This commit migrates "deno_core" from using "FuturesUnordered" to
"tokio::task::JoinSet". This makes every op to be a separate Tokio task
and should unlock better utilization of kqueue/epoll.
There were two quirks added to this PR:
- because of the fact that "JoinSet" immediately polls spawn tasks,
op sanitizers can give false positives in some cases, this was
alleviated by polling event loop once before running a test with
"deno test", which gives canceled ops an opportunity to settle
- "JsRuntimeState::waker" was moved to "OpState::waker" so that FFI
API can still use threadsafe functions - without this change the
registered wakers were wrong as they would not wake up the
whole "JsRuntime" but the task associated with an op
---------
Co-authored-by: Matt Mastracci <matthew@mastracci.com>
This cleans up `JsRuntime` a bit more:
* We no longer print cargo's rerun-if-changed messages in `JsRuntime` --
those are printed elsewhere
* We no longer special case the OwnedIsolate for snapshots. Instead we
make use of an inner object that has the `Drop` impl and allows us to
`std::mem::forget` it if we need to extract the isolate for a snapshot
* The `snapshot` method is only available on `JsRuntimeForSnapshot`, not
`JsRuntime`.
* `OpState` construction is slightly cleaner, though I'd still like to
extract more
---------
Co-authored-by: Bartek Iwańczuk <biwanczuk@gmail.com>
Dispatches op per-realm, and allows JsRealm to be garbage collected.
Slight improvement to benchmarks, but opens opportunity to clean up event loop.
---------
Co-authored-by: Matt Mastracci <matthew@mastracci.com>
About 2% improvement on WS/HTTP benchmarks, possibly unlocking more
optimizations in the future.
---------
Co-authored-by: Matt Mastracci <matthew@mastracci.com>
This commit changes the build process in a way that preserves already
registered ops in the snapshot. This allows us to skip creating hundreds of
"v8::String" on each startup, but sadly there is still some op registration
going on startup (however we're registering 49 ops instead of >200 ops).
This situation could be further improved, by moving some of the ops
from "runtime/" to a separate extension crates.
---------
Co-authored-by: Divy Srivastava <dj.srivastava23@gmail.com>
```
Benchmark 1: deno run -A ../empty.js
Time (mean ± σ): 20.5 ms ± 0.5 ms [User: 13.4 ms, System: 5.1 ms]
Range (min … max): 19.8 ms … 24.0 ms 119 runs
Benchmark 2: target/release/deno run -A ../empty.js
Time (mean ± σ): 18.8 ms ± 0.3 ms [User: 13.0 ms, System: 4.9 ms]
Range (min … max): 18.3 ms … 19.9 ms 129 runs
```
---------
Co-authored-by: Bartek Iwańczuk <biwanczuk@gmail.com>
Currently realms are supported on `deno_core`, but there was no support
for async ops anywhere other than the main realm. The main issue is that
the `js_recv_cb` callback, which resolves promises corresponding to
async ops, was only set for the main realm, so async ops in other realms
would never resolve. Furthermore, promise ID's are specific to each
realm, which meant that async ops from other realms would result in a
wrong promise from the main realm being resolved.
This change takes the `ContextState` struct added in #17050, and adds to
it a `js_recv_cb` callback for each realm. Combined with the fact that
that same PR also added a list of known realms to `JsRuntimeState`, and
that #17174 made `OpCtx` instances realm-specific and had them include
an index into that list of known realms, this makes it possible to know
the current realm in the `queue_async_op` and `queue_fast_async_op`
methods, and therefore to send the results of promises for each realm to
that realm, and prevent the ID's from getting mixed up.
Additionally, since promise ID's are no longer unique to the isolate,
having a single set of unrefed ops doesn't work. This change therefore
also moves `unrefed_ops` from `JsRuntimeState` to `ContextState`, and
adds the lengths of the unrefed op sets for all known realms to get the
total number of unrefed ops to compare in the event loop.
This PR is a reland of #14734 after it was reverted in #16366, except
that `ContextState` and `JsRuntimeState::known_realms` were previously
relanded in #17050. Another significant difference with the original PR
is passing around an index into `JsRuntimeState::known_realms` instead
of a `v8::Global<v8::Context>` to identify the realm, because async op
queuing in fast calls cannot call into V8, and therefore cannot have
access to V8 globals. This also simplified the implementation of
`resolve_async_ops`.
