Adds a `parallel` flag to `deno serve`. When present, we spawn multiple
workers to parallelize serving requests.
```bash
deno serve --parallel main.ts
```
Currently on linux we use `SO_REUSEPORT` and rely on the fact that the
kernel will distribute connections in a round-robin manner.
On mac and windows, we sort of emulate this by cloning the underlying
file descriptor and passing a handle to each worker. The connections
will not be guaranteed to be fairly distributed (and in practice almost
certainly won't be), but the distribution is still spread enough to
provide a significant performance increase.
---
(Run on an Macbook Pro with an M3 Max, serving `deno.com`
baseline::
```
❯ wrk -d 30s -c 125 --latency http://127.0.0.1:8000
Running 30s test @ http://127.0.0.1:8000
2 threads and 125 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 239.78ms 13.56ms 330.54ms 79.12%
Req/Sec 258.58 35.56 360.00 70.64%
Latency Distribution
50% 236.72ms
75% 248.46ms
90% 256.84ms
99% 268.23ms
15458 requests in 30.02s, 2.47GB read
Requests/sec: 514.89
Transfer/sec: 84.33MB
```
this PR (`with --parallel` flag)
```
❯ wrk -d 30s -c 125 --latency http://127.0.0.1:8000
Running 30s test @ http://127.0.0.1:8000
2 threads and 125 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 117.40ms 142.84ms 590.45ms 79.07%
Req/Sec 1.33k 175.19 1.77k 69.00%
Latency Distribution
50% 22.34ms
75% 223.67ms
90% 357.32ms
99% 460.50ms
79636 requests in 30.07s, 12.74GB read
Requests/sec: 2647.96
Transfer/sec: 433.71MB
```
Changes:
- Implements a TCP socket listener that will allow for round-robin
load-balancing in-process.
- Cleans up the raw networking code to make it easier to work with.
<!--
Before submitting a PR, please read https://deno.com/manual/contributing
1. Give the PR a descriptive title.
Examples of good title:
- fix(std/http): Fix race condition in server
- docs(console): Update docstrings
- feat(doc): Handle nested reexports
Examples of bad title:
- fix #7123
- update docs
- fix bugs
2. Ensure there is a related issue and it is referenced in the PR text.
3. Ensure there are tests that cover the changes.
4. Ensure `cargo test` passes.
5. Ensure `./tools/format.js` passes without changing files.
6. Ensure `./tools/lint.js` passes.
7. Open as a draft PR if your work is still in progress. The CI won't
run
all steps, but you can add '[ci]' to a commit message to force it to.
8. If you would like to run the benchmarks on the CI, add the 'ci-bench'
label.
-->
---------
Signed-off-by: Matt Mastracci <matthew@mastracci.com>
Co-authored-by: Matt Mastracci <matthew@mastracci.com>
Implementation of generics for `#[op2]`, along with some refactoring to
improve the ergonomics of ops with generics parameters:
- The ops have generics on the struct rather than the associated
methods, which allows us to trait-ify ops (impossible when they are on
the methods)
- The decl() method can become a trait-associated const field which
unlocks future optimizations
Callers of ops need to switch from:
`op_net_connect_tcp::call::<TestPermission>(conn_state, ip_addr)` to
`op_net_connect_tcp::<TestPermission>::call(conn_state, ip_addr)`.
…nclusion" (#19519)"
This reverts commit 28a4f3d0f5.
This change causes failures when used outside Deno repo:
```
============================================================
Deno has panicked. This is a bug in Deno. Please report this
at https://github.com/denoland/deno/issues/new.
If you can reliably reproduce this panic, include the
reproduction steps and re-run with the RUST_BACKTRACE=1 env
var set and include the backtrace in your report.
Platform: linux x86_64
Version: 1.34.3+b37b286
Args: ["/opt/hostedtoolcache/deno/0.0.0-b37b286f7fa68d5656f7c180f6127bdc38cf2cf5/x64/deno", "test", "--doc", "--unstable", "--allow-all", "--coverage=./cov"]
thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Failed to read "/home/runner/work/deno/deno/core/00_primordials.js"
Caused by:
No such file or directory (os error 2)', core/runtime/jsruntime.rs:699:8
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
```
Relands #19463. This time the `ExtensionFileSourceCode` enum is
preserved, so this effectively just splits feature
`include_js_for_snapshotting` into `exclude_js_sources` and
`runtime_js_sources`, adds a `force_include_js_sources` option on
`extension!()`, and unifies `ext::Init_ops_and_esm()` and
`ext::init_ops()` into `ext::init()`.
