests.rs` that check the different combinations.
Fixes #24632
Fixes #24691
The intent is that those tests will be executed, but our check that the
files are up to date won't overwrite the contents of the tests. This is
useful when a test needs some manual edits to work.
It turns out we weren't actually running them.
---
This ended up turning into a couple of small bug fixes to get the tests
passing:
- We weren't canonicalizing the exec path properly (it sometimes still
had `..` or `.` in it)
- We weren't accepting strings in `process.exit`
There was one failure I couldn't figure out quickly, so I disabled the
test for now, and filed a follow up issue: #24694
