Utilize idle tabs in puppeteer for concurrent rendering #3860

habdelra · 2026-01-16T21:43:17Z

This PR updates how our prerenderer works so that all chrome tabs are used for rendering, such that a single server can concurrently handle multiple render requests.

Update AWS infra to scale down the number of prerender servers by 25%
Add PRERENDER_REALM_TAB_MAX env var to parameter store and infra

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 234084aa5b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-01-16T21:49:08Z

packages/realm-server/prerender/page-pool.ts

+    let entries = this.#realmPages.get(realm);
+    let entryList = entries
+      ? [...entries].filter((entry) => !entry.closing && !entry.transitioning)
+      : [];


Allow same-realm queueing during tab transitions

In #selectEntryForRealm, the entry list for a realm excludes any tab marked transitioning (filter((entry) => !entry.closing && !entry.transitioning)). When a cross‑realm request queues on a busy tab, the method sets transitioning = true for that tab later in this function, so a subsequent request for the original realm no longer sees its own tab to queue behind. If the pool is full (e.g., maxPages = 1 and no standby), the same‑realm request now throws “No standby page available” even though the tab would become free shortly. This only happens under concurrent cross‑realm load, but it turns what could be a queued render into a hard failure. Consider letting same‑realm requests queue on transitioning tabs (or scoping transitioning to cross‑realm selection) to avoid these false “no capacity” errors.

Useful? React with 👍 / 👎.

github-actions · 2026-01-16T22:05:59Z

Host Test Results

1 files ± 0 1 suites ±0 1h 35m 59s ⏱️ + 7m 2s
1 887 tests +218 1 872 ✅ +220 15 💤 ±0 0 ❌ - 1
1 902 runs +219 1 887 ✅ +222 15 💤 ±0 0 ❌ - 2

Results for commit 234084a. ± Comparison against base commit 90d8c13.

This pull request removes 1 and adds 219 tests. Note that renamed tests count towards both.

Chrome ‑ Global error: Uncaught TypeError: Failed to fetch at http://localhost:7357/assets/chunk.b9c2f4bf6c1283ea8e7a.js, line 150984  While executing test: Integration | card-copy: copy button does not appear when there is 1 stack for single card item

Chrome ‑ Integration | Store: added instance that was previously not saved will begin to auto save after being added
Chrome ‑ Integration | Store: an instance can be restored after a loader reset
Chrome ‑ Integration | Store: an instance can debounce auto saves
Chrome ‑ Integration | Store: an instance can live update thru an error state
Chrome ‑ Integration | Store: an instance live updates from indexing events for a code update
Chrome ‑ Integration | Store: an instance live updates from indexing events for an instance update
Chrome ‑ Integration | Store: an instance that started out with a local ID can be restored after a loader reset
Chrome ‑ Integration | Store: an instance will NOT auto save when its data changes, if the user does not have write permissions
Chrome ‑ Integration | Store: an instance will auto save when its data changes
Chrome ‑ Integration | Store: an unsaved instance live updates when realm event matching local ID is received
…

Copilot

Pull request overview

This PR refactors the prerenderer to enable concurrent rendering across multiple Chrome tabs per server, replacing the previous per-realm serialization model. The changes introduce tab-level queuing and a configurable per-realm tab maximum to optimize resource utilization.

Changes:

Removed global per-realm serialization, moving concurrency control to tab-level queuing within PagePool
Introduced PRERENDER_REALM_TAB_MAX environment variable to limit concurrent tabs per realm
Updated RenderRunner to use a lease/release pattern with try-finally blocks for proper resource cleanup

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
packages/realm-server/tests/prerendering-test.ts	Added TestSemaphore and comprehensive tests for concurrent tab usage, queuing behavior, and tab reassignment logic
packages/realm-server/prerender/render-runner.ts	Refactored to acquire page leases with release callbacks and ensure cleanup via try-finally blocks; simplified auth change detection
packages/realm-server/prerender/prerenderer.ts	Removed per-realm promise chaining and global semaphore acquisition; simplified retry logic by delegating concurrency control to PagePool
packages/realm-server/prerender/page-pool.ts	Implemented TabQueue for tab-level queuing, added tab selection logic favoring idle aligned tabs, and introduced transitioning/closing flags to prevent race conditions

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-16T22:27:34Z

packages/realm-server/tests/prerendering-test.ts

+          'tab realigned to realm-b when queued request starts',
+        );
+        third.release();
+        blocker.release();


The blocker lease is released twice (lines 2263 and 2276). This could cause issues if the release function is not idempotent. While the PagePool's release function appears to check for double-release, this is confusing and may indicate a logic error in the test.