`ZeroCopyBuf` was convenient to use, but sometimes it did hide details
that some copies were necessary in certain cases. Also it made it way to easy
for the caller to pass around and convert into different values. This commit
splits `ZeroCopyBuf` into `JsBuffer` (an array buffer coming from V8) and
`ToJsBuffer` (a Rust buffer that will be converted into a V8 array buffer).
As a result some magical conversions were removed (they were never used)
limiting the API surface and preparing for changes in #19534.
… (#19463)"
This reverts commit ceb03cfb03.
This is being reverted because it causes 3.5Mb increase in the binary
size,
due to runtime JS code being included in the binary, even though it's
already snapshotted.
CC @nayeemrmn
Remove `ExtensionFileSourceCode::LoadedFromFsDuringSnapshot` and feature
`include_js_for_snapshotting` since they leak paths that are only
applicable in this repo to embedders. Replace with feature
`exclude_js_sources`. Additionally the feature
`force_include_js_sources` allows negating it, if both features are set.
We need both of these because features are additive and there must be a
way of force including sources for snapshot creation while still having
the `exclude_js_sources` feature. `force_include_js_sources` is only set
for build deps, so sources are still excluded from the final binary.
You can also specify `force_include_js_sources` on any extension to
override the above features for that extension. Towards #19398.
But there was still the snapshot-from-snapshot situation where code
could be executed twice, I addressed that by making `mod_evaluate()` and
scripts like `core/01_core.js` behave idempotently. This allowed
unifying `ext::init_ops()` and `ext::init_ops_and_esm()` into
`ext::init()`.
This function was inherently racy which showed on slow machines -
the connect future started before the spawned task was first polled.
This change makes it so we're already accepting a connection when
the connect future is first polled.
This implements two macros to simplify extension registration and centralize a lot of the boilerplate as a base for future improvements:
* `deno_core::ops!` registers a block of `#[op]`s, optionally with type
parameters, useful for places where we share lists of ops
* `deno_core::extension!` is used to register an extension, and creates
two methods that can be used at runtime/snapshot generation time:
`init_ops` and `init_ops_and_esm`.
---------
Co-authored-by: Bartek Iwańczuk <biwanczuk@gmail.com>
There's no point for this API to expect result. If something fails it should
result in a panic during build time to signal to embedder that setup is
wrong.
The whole point of creating this alternative operation was to allow
usage in node, without `--unstable` flag.
Introduced and I believe missed in
https://github.com/denoland/deno/pull/16520/
This PR adds copies of several unstable APIs that are available
in "Deno[Deno.internal].nodeUnstable" namespace.
These copies do not perform unstable check (ie. don't require
"--unstable" flag to be present). Otherwise they work exactly
the same, including permission checks.
These APIs are not meant to be used by users directly and
can change at any time.
Copies of following APIs are available in that namespace:
- Deno.spawnChild
- Deno.spawn
- Deno.spawnSync
- Deno.serve
- Deno.upgradeHttpRaw
- Deno.listenDatagram
Previously `op_net_listen`, `op_net_accept`, and various other ops in
ext/net where variadic on the transport. This created a lot of code
bloat. This commit updates the code to instead have separate ops for
each transport.
This commit adds a `reuseAddress` option for UDP sockets. When this
option is enabled, one can listen on an address even though it is
already being listened on from a different process or thread. The new
socket will steal the address from the existing socket.
On Windows and Linux this uses the `SO_REUSEADDR` option, while on other
Unixes this is done with `SO_REUSEPORT`.
This behavior aligns with what libuv does.
TCP sockets still unconditionally set the `SO_REUSEADDR` flag - this
behavior matches Node.js and Go. This PR does not change this behaviour.
Co-authored-by: Luca Casonato <hello@lcas.dev>
Stop allowing clippy::derive-partial-eq-without-eq and fix warnings
about deriving PartialEq without also deriving Eq.
In one case I removed the PartialEq because it a) wasn't necessary,
and b) sketchy because it was comparing floating point numbers.
IMO, that's a good argument for enforcing the lint rule, because it
would most likely have been caught during review if it had been enabled.
