feat: add gblob scheme support for extension (#1358)

This commit is contained in:
Levi authored and GitHub committed 2026-04-19 19:15:48 +08:00
1 parent ae70dc240c
commit a205c0b8d7
18 files changed
+2664 -147

No files matched your search

+233
View File
@@ -0,0 +1,233 @@
# gblob Protocol
## Background
Gopeed extensions can resolve a download target into something that is not a normal remote URL.
Typical examples are:
- text generated by an extension
- binary data assembled in JavaScript
- a stream produced incrementally inside the extension runtime
Returning a plain `http` or `https` URL is not enough for these cases.
The `gblob` protocol provides an internal object-URL style transport so extensions can hand generated content to the downloader in a browser-like way.
The extension-facing model is intentionally close to the browser:
- `URL.createObjectURL(new Blob(...))`
- `URL.createObjectURL(writableStream)`
- `URL.revokeObjectURL(url)`
The returned URL looks like:
```text
gblob:<id>
```
`gblob` is a process-local protocol. The source only exists inside the current Gopeed process.
## Design Goals
- make extension-generated data download as a normal task
- support both eager `Blob` data and incremental `WritableStream` data
- avoid large in-memory buffering by spooling to disk
- keep pause/continue behavior intuitive
- allow an extension to recreate a new `gblob` URL after source expiration
## High-Level Architecture
The implementation has two layers in this directory:
- [`registry.go`](./registry.go): process-local source registry and spool-file manager
- [`fetcher.go`](./fetcher.go): downloader fetcher implementation for the `gblob` scheme
The end-to-end flow is:
1. JavaScript creates a `Blob` or `WritableStream` object URL.
2. Go registers a `gblob` source and assigns it a unique `gblob:<id>` URL.
3. Source bytes are written into a local spool file.
4. The downloader resolves the `gblob` URL through the `gblob` fetcher.
5. The fetcher copies bytes from the spool file into the final download target.
## Registry
The registry owns all active `gblob` sources for the current process.
Each source tracks:
- source id and `gblob` URL
- local spool-file path
- source type: `blob` or `writable_stream`
- written byte count
- source state: `open`, `closed`, `aborted`, `failed`
- revoke state
- pin count
- optional extension session reference
Key behaviors:
- `Blob` sources are written once and immediately closed.
- `WritableStream` sources stay open until the writer closes or aborts.
- readers wait on a per-source notification channel when no new bytes are available yet.
- revocation only prevents future use; cleanup happens when the source is no longer pinned.
## Fetcher
The fetcher treats `gblob` as a local file-backed download source.
### Resolve
`Resolve()` looks up the source in the registry and builds a single-file `Resource`.
Behavior differences by source type:
- `Blob`: resolved size is the current written size
- `WritableStream`: resolved size is reported as `0` until the writer closes
`Range` is always `false`.
### Start
`Start()` opens the final target file and reads from the source spool file.
There are two resume modes:
- same `gblob` id: continue from the current target-file size
- different `gblob` id: truncate the target file and restart from byte `0`
The fetcher persists the last started source id so it can distinguish these two cases across pause/continue and task restore.
### Pause / Continue
`Pause()` stops the current copy loop by canceling the reader side.
It does not abort the `gblob` source.
If the source id stays the same, `Continue()` resumes from the existing target-file offset.
This is how normal in-process `Pause -> Continue` works.
### Source Replacement
If an extension replaces `req.url` with a new `gblob:<new-id>` before the next start, the fetcher detects the id change and restarts from scratch.
This behavior is important for recovery flows such as:
- original `gblob` source expired
- extension recreates a new `gblob` URL in `onError`
- task continues with the new source
## Lifecycle Notes
### In-process pause and continue
For the same `gblob` id:
- the partially downloaded output file is kept
- the fetcher resumes from the current file size
- `WritableStream` producers may continue writing while the task is paused
### Process restart
`gblob` sources are not persisted across process restarts.
An unfinished `gblob` task is restored without a live source behind its old URL.
