fix(chromium)!: isolate each conversion in its own browser context

This commit is contained in:
Julien Neuhart
2026-10-13 18:31:38 +02:00
parent 887727b40e
commit 7b604da4f1
6 changed files with 148 additions and 26 deletions

View File

@@ -12,10 +12,12 @@ import (
"time"
cdprotobrowser "github.com/chromedp/cdproto/browser"
"github.com/chromedp/cdproto/cdp"
"github.com/chromedp/cdproto/fetch"
"github.com/chromedp/cdproto/network"
"github.com/chromedp/cdproto/page"
"github.com/chromedp/cdproto/runtime"
"github.com/chromedp/cdproto/target"
"github.com/chromedp/chromedp"
"github.com/dlclark/regexp2/v2"
"github.com/shirou/gopsutil/v4/process"
@@ -62,14 +64,15 @@ type browserArguments struct {
hyphenDataDirPath string
// Tasks specific.
allowList []*regexp2.Regexp
denyList []*regexp2.Regexp
denyPrivateIPs bool
denyPublicIPs bool
clearCache bool
clearCookies bool
clearStorage bool
disableJavaScript bool
allowList []*regexp2.Regexp
denyList []*regexp2.Regexp
denyPrivateIPs bool
denyPublicIPs bool
sharedBrowserContext bool
clearCache bool
clearCookies bool
clearStorage bool
disableJavaScript bool
}
type chromiumBrowser struct {
@@ -435,6 +438,55 @@ func (b *chromiumBrowser) screenshot(ctx context.Context, logger *slog.Logger, u
})
}
// newIsolatedTarget creates a browser context and opens a blank tab in it, so
// cookies, cache, and storage never leak to another conversion, even a
// concurrent one. The caller must dispose the browser context.
//
// chromedp.WithNewBrowserContext does not work here: it opens the tab without
// a new window, which headless Chromium rejects with "no browser is open", and
// it leaks the browser context on that failure.
func (b *chromiumBrowser) newIsolatedTarget(ctx context.Context, logger *slog.Logger) (cdp.BrowserContextID, target.ID, error) {
c := chromedp.FromContext(b.ctx)
if c == nil || c.Browser == nil {
return "", "", errors.New("browser not allocated")
}
executor := cdp.WithExecutor(ctx, c.Browser)
browserContextID, err := target.CreateBrowserContext().WithDisposeOnDetach(true).Do(executor)
if err != nil {
return "", "", fmt.Errorf("create browser context: %w", err)
}
targetID, err := target.CreateTarget("about:blank").
WithBrowserContextID(browserContextID).
WithNewWindow(true).
Do(executor)
if err != nil {
b.disposeBrowserContext(logger, browserContextID)
return "", "", fmt.Errorf("create target: %w", err)
}
return browserContextID, targetID, nil
}
// disposeBrowserContext disposes a browser context created by
// [chromiumBrowser.newIsolatedTarget]. It uses a fresh context, as the
// conversion's one may already be done.
func (b *chromiumBrowser) disposeBrowserContext(logger *slog.Logger, browserContextID cdp.BrowserContextID) {
c := chromedp.FromContext(b.ctx)
if c == nil || c.Browser == nil {
return
}
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
err := target.DisposeBrowserContext(browserContextID).Do(cdp.WithExecutor(ctx, c.Browser))
if err != nil {
logger.ErrorContext(ctx, fmt.Sprintf("dispose browser context: %s", err))
}
}
func (b *chromiumBrowser) do(ctx context.Context, logger *slog.Logger, url string, options Options, aggregate *networkAggregate, tasks chromedp.Tasks) error {
if !b.isStarted.Load() {
return errors.New("browser not started, cannot handle tasks")
@@ -461,7 +513,18 @@ func (b *chromiumBrowser) do(ctx context.Context, logger *slog.Logger, url strin
timeoutCtx, timeoutCancel := context.WithTimeout(b.ctx, time.Until(deadline))
defer timeoutCancel()
taskCtx, taskCancel := chromedp.NewContext(timeoutCtx)
var taskOpts []chromedp.ContextOption
if !b.arguments.sharedBrowserContext {
browserContextID, targetID, err := b.newIsolatedTarget(timeoutCtx, logger)
if err != nil {
return fmt.Errorf("create isolated browser context: %w", err)
}
// Deferred before taskCancel, so it runs after the tab is closed.
defer b.disposeBrowserContext(logger, browserContextID)
taskOpts = append(taskOpts, chromedp.WithTargetID(targetID))
}
taskCtx, taskCancel := chromedp.NewContext(timeoutCtx, taskOpts...)
defer taskCancel()
// Accumulate per-conversion network activity for telemetry.