Wrangler destroys prod site when publishing, letting down paid Worker Sites customers

During publishing an old and trusted setup which worked before, worker site recently gets destroyed — instead of publishing new website, I get “cound not find index.html in your content namespace”.

The GitHub issue was raised Wrangler 1.12 publish destroys Worker site · Issue #1625 · cloudflare/wrangler · GitHub and closed but the problem seems to persist on latest versions of wrangler.

Since I posted on that closed ticket, two more Cloudflare customers reported the same problem: site wipeout bug is intermittent, after publishing many times eventually it does publish and site eventually goes up.

From a customer perspective, it’s a horrible experience since live production websites are taken down. People start to not trust Cloudflare workers:

“…it does not make me comfortable with future publishes”
—GitHub user @zwily Wrangler 1.12 publish destroys Worker site · Issue #1625 · cloudflare/wrangler · GitHub

That’s exactly what I feel currently. Fear to deploy to Cloudflare Workers.

What’s worse, even though I’m a paying Cloudflare customer, I can’t complain about workers on general Support pages! Support cases are closed instantly!

Personally, when my deployment destroyed my site the first time, on Sunday 11 pm, I spent 5 hours first trying to fix it, later giving up and in the end deploying to Vercel, with name server changes and all.

If anybody at Cloudflare is reading, here are constructive suggestions:

  1. Look at Wrangler 1.12 publish destroys Worker site · Issue #1625 · cloudflare/wrangler · GitHub — wrangler team just patched what they thought was an issue. To fix properly, they should have recreated the bug first, then written integration tests that prove the fix works. Otherwise, bug will happen again (as it is currently).
  2. Consider allowing paying Worker Sites customers to complain via general Cloudflare Support channel. This way when wrangler team bungs up something, they will have to actually fix their software. Also wrangler issues will get exposure in the general Cloudflare Support ticket pool.
  3. Consider adding an automated preventive measure — some sort of circuit breaker, which safeguards against deployments where there would be no index.html in the root of the namespace, if it would proceed further. That is, you should work from both ends: try to fix Wrangler, but also don’t let Wrangler to bung up the deployment if ever it was buggy.

Thank you.