The intended recovery pattern is:
1. user clicks continue
2. the old `gblob` source fails
3. extension `onError` runs
4. extension recreates a new `gblob` URL from `req.rawUrl`
5. extension updates `ctx.task.meta.req.url`
6. extension calls `ctx.task.continue()`
Because the new URL uses a different `gblob` id, the fetcher restarts from byte `0`.
## Why `rawUrl` Exists
When an extension resolves a request into `gblob:<id>`, the original site URL would otherwise be lost.
`Request.RawURL` keeps the original creation URL so lifecycle events can still reason about the original target.
Extension event URL matching prefers `req.rawUrl` when present.
This is especially useful for `onStart`, `onError`, and `onDone` on `gblob` tasks.
## Extension Usage
### 1. Create a `Blob` object URL
```js
gopeed.events.onResolve(async function (ctx) {
const blob = new Blob(["hello world"], { type: "text/plain" });
const url = URL.createObjectURL(blob);
ctx.res = {
name: "example",
files: [
{
name: "hello.txt",
size: 11,
req: { url }
}
]
};
});
```
### 2. Create a streaming object URL
```js
gopeed.events.onResolve(async function (ctx) {
const stream = new TransformStream();
const writer = stream.writable.getWriter();
const url = URL.createObjectURL(stream.writable);
(async () => {
await writer.write(new TextEncoder().encode("line 1\n"));
await writer.write(new TextEncoder().encode("line 2\n"));
await writer.close();
})();
ctx.res = {
name: "stream-example",
files: [
{
name: "stream.txt",
req: { url }
}
]
};
});
```
### 3. Recover in `onError`
```js
gopeed.events.onError(async function (ctx) {
const req = ctx.task.meta.req;
if (!req.rawUrl) return;
if (!req.url.startsWith("gblob:")) return;
req.url = URL.createObjectURL(new Blob(["recovered\n"]));
ctx.task.continue();
});
```
This creates a new `gblob` id, so the downloader restarts from byte `0`.
## Limitations
- `gblob` is not a network protocol and does not support range requests.
- unfinished `gblob` sources do not survive process restarts.
- recovery after restart depends on extension logic recreating a new source.
- the current restart behavior is full restart on source-id change, not byte-range recovery across recreated sources.
- `WritableStream` tasks with unknown total size report size `0` until the writer closes.
## Summary
`gblob` lets Gopeed extensions treat generated content as a first-class download source.
The registry handles source lifetime and disk spooling.
The fetcher turns those sources into normal download tasks with predictable pause/continue behavior.
The most important rule is simple:
- same `gblob` id means resume
- different `gblob` id means restart
+377
View File
@@ -0,0 +1,377 @@
package gblob
import (
"context"
"errors"
"io"
"os"
"path/filepath"
"strings"
"sync"
"sync/atomic"
"github.com/GopeedLab/gopeed/internal/controller"
"github.com/GopeedLab/gopeed/internal/fetcher"
"github.com/GopeedLab/gopeed/pkg/base"
)
type config struct{}
type Fetcher struct {
ctl *controller.Controller
registry *Registry
meta *fetcher.FetcherMeta
doneCh chan error
mu sync.Mutex
cancel context.CancelFunc
runDone chan struct{}
paused bool
closed bool
downloaded atomic.Int64
lastSourceID string
}
type fetcherData struct {
LastSourceID string `json:"lastSourceId"`
}
func (f *Fetcher) Setup(ctl *controller.Controller) {
f.ctl = ctl
if f.meta == nil {
f.meta = &fetcher.FetcherMeta{}
}
f.doneCh = make(chan error, 1)
}
func (f *Fetcher) Resolve(req *base.Request, opts *base.Options) error {
if f.registry == nil {
return ErrSourceNotFound
}
src, err := f.registry.Get(req.URL)
if err != nil {
return err
}
f.meta.Req = req
f.meta.Opts = opts
if f.meta.Opts == nil {
f.meta.Opts = &base.Options{}
}
name := f.meta.Opts.Name
if name == "" {
name = src.ID
}
size := src.Snapshot().Written
if src.Type == SourceTypeWritableStream {
// Writable streams may continue producing bytes after Resolve,
// so treat the total size as unknown until the writer closes.