Suggested change

blocker.release();

Copilot · 2026-01-16T22:27:35Z

packages/realm-server/prerender/render-runner.ts

+              shortCircuit = true;
+              break;
+            }
+            if (this.#isAuthError(error)) {
+              shortCircuit = true;


The value assigned to shortCircuit here is unused.

Suggested change

shortCircuit = true;

break;

}

if (this.#isAuthError(error)) {

shortCircuit = true;

break;

}

if (this.#isAuthError(error)) {

Copilot · 2026-01-16T22:27:35Z

packages/realm-server/prerender/render-runner.ts

+              break;
+            }
+            if (this.#isAuthError(error)) {
+              shortCircuit = true;


The value assigned to shortCircuit here is unused.

Suggested change

shortCircuit = true;

Copilot · 2026-01-16T22:27:35Z

packages/realm-server/prerender/render-runner.ts

+          isolatedHTML = capture.value;
+        } else {
+          let capErr = this.#captureToError(capture);
+          if (!error && capErr) {


This negation always evaluates to true.

Suggested change

if (!error && capErr) {

if (capErr) {

Copilot · 2026-01-16T22:27:35Z

packages/realm-server/prerender/render-runner.ts

-            shortCircuit = true;
-            break;
+        for (let step of steps) {
+          if (shortCircuit) break;


This use of variable 'shortCircuit' always evaluates to false.

Suggested change

if (shortCircuit) break;

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 234084aa5b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-01-16T22:39:26Z

packages/realm-server/prerender/page-pool.ts

+    } else {
+      for (let entry of entries) {
+        void this.#closeEntry(entry);


Prevent oversubscribing contexts during async eviction

In the awaitIdle: false branch, the realm is removed from #realmPages and each entry is closed asynchronously. Because pool sizing (#poolEntryCount/#totalContextCount) only counts entries still in #realmPages, the soon‑to‑be‑closed contexts are no longer counted and #ensureStandbyPool() can spin up new standbys while the old contexts are still running. Under timeout/unusable evictions (the new awaitIdle: false path), this can temporarily exceed the configured maxPages+1 and spike Chrome context count/memory. Consider keeping closing entries in the map (marked closing) or deferring standby creation until closes finish to preserve the capacity limit.

Useful? React with 👍 / 👎.

Utilize idle tabs in puppeteer for concurrent rendering

234084a

chatgpt-codex-connector bot reviewed Jan 16, 2026

View reviewed changes

habdelra marked this pull request as draft January 16, 2026 21:56

habdelra requested a review from Copilot January 16, 2026 22:24

habdelra marked this pull request as ready for review January 16, 2026 22:24

Copilot started reviewing on behalf of habdelra January 16, 2026 22:25 View session

Copilot AI reviewed Jan 16, 2026

View reviewed changes

chatgpt-codex-connector bot reviewed Jan 16, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Utilize idle tabs in puppeteer for concurrent rendering #3860

Utilize idle tabs in puppeteer for concurrent rendering #3860

habdelra commented Jan 16, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Jan 16, 2026

Uh oh!

github-actions bot commented Jan 16, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Jan 16, 2026

Uh oh!

Copilot AI Jan 16, 2026

Uh oh!

Copilot AI Jan 16, 2026

Uh oh!

Copilot AI Jan 16, 2026

Uh oh!

Copilot AI Jan 16, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Jan 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Utilize idle tabs in puppeteer for concurrent rendering #3860

Are you sure you want to change the base?

Utilize idle tabs in puppeteer for concurrent rendering #3860

Conversation

habdelra commented Jan 16, 2026

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Jan 16, 2026

Host Test Results

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants