Uh oh!
There was an error while loading.Please reload this page.
- Notifications
You must be signed in to change notification settings - Fork32.2k
src: implement DataQueue and non-memory resident Blob#45258
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.
Already on GitHub?Sign in to your account
Uh oh!
There was an error while loading.Please reload this page.
Conversation
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
@mcollina ... this is ready for review! |
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others.Learn more.
lgtm
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
9b037f1
to18ba98e
Compare This comment was marked as outdated.
This comment was marked as outdated.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Uh oh!
There was an error while loading.Please reload this page.
Moving this back to Draft status. Trying to work through@addaleax's feedback and wanted to make it so that the |
Would you mind to backporting this to v18.x? |
See documentation in dataqueue/queue.h for detailsCo-authored-by: flakey5 <73616808+flakey5@users.noreply.github.com>PR-URL:#45258Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Co-authored-by: flakey5 <73616808+flakey5@users.noreply.github.com>PR-URL:#45258Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Co-authored-by: flakey5 <73616808+flakey5@users.noreply.github.com>PR-URL:#45258Reviewed-By: Matteo Collina <matteo.collina@gmail.com>
Notable changes:buffer: * (SEMVER-MINOR) add Buffer.copyBytesFrom(...) (James M Snell)#46500doc: * add marco-ippolito to collaborators (Marco Ippolito)#46816events: * (SEMVER-MINOR) add listener argument to listenerCount (Paolo Insogna)#46523lib: * (SEMVER-MINOR) add AsyncLocalStorage.bind() and .snapshot() (flakey5)#46387src: * (SEMVER-MINOR) add `fs.openAsBlob` to support File-backed Blobs (James M Snell)#45258tls: * (SEMVER-MINOR) support automatic DHE (Tobias Nießen)#46978url: * (SEMVER-MINOR) implement URLSearchParams size getter (James M Snell)#46308wasi: * (SEMVER-MINOR) add support for version when creating WASI (Michael Dawson)#46469worker: * (SEMVER-MINOR) add support for worker name in inspector and trace_events (Debadree Chatterjee)#46832PR-URL:#47086
Notable changes:buffer: * (SEMVER-MINOR) add Buffer.copyBytesFrom(...) (James M Snell)#46500doc: * add marco-ippolito to collaborators (Marco Ippolito)#46816events: * (SEMVER-MINOR) add listener argument to listenerCount (Paolo Insogna)#46523lib: * (SEMVER-MINOR) add AsyncLocalStorage.bind() and .snapshot() (flakey5)#46387src: * (SEMVER-MINOR) add `fs.openAsBlob` to support File-backed Blobs (James M Snell)#45258tls: * (SEMVER-MINOR) support automatic DHE (Tobias Nießen)#46978url: * (SEMVER-MINOR) implement URLSearchParams size getter (James M Snell)#46308wasi: * (SEMVER-MINOR) add support for version when creating WASI (Michael Dawson)#46469worker: * (SEMVER-MINOR) add support for worker name in inspector and trace_events (Debadree Chatterjee)#46832PR-URL:#47087
Notable changes:buffer: * (SEMVER-MINOR) add Buffer.copyBytesFrom(...) (James M Snell)#46500doc: * add marco-ippolito to collaborators (Marco Ippolito)#46816events: * (SEMVER-MINOR) add listener argument to listenerCount (Paolo Insogna)#46523lib: * (SEMVER-MINOR) add AsyncLocalStorage.bind() and .snapshot() (flakey5)#46387src: * (SEMVER-MINOR) add `fs.openAsBlob` to support File-backed Blobs (James M Snell)#45258tls: * (SEMVER-MINOR) support automatic DHE (Tobias Nießen)#46978url: * (SEMVER-MINOR) implement URLSearchParams size getter (James M Snell)#46308wasi: * (SEMVER-MINOR) add support for version when creating WASI (Michael Dawson)#46469worker: * (SEMVER-MINOR) add support for worker name in inspector and trace_events (Debadree Chatterjee)#46832PR-URL:#47087
Notable changes:buffer: * (SEMVER-MINOR) add Buffer.copyBytesFrom(...) (James M Snell)#46500doc: * add marco-ippolito to collaborators (Marco Ippolito)#46816events: * (SEMVER-MINOR) add listener argument to listenerCount (Paolo Insogna)#46523lib: * (SEMVER-MINOR) add AsyncLocalStorage.bind() and .snapshot() (flakey5)#46387src: * (SEMVER-MINOR) add `fs.openAsBlob` to support File-backed Blobs (James M Snell)#45258tls: * (SEMVER-MINOR) support automatic DHE (Tobias Nießen)#46978url: * (SEMVER-MINOR) implement URLSearchParams size getter (James M Snell)#46308wasi: * (SEMVER-MINOR) add support for version when creating WASI (Michael Dawson)#46469worker: * (SEMVER-MINOR) add support for worker name in inspector and trace_events (Debadree Chatterjee)#46832PR-URL:#47087
jimmywarting commentedMar 18, 2023
I notice that blobs transfered over workers aren't readable at all.