size = 0
}
f.meta.Res = &base.Resource{
Size: size,
Range: false,
Files: []*base.FileInfo{
{
Name: name,
Path: "",
Size: size,
Req: req,
},
},
}
return nil
}
func (f *Fetcher) Start() error {
if f.registry == nil {
return ErrSourceNotFound
}
currentSourceID, err := ParseURL(f.meta.Req.URL)
if err != nil {
return err
}
f.mu.Lock()
if f.runDone != nil {
f.mu.Unlock()
return nil
}
f.paused = false
runDone := make(chan struct{})
ctx, cancel := context.WithCancel(context.Background())
f.cancel = cancel
f.runDone = runDone
f.mu.Unlock()
path := f.meta.SingleFilepath()
if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
return err
}
target, err := os.OpenFile(path, os.O_CREATE|os.O_WRONLY, 0644)
if err != nil {
return err
}
offset := int64(0)
if f.lastSourceID != "" && f.lastSourceID != currentSourceID {
if err := target.Truncate(0); err != nil {
target.Close()
return err
}
if _, err := target.Seek(0, io.SeekStart); err != nil {
target.Close()
return err
}
} else {
stat, err := target.Stat()
if err != nil {
target.Close()
return err
}
offset = stat.Size()
if _, err := target.Seek(offset, io.SeekStart); err != nil {
target.Close()
return err
}
}
f.lastSourceID = currentSourceID
f.downloaded.Store(offset)
go f.copyLoop(ctx, target, offset)
return nil
}
func (f *Fetcher) copyLoop(ctx context.Context, target *os.File, offset int64) {
defer target.Close()
defer func() {
f.mu.Lock()
runDone := f.runDone
f.runDone = nil
f.cancel = nil
paused := f.paused
closed := f.closed
f.mu.Unlock()
close(runDone)
if closed {
select {
case f.doneCh <- nil:
default:
}
return
}
if paused {
return
}
}()
src, err := f.registry.Get(f.meta.Req.URL)
if err != nil {
select {
case f.doneCh <- err:
default:
}
return
}
sourceFile, err := os.Open(src.Path)
if err != nil {
select {
case f.doneCh <- err:
default:
}
return
}
defer sourceFile.Close()
buf := make([]byte, 32*1024)
for {
if err := f.registry.WaitForReadable(ctx, f.meta.Req.URL, offset); err != nil {
if errors.Is(err, context.Canceled) {
return
}
select {
case f.doneCh <- err:
default:
}
return
}
snapshot, serr := f.registry.Get(f.meta.Req.URL)
if serr != nil {
select {
case f.doneCh <- serr:
default:
}
return
}
current := snapshot.Snapshot()
for offset < current.Written {
remain := current.Written - offset
readSize := int64(len(buf))
if remain < readSize {
readSize = remain
}
n, err := sourceFile.ReadAt(buf[:readSize], offset)
if err != nil && !errors.Is(err, io.EOF) {
select {
case f.doneCh <- err:
default:
}
return
}
if n == 0 {
break
}
if _, err := target.Write(buf[:n]); err != nil {
select {
case f.doneCh <- err:
default:
}
return
}
offset += int64(n)
f.downloaded.Store(offset)
}
if current.State == SourceStateClosed && offset >= current.Written {
if f.meta.Res != nil {
f.meta.Res.Size = current.Written
f.meta.Res.Files[0].Size = current.Written
}
select {
case f.doneCh <- nil:
default:
}
return
}
if current.State == SourceStateAborted || current.State == SourceStateFailed {
err := current.Err
if err == nil {
err = ErrSourceAborted
}
select {
case f.doneCh <- err:
default:
}
return
}
}
}
func (f *Fetcher) Patch(req *base.Request, opts *base.Options) error {
return nil
}
func (f *Fetcher) Pause() error {
f.mu.Lock()
f.paused = true
cancel := f.cancel
runDone := f.runDone
f.mu.Unlock()
if cancel != nil {
cancel()
}
if runDone != nil {
<-runDone
}
return nil
}
func (f *Fetcher) Close() error {
f.mu.Lock()
f.closed = true
cancel := f.cancel
runDone := f.runDone
f.mu.Unlock()
if cancel != nil {
cancel()
}
if runDone != nil {
<-runDone
}
select {
case f.doneCh <- nil:
default:
}
return nil
}
func (f *Fetcher) Stats() any {
return map[string]any{}
}
func (f *Fetcher) Meta() *fetcher.FetcherMeta {
return f.meta
}
func (f *Fetcher) Progress() fetcher.Progress {
return fetcher.Progress{f.downloaded.Load()}
}
func (f *Fetcher) Wait() error {
return <-f.doneCh
}
type FetcherManager struct {
registry *Registry
}
func (fm *FetcherManager) SetRegistry(registry *Registry) {
fm.registry = registry
}
func (fm *FetcherManager) Name() string {
return Scheme
}
func (fm *FetcherManager) Filters() []*fetcher.SchemeFilter {
return []*fetcher.SchemeFilter{
{
Type: fetcher.FilterTypeUrl,
Pattern: strings.ToUpper(Scheme),
},
}
}
func (fm *FetcherManager) Build() fetcher.Fetcher {
return &Fetcher{
registry: fm.registry,
}
}
func (fm *FetcherManager) ParseName(u string) string {
id, err := ParseURL(u)
if err != nil {
return ""
}
return id
}
func (fm *FetcherManager) AutoRename() bool {
return true
}
func (fm *FetcherManager) DefaultConfig() any {
return &config{}
}
func (fm *FetcherManager) Store(f fetcher.Fetcher) (data any, err error) {
_f := f.(*Fetcher)
return &fetcherData{
LastSourceID: _f.lastSourceID,
}, nil
}
func (fm *FetcherManager) Restore() (v any, f func(meta *fetcher.FetcherMeta, v any) fetcher.Fetcher) {
return &fetcherData{}, func(meta *fetcher.FetcherMeta, v any) fetcher.Fetcher {
fd := v.(*fetcherData)
return &Fetcher{
registry: fm.registry,
meta: meta,
doneCh: make(chan error, 1),
lastSourceID: fd.LastSourceID,
}
}
}
func (fm *FetcherManager) Close() error {
return nil
}
+447
View File
@@ -0,0 +1,447 @@
package gblob
import (
"context"
"errors"
"os"
"path/filepath"
"strings"
"sync"
gonanoid "github.com/matoous/go-nanoid/v2"
)
const Scheme = "gblob"
var (
ErrInvalidURL = errors.New("invalid gblob url")
ErrSourceNotFound = errors.New("gblob source not found")
ErrSourceRevoked = errors.New("gblob source revoked")
ErrSourceClosed = errors.New("gblob source closed")
ErrSourceAborted = errors.New("gblob source aborted")
)
type SourceType string
const (
SourceTypeBlob SourceType = "blob"
SourceTypeWritableStream SourceType = "writable_stream"
)
type SourceState string
const (
SourceStateOpen SourceState = "open"
SourceStateClosed SourceState = "closed"
SourceStateAborted SourceState = "aborted"
SourceStateFailed SourceState = "failed"
)
type SessionRef interface {
Retain()
Release()
}
type Snapshot struct {
Written int64
State SourceState
Err error
WaitCh <-chan struct{}
}
type Source struct {
ID string
URL string
Path string
Type SourceType
ContentType string
mu sync.Mutex
file *os.File
written int64
state SourceState
err error
revoked bool
pins int
waitCh chan struct{}
session SessionRef
sessionReleased bool
}
func (s *Source) Snapshot() Snapshot {
s.mu.Lock()
defer s.mu.Unlock()
return Snapshot{
Written: s.written,
