Deploy · Infra · External APIs

30 entries

Missing `requirements.txt` entry → symptomless rollback

Symptom

A file the runtime reads isn't in the image

Symptom

On a 512MB instance, ffmpeg takes the whole worker down

Symptom (it doesn't say OOM, so you don't recognize it)

A PaaS `/tmp` is wiped on every redeploy and restart

Symptom

With auto-deploy off, pushing still serves the old version

If some services need a Manual Deploy after push, you'll assume it deployed and start debugging.

A synchronous bulk route dies at the caller's timeout

Symptom

Pinning the API spec doesn't pin the *format* of the values

Symptom

Disk full: SSH drops and JSON writes truncate

Symptom

Misdiagnosing a public API's concurrency slot limit as a server outage

Symptom

Switch away from a free geocoder with tight rate limits

Symptom

Calling a Gradio Space directly over REST

Call it with a two-step REST flow without the gradio_client package (no runtime dependency to add).

The Supabase key overhaul and an RLS misdiagnosis

service_role moved to the Legacy tab. Current path: Settings → API Keys → Secret keys.

WordPress REST: an encoded slash 404s

Symptom

Play Console: "Released to production" may not be released

Symptom

The upload key fingerprint the console shows ≠ my keystore's fingerprint

Symptom

Three OAuth traps where the error message misleads

1. youtubeSignupRequired ≠ no channel

An account locked for exhausted balance doesn't unlock on top-up

Symptom

A performance diet comes back as "it feels unstable"

Symptom

Three traps in TWA/Android builds on Windows

1. The generator's update overwrites your settings

Check the console before you throw a keystore away

What happened — this cost real time

The headroom your monitoring shows is not your headroom

Symptom

NXDOMAIN after moving nameservers to Cloudflare: records weren't imported

Symptom

A PaaS shared outbound IP in a partner allow-list stops working days later

Symptom

Gmail IMAP [AUTHENTICATIONFAILED] Invalid credentials is usually not a typo

Symptom

A large file pasted as a heredoc lands with broken newlines and an early EOF

Symptom

Three per-task limits that summed past the global one — and the global one was in the docs all along

Symptom

Three metrics blamed, three times wrong — the real rule was in the documentation

Symptom

Stopped fighting the limit and moved to a different resource pool — and three things came with it

Background

Silence monitoring must watch something written unconditionally — and it only catches silence

Background

A migration is not "add it in the new place" — it is "remove it from the old place"

Symptom