mini repro: import{Worker}from'worker_threads'newWorker('./worker.js').once('message',blob=>blob.text()) // worker.jsimport{parentPort}from'node:worker_threads'import{openAsBlob}from'node:fs'constblob=awaitopenAsBlob(import.meta.url.slice(7))// file:// doesn't workparentPort.postMessage(blob) I would assume that if you also closed the worker it would still work just fine to read the blob, even if it came from another thread... Error stacktrace
|
Oops, yeah just a bug. Will take a look in the next day or so. |
I also noticed that the new method is on |
* chore: upgrade to Node.js v20* src: allow embedders to override NODE_MODULE_VERSIONnodejs/node#49279* src: fix missing trailing ,nodejs/node#46909* src,tools: initialize cppgcnodejs/node#45704* tools: allow passing absolute path of config.gypi in js2cnodejs/node#49162* tools: port js2c.py to C++nodejs/node#46997* doc,lib: disambiguate the old term, NativeModulenodejs/node#45673* chore: fixup Node.js BSSL tests*nodejs/node#49492*nodejs/node#44498* deps: upgrade to libuv 1.45.0nodejs/node#48078* deps: update V8 to 10.7nodejs/node#44741* test: use gcUntil() in test-v8-serialize-leaknodejs/node#49168* module: make CJS load from ESM loadernodejs/node#47999* src: make BuiltinLoader threadsafe and non-globalnodejs/node#45942* chore: address changes to CJS/ESM loading* module: make CJS load from ESM loader (nodejs/node#47999)* lib: improve esm resolve performance (nodejs/node#46652)* bootstrap: optimize modules loaded in the built-in snapshotnodejs/node#45849* test: mark test-runner-output as flakynodejs/node#49854* lib: lazy-load deps in modules/run_main.jsnodejs/node#45849* url: use private properties for brand checknodejs/node#46904* test: refactor `test-node-output-errors`nodejs/node#48992* assert: deprecate callTrackernodejs/node#47740* src: cast v8::Object::GetInternalField() return value to v8::Valuenodejs/node#48943* test: adapt test-v8-stats for V8 updatenodejs/node#45230* tls: ensure TLS Sockets are closed if the underlying wrapclosesnodejs/node#49327* test: deflake test-tls-socket-closenodejs/node#49575* net: fix crash due to simultaneous close/shutdown on JS Stream Socketsnodejs/node#49400* net: use asserts in JS Socket Stream to catch races in futurenodejs/node#49400* lib: fix BroadcastChannel initialization locationnodejs/node#46864* src: create BaseObject with node::Realmnodejs/node#44348* src: implement DataQueue and non-memory resident Blobnodejs/node#45258* sea: add support for V8 bytecode-only cachingnodejs/node#48191* chore: fixup patch indices* gyp: put filenames in variablesnodejs/node#46965* build: modify js2c.py into GN executable* fix: (WIP) handle string replacement of fs -> original-fs* [v20.x] backport vm-related memoryfixesnodejs/node#49874* src: make BuiltinLoader threadsafe and non-globalnodejs/node#45942* src: avoid copying string in fs_permissionnodejs/node#47746* look upon my works ye mightyand dispair* chore: patch cleanup* [api] Remove AllCan Read/Writehttps://chromium-review.googlesource.com/c/v8/v8/+/5006387* fix: missing include for NODE_EXTERN* chore: fixup patch indices* fix: fail properly when js2c fails in Node.js* build: fix js2c root_gen_dir* fix: lib/fs.js -> lib/original-fs.js* build: fix original-fs file xforms* fixup! module: make CJS load from ESM loader* build: get rid of CppHeap for now* build: add patch to prevent extra fs lookup on esm load* build: greatly simplify js2c modificationsMoves our original-fs modifications back into a super simple python script action, wires up the output of that action into our call to js2c* chore: update to handle moved internal/modules/helpers file* test: update @types/node test* feat: enable preventing cppgc heap creation* feat: optionally prevent calling V8::EnableWebAssemblyTrapHandler* fix: no cppgc initialization in the renderer* gyp: put filenames in variablesnodejs/node#46965* test: disable single executable tests* fix: nan tests failing on node headers missing file* tls,http2: send fatal alert on ALPN mismatchnodejs/node#44031* test: disable snapshot tests*nodejs/node#47887*nodejs/node#49684*nodejs/node#44193* build: use deps/v8 for v8/toolsNode.js hard depends on these in their builtins* test: fix edge snapshot stack tracesnodejs/node#49659* build: remove js2c //base dep* build: use electron_js2c_toolchain to build node_js2c* fix: don't create SafeSet outside packageResolveFixes failure in parallel/test-require-delete-array-iterator:=== release test-require-delete-array-iterator ===Path: parallel/test-require-delete-array-iteratornode:internal/per_context/primordials:426 constructor(i) { super(i); } // eslint-disable-line no-useless-constructor ^TypeError: object is not iterable (cannot read property Symbol(Symbol.iterator)) at new Set (<anonymous>) at new SafeSet (node:internal/per_context/primordials:426:22)* fix: failing crashReporter tests on LinuxThese were failing because our change from node::InitializeNodeWithArgs tonode::InitializeOncePerProcess meant that we now inadvertently calledPlatformInit, which reset signal handling. This meant that our intentionalcrash function ElectronBindings::Crash no longer worked and the renderer processno longer crashed when process.crash() was called. We don't want to use Node.js'default signal handling in the renderer process, so we disable it by passingkNoDefaultSignalHandling to node::InitializeOncePerProcess.