State: s.state,
Err: s.err,
WaitCh: s.waitCh,
}
}
func (s *Source) State() SourceState {
s.mu.Lock()
defer s.mu.Unlock()
return s.state
}
func (s *Source) IsRevoked() bool {
s.mu.Lock()
defer s.mu.Unlock()
return s.revoked
}
func (s *Source) pin() {
s.mu.Lock()
defer s.mu.Unlock()
s.pins++
}
func (s *Source) unpin() int {
s.mu.Lock()
defer s.mu.Unlock()
if s.pins > 0 {
s.pins--
}
return s.pins
}
func (s *Source) shouldCleanupLocked() bool {
return s.revoked && s.pins == 0 && s.state != SourceStateOpen
}
func (s *Source) notifyLocked() {
close(s.waitCh)
s.waitCh = make(chan struct{})
}
func (s *Source) releaseSessionLocked() {
if s.session != nil && !s.sessionReleased {
s.session.Release()
s.sessionReleased = true
}
}
func (s *Source) write(data []byte) error {
s.mu.Lock()
defer s.mu.Unlock()
if s.revoked {
return ErrSourceRevoked
}
if s.state != SourceStateOpen {
if s.state == SourceStateAborted || s.state == SourceStateFailed {
if s.err != nil {
return s.err
}
return ErrSourceAborted
}
return ErrSourceClosed
}
if len(data) == 0 {
return nil
}
n, err := s.file.Write(data)
if err != nil {
s.state = SourceStateFailed
s.err = err
s.file.Close()
s.file = nil
s.releaseSessionLocked()
s.notifyLocked()
return err
}
s.written += int64(n)
s.notifyLocked()
return nil
}
func (s *Source) close() error {
s.mu.Lock()
defer s.mu.Unlock()
if s.state != SourceStateOpen {
return nil
}
s.state = SourceStateClosed
if s.file != nil {
if err := s.file.Close(); err != nil {
s.err = err
}
s.file = nil
}
s.releaseSessionLocked()
s.notifyLocked()
return nil
}
func (s *Source) abort(err error) error {
s.mu.Lock()
defer s.mu.Unlock()
if s.state != SourceStateOpen {
return nil
}
s.state = SourceStateAborted
if err != nil {
s.err = err
}
if s.file != nil {
_ = s.file.Close()
s.file = nil
}
s.releaseSessionLocked()
s.notifyLocked()
return nil
}
func (s *Source) revoke() bool {
s.mu.Lock()
defer s.mu.Unlock()
s.revoked = true
return s.shouldCleanupLocked()
}
func (s *Source) waitForReadable(ctx context.Context, offset int64) error {
for {
snapshot := s.Snapshot()
if offset < snapshot.Written {
return nil
}
switch snapshot.State {
case SourceStateClosed:
return nil
case SourceStateAborted, SourceStateFailed:
if snapshot.Err != nil {
return snapshot.Err
}
return ErrSourceAborted
}
select {
case <-ctx.Done():
return ctx.Err()
case <-snapshot.WaitCh:
}
}
}
type Registry struct {
dir string
mu sync.RWMutex
sources map[string]*Source
}
func NewRegistry(baseDir string) *Registry {
dir := baseDir
if dir == "" {
dir = filepath.Join(os.TempDir(), "gopeed-gblob")
}
return &Registry{
dir: dir,
sources: make(map[string]*Source),
}
}
func BuildURL(id string) string {
return Scheme + ":" + id
}
func ParseURL(raw string) (string, error) {
if !strings.HasPrefix(strings.ToLower(raw), Scheme+":") {
return "", ErrInvalidURL
}
id := raw[len(Scheme)+1:]
if idx := strings.IndexByte(id, '/'); idx >= 0 {
id = id[:idx]
}
if id == "" {
return "", ErrInvalidURL
}
return id, nil
}
func (r *Registry) createSource(sourceType SourceType, contentType string, session SessionRef) (*Source, error) {
if err := os.MkdirAll(r.dir, 0755); err != nil {
return nil, err
}
id, err := gonanoid.New()
