Issue Review: cloudflare/workers-sdk#7570

Summary

Enterprise user reports ~0.0003% of requests experiencing high TTFB (500ms-7s) in production Workers using node:diagnostics_channel to communicate with a tail worker. Issue is a Workers Runtime concern, not a wrangler tooling bug.

Findings

• Created: 2024-12-17
• Updated: 2025-10-30
• Version: wrangler 3.91.0 (reported) → 4.60.0 (current)
• Component: Workers Runtime (production edge behavior)
• Labels: bug
• Comments: 0

Key Evidence

1. Misattributed repository: This issue pertains to the Workers Runtime behavior in production (edge network), NOT wrangler or local development tooling. The reporter explicitly states "issue observed on edge."

2. No fix exists: No PRs reference this issue, changelogs don't mention #7570, and no merged PRs address TTFB outliers specifically.

3. Related workerd issue: cloudflare/workerd#2372 documents a similar TTFB problem (larger workers = slower cold starts) that remains open. However, that issue focuses on dynamic imports affecting all requests proportionally, whereas this issue describes sporadic outliers (0.0003% of requests).

4. Different root cause indicated: The reporter's Cloudflare support contact suggested the TTFB outliers may be caused by subrequests to a third-party metrics vendor reaching IAD. The reporter disputes this, stating only the tail worker makes such requests.

5. Confidential enterprise data: The reporter indicates "details are confidential" and "this is part of enterprise support," suggesting this is actively being handled through enterprise support channels.

6. Insufficient reproduction info: The steps to reproduce state "This is a part of the enterprise support. Details are confidential." - no public reproduction is possible.

7. Stale issue: 13+ months old with no comments, no community engagement, and no updates from the reporter.

Analysis

The issue describes sporadic TTFB outliers that:

• Affect only ~0.0003% of requests (extremely rare)
• Have cache hit status despite high latency
• May involve node:diagnostics_channel and tail workers

This appears to be a production infrastructure behavior question best addressed through:

1. Enterprise support channels (already engaged per the issue)
2. The Workers Runtime team (cloudflare/workerd repository)

The workers-sdk repository contains wrangler (CLI tooling), miniflare (local simulator), and related build tools - none of which control production edge network behavior.

Recommendation

Status: CLOSE (as misattributed) or NEEDS MORE INFO

Reasoning: This issue describes Workers Runtime production behavior, not a bug in wrangler or workers-sdk tooling. The issue has been open for 13+ months with no community engagement, no reproducible steps, and the reporter indicates it's being handled through enterprise support. Additionally, the extremely rare occurrence (0.0003%) and the fact that the issue involves confidential enterprise data make this unsuitable for public debugging.

Action: Close with a comment redirecting to the appropriate venue and requesting follow-up if still relevant.

Suggested Comment

Hi @nicnocquee, thank you for this detailed report!

After review, this issue appears to describe Workers Runtime production behavior (edge network TTFB) rather than a bug in wrangler or the workers-sdk tooling. The workers-sdk repository covers the CLI tools and local development experience, while production edge behavior is handled by the Workers Runtime team.

Since you mentioned this is being handled through enterprise support, that's likely the best channel for investigating these sporadic TTFB outliers, as they can access production telemetry and infrastructure data that isn't available publicly.

If you believe there's a specific bug in wrangler or the local development tools related to this, please feel free to open a new issue with:

1. A reproducible test case
2. Steps to reproduce locally or in a non-enterprise context

For runtime-level investigations, you might also consider filing an issue at https://github.com/cloudflare/workerd if you identify a specific runtime behavior that can be reproduced.

I'm going to close this for now, but please reopen if you have additional information that points to a workers-sdk tooling bug specifically.