Co-authored-by: Luis Malheiro <luismalheiro@gmail.com>
When an op returns an `anyhow` error with a cause (usually added using
the `.context()` method), the `Error` thrown into JavaScript contains
only the message of the outernmost error in the chain.
This PR simply changes the formatting of `anyhow::Error` from `"{}"` to
`"{:#}"`:
This significantly improves errors for code that embeds Deno and defines
custom ops. For example, in
[chiselstrike/chiselstrike](https://github.com/chiselstrike/chiselstrike),
this PR improves an error message like
```
Error: could not plan migration
```
to
```
Error: could not plan migration: could not migrate table for entity "E": could not add column for field "title": the field does not have a default value
```
example writeFile benchmark:
```
# before
time 188 ms rate 53191
time 168 ms rate 59523
time 167 ms rate 59880
time 166 ms rate 60240
time 168 ms rate 59523
time 173 ms rate 57803
time 183 ms rate 54644
# after
time 157 ms rate 63694
time 152 ms rate 65789
time 151 ms rate 66225
time 151 ms rate 66225
time 152 ms rate 65789
```
This commit adds an ability to "ref" or "unref" pending ops.
Up to this point Deno had a notion of "async ops" and "unref async ops";
the former keep event loop alive, while the latter do not block event loop
from finishing. It was not possible to change between op types after
dispatching, one had to decide which type to use before dispatch.
Instead of storing ops in two separate "FuturesUnordered" collections,
now ops are stored in a single collection, with supplemental "HashSet"
storing ids of promises that were "unrefed".
Two APIs were added to "Deno.core":
"Deno.core.refOp(promiseId)" which allows to mark promise id
to be "refed" and keep event loop alive (the default behavior)
"Deno.core.unrefOp(promiseId)" which allows to mark promise
id as "unrefed" which won't block event loop from exiting
This adds `.code` attributes to errors returned by the op-layer, facilitating classifying OS errors and helping node-compat.
Similar to Node, these `.code` attributes are stringified names of unix ERRNOs, the mapping tables are generated by [tools/codegen_error_codes.js](https://gist.github.com/AaronO/dfa1106cc6c7e2a6ebe4dba9d5248858) and derived from libuv and rust's std internals
Currently all async ops are polled lazily, which means that op
initialization code is postponed until control is yielded to the event
loop. This has some weird consequences, e.g.
```js
let listener = Deno.listen(...);
let conn_promise = listener.accept();
listener.close();
// `BadResource` is thrown. A reasonable error would be `Interrupted`.
let conn = await conn_promise;
```
JavaScript promises are expected to be eagerly evaluated. This patch
makes ops actually do that.
- Improves op performance.
- Handle op-metadata (errors, promise IDs) explicitly in the op-layer vs
per op-encoding (aka: out-of-payload).
- Remove shared queue & custom "asyncHandlers", all async values are
returned in batches via js_recv_cb.
- The op-layer should be thought of as simple function calls with little
indirection or translation besides the conceptually straightforward
serde_v8 bijections.
- Preserve concepts of json/bin/min as semantic groups of their
inputs/outputs instead of their op-encoding strategy, preserving these
groups will also facilitate partial transitions over to v8 Fast API for the
"min" and "bin" groups
This commit moves implementation of bin ops to "deno_core" crates
as well as unifying logic between bin ops and json ops to reuse
as much code as possible (both in Rust and JavaScript).
This PR makes json_op_sync/async generic to all Deserialize/Serialize types
instead of the loosely-typed serde_json::Value. Since serde_json::Value
implements Deserialize/Serialize, very little existing code needs to be updated,
however as json_op_sync/async are now generic, type inference is broken in some
cases (see cli/build.rs:146). I've found this reduces a good bit of boilerplate,
as seen in the updated deno_core examples.
This change may also reduce serialization and deserialization overhead as serde
has a better idea of what types it is working with. I am currently working on
benchmarks to confirm this and I will update this PR with my findings.
This commit rewrites initialisation of the "shared queue" and
in effect prevents from double execution of "core/core.js" and
"core/error.js".
Previously both of these files were executed every time a "JsRuntime"
was created. That lead to a situation where one copy of each script
was included in the snapshot and then another copy would be
executed after loading the snapshot.
Effectively "JsRuntime::shared_init" was removed; instead execution
of those scripts and actual initialisation of shared queue
was split into two helper functions: "JsRuntime::js_init" and
"JsRuntime::share_queue_init".
Additionally stale TODO comments were removed.