if err != nil {
return nil, err
}
filePath := filepath.Join(r.dir, id)
file, err := os.Create(filePath)
if err != nil {
return nil, err
}
if session != nil {
session.Retain()
}
src := &Source{
ID: id,
URL: BuildURL(id),
Path: filePath,
Type: sourceType,
ContentType: contentType,
file: file,
state: SourceStateOpen,
waitCh: make(chan struct{}),
session: session,
}
r.mu.Lock()
r.sources[id] = src
r.mu.Unlock()
return src, nil
}
func (r *Registry) CreateBlob(data []byte, contentType string) (string, error) {
src, err := r.createSource(SourceTypeBlob, contentType, nil)
if err != nil {
return "", err
}
if err := src.write(data); err != nil {
_ = os.Remove(src.Path)
return "", err
}
if err := src.close(); err != nil {
_ = os.Remove(src.Path)
return "", err
}
return src.URL, nil
}
func (r *Registry) CreateWritableStream(session SessionRef) (string, error) {
src, err := r.createSource(SourceTypeWritableStream, "", session)
if err != nil {
return "", err
}
return src.URL, nil
}
func (r *Registry) get(raw string) (*Source, error) {
id, err := ParseURL(raw)
if err != nil {
return nil, err
}
r.mu.RLock()
defer r.mu.RUnlock()
src := r.sources[id]
if src == nil {
return nil, ErrSourceNotFound
}
return src, nil
}
func (r *Registry) Get(raw string) (*Source, error) {
return r.get(raw)
}
func (r *Registry) Write(raw string, data []byte) error {
src, err := r.get(raw)
if err != nil {
return err
}
return src.write(data)
}
func (r *Registry) CloseSource(raw string) error {
src, err := r.get(raw)
if err != nil {
return err
}
if err := src.close(); err != nil {
return err
}
r.cleanupIfNeeded(src)
return nil
}
func (r *Registry) AbortSource(raw string, err error) error {
src, gerr := r.get(raw)
if gerr != nil {
return gerr
}
if err := src.abort(err); err != nil {
return err
}
r.cleanupIfNeeded(src)
return nil
}
func (r *Registry) Revoke(raw string) error {
src, err := r.get(raw)
if err != nil {
return err
}
if src.revoke() {
r.cleanupIfNeeded(src)
}
return nil
}
func (r *Registry) Pin(raw string) error {
src, err := r.get(raw)
if err != nil {
return err
}
if src.IsRevoked() {
return ErrSourceRevoked
}
src.pin()
return nil
}
func (r *Registry) Unpin(raw string) {
src, err := r.get(raw)
if err != nil {
return
}
src.unpin()
r.cleanupIfNeeded(src)
}
func (r *Registry) WaitForReadable(ctx context.Context, raw string, offset int64) error {
src, err := r.get(raw)
if err != nil {
return err
}
return src.waitForReadable(ctx, offset)
}
func (r *Registry) cleanupIfNeeded(src *Source) {
src.mu.Lock()
shouldCleanup := src.shouldCleanupLocked()
file := src.file
src.mu.Unlock()
if !shouldCleanup {
return
}
r.mu.Lock()
delete(r.sources, src.ID)
r.mu.Unlock()
if file != nil {
_ = file.Close()
}
_ = os.Remove(src.Path)
}
func (r *Registry) Close() error {
r.mu.Lock()
sources := r.sources
r.sources = make(map[string]*Source)
r.mu.Unlock()
var lastErr error
for _, src := range sources {
src.mu.Lock()
if src.file != nil {
if err := src.file.Close(); err != nil {
lastErr = err
}
src.file = nil
}
src.releaseSessionLocked()
src.notifyLocked()
src.mu.Unlock()
if err := os.Remove(src.Path); err != nil && !errors.Is(err, os.ErrNotExist) {
lastErr = err
}
}
return lastErr
}