* build: only create cppgc heap on non-32 bit platforms* chore: clean up util:CompileAndCall* src: fix compatility with upcoming V8 12.1 APIsnodejs/node#50709* fix: use thread_local BuiltinLoader* chore: fixup v8 patch indices---------Co-authored-by: Keeley Hammond <vertedinde@electronjs.org>Co-authored-by: Samuel Attard <marshallofsound@electronjs.org>
* chore: upgrade to Node.js v20* src: allow embedders to override NODE_MODULE_VERSIONnodejs/node#49279* src: fix missing trailing ,nodejs/node#46909* src,tools: initialize cppgcnodejs/node#45704* tools: allow passing absolute path of config.gypi in js2cnodejs/node#49162* tools: port js2c.py to C++nodejs/node#46997* doc,lib: disambiguate the old term, NativeModulenodejs/node#45673* chore: fixup Node.js BSSL tests*nodejs/node#49492*nodejs/node#44498* deps: upgrade to libuv 1.45.0nodejs/node#48078* deps: update V8 to 10.7nodejs/node#44741* test: use gcUntil() in test-v8-serialize-leaknodejs/node#49168* module: make CJS load from ESM loadernodejs/node#47999* src: make BuiltinLoader threadsafe and non-globalnodejs/node#45942* chore: address changes to CJS/ESM loading* module: make CJS load from ESM loader (nodejs/node#47999)* lib: improve esm resolve performance (nodejs/node#46652)* bootstrap: optimize modules loaded in the built-in snapshotnodejs/node#45849* test: mark test-runner-output as flakynodejs/node#49854* lib: lazy-load deps in modules/run_main.jsnodejs/node#45849* url: use private properties for brand checknodejs/node#46904* test: refactor `test-node-output-errors`nodejs/node#48992* assert: deprecate callTrackernodejs/node#47740* src: cast v8::Object::GetInternalField() return value to v8::Valuenodejs/node#48943* test: adapt test-v8-stats for V8 updatenodejs/node#45230* tls: ensure TLS Sockets are closed if the underlying wrapclosesnodejs/node#49327* test: deflake test-tls-socket-closenodejs/node#49575* net: fix crash due to simultaneous close/shutdown on JS Stream Socketsnodejs/node#49400* net: use asserts in JS Socket Stream to catch races in futurenodejs/node#49400* lib: fix BroadcastChannel initialization locationnodejs/node#46864* src: create BaseObject with node::Realmnodejs/node#44348* src: implement DataQueue and non-memory resident Blobnodejs/node#45258* sea: add support for V8 bytecode-only cachingnodejs/node#48191* chore: fixup patch indices* gyp: put filenames in variablesnodejs/node#46965* build: modify js2c.py into GN executable* fix: (WIP) handle string replacement of fs -> original-fs* [v20.x] backport vm-related memoryfixesnodejs/node#49874* src: make BuiltinLoader threadsafe and non-globalnodejs/node#45942* src: avoid copying string in fs_permissionnodejs/node#47746* look upon my works ye mightyand dispair* chore: patch cleanup* [api] Remove AllCan Read/Writehttps://chromium-review.googlesource.com/c/v8/v8/+/5006387* fix: missing include for NODE_EXTERN* chore: fixup patch indices* fix: fail properly when js2c fails in Node.js* build: fix js2c root_gen_dir* fix: lib/fs.js -> lib/original-fs.js* build: fix original-fs file xforms* fixup! module: make CJS load from ESM loader* build: get rid of CppHeap for now* build: add patch to prevent extra fs lookup on esm load* build: greatly simplify js2c modificationsMoves our original-fs modifications back into a super simple python script action, wires up the output of that action into our call to js2c* chore: update to handle moved internal/modules/helpers file* test: update @types/node test* feat: enable preventing cppgc heap creation* feat: optionally prevent calling V8::EnableWebAssemblyTrapHandler* fix: no cppgc initialization in the renderer* gyp: put filenames in variablesnodejs/node#46965* test: disable single executable tests* fix: nan tests failing on node headers missing file* tls,http2: send fatal alert on ALPN mismatchnodejs/node#44031* test: disable snapshot tests*nodejs/node#47887*nodejs/node#49684*nodejs/node#44193* build: use deps/v8 for v8/toolsNode.js hard depends on these in their builtins* test: fix edge snapshot stack tracesnodejs/node#49659* build: remove js2c //base dep* build: use electron_js2c_toolchain to build node_js2c* fix: don't create SafeSet outside packageResolveFixes failure in parallel/test-require-delete-array-iterator:=== release test-require-delete-array-iterator ===Path: parallel/test-require-delete-array-iteratornode:internal/per_context/primordials:426 constructor(i) { super(i); } // eslint-disable-line no-useless-constructor ^TypeError: object is not iterable (cannot read property Symbol(Symbol.iterator)) at new Set (<anonymous>) at new SafeSet (node:internal/per_context/primordials:426:22)* fix: failing crashReporter tests on LinuxThese were failing because our change from node::InitializeNodeWithArgs tonode::InitializeOncePerProcess meant that we now inadvertently calledPlatformInit, which reset signal handling. This meant that our intentionalcrash function ElectronBindings::Crash no longer worked and the renderer processno longer crashed when process.crash() was called. We don't want to use Node.js'default signal handling in the renderer process, so we disable it by passingkNoDefaultSignalHandling to node::InitializeOncePerProcess.* build: only create cppgc heap on non-32 bit platforms* chore: clean up util:CompileAndCall* src: fix compatility with upcoming V8 12.1 APIsnodejs/node#50709* fix: use thread_local BuiltinLoader* chore: fixup v8 patch indices---------Co-authored-by: Keeley Hammond <vertedinde@electronjs.org>Co-authored-by: Samuel Attard <marshallofsound@electronjs.org>
Uh oh!
There was an error while loading.Please reload this page.
This is
the start ofworkbeingcollaboratively done by myself and@flakey5 to update the implementation ofBlob
to support non-memory resident data sources such as files as well as enable proper streaming support.It is a work in progress and remain draft status until it is ready to go. Opening it for transparency.Some background:
Currently,
Blob
objects consist solely of memory-resident data, backed by a collection ofv8::BackingStore
instances. While this behavior is correct standard/spec defined behavior, it is extremely inefficient. To support other use cases ofBlob
, we want to be able to efficiently and effectively support non-memory resident data sources such as files. To do so will require a number of key changes on the internals of theBlob
class.Further, the current implementation of
Blob
suffers from an overly simplistic streaming model where the entireBlob
is first read into a singleArrayBuffer
before passing the data on to the stream API. Effectively meaningBlobs
are never actually streamed.While going through this evaluation, we came to realization that many of the underlying requirements for
Blob
are (surprisingly) shared by the internal data management requirements for the QUIC implementation that is in progress, and can enable a range of other use cases also, so we have decided to take an approach that addresses those multiple use cases.When complete, this PR is going to do two things:This PR does three things:
DataQueue
which acts as a sequence of memory-resident and non-memory-resident data sources that can be consumed as a single logical stream of data. This is designed to meet the needs of bothBlob
and QUIC initially (the bits relevant to QUIC will come later)Blob
implementation to useDataQueue
internally in support of the existing all-memory-resident model.Blob
that will support efficient, non-memory resident use cases.This PR is far from complete. I would strongly recommend that you hold off performing any review on it at all until after we flip the bit to mark it ready for review. We are opening this draft PR now purely for the sake of transparency so that folks can be aware of what we are working on.I also want to avoid any and all bikeshedding on the design until we at least have the updatedBlob
implementation ready to go, and many of the design decisions made will likely not be clear until that concrete case is implemented.Specifically for Blob, this PR
will enableenables:Blob
data be collected into a singleArrayBuffer
and then passed on to the stream, but data will flow into the stream as one would actually expect.Blob
. The ability to acquire aBlob
that is backed by an on-disc file. As well as the ability to use suchBlob
instances efficiently within otherBlob
instances, e.g.new Blob(['string', fileBackedBlob])
For QUIC, eventually the
DataQueue
implemented here will replace theStream::Queue
mechanism implemented in that PR.DataQueue
will not completely replace that mechanism but covers a sizable chunk of it, which should further simplify and reduce the size of that rather complex bit of work.As for other use cases of
DataQueue
... we'll address those later, but we do have a range of cases in mind (including the recent discussions around Node.js providing an efficient http static file server)./cc@mcollina (who prompted this work due to Undici and Fetch use case requirements)