Keep the existing backups
The infrastructure rollout preserves backup files, destinations, and schedules. The portal shows whether backups are recent and whether restore verification has evidence.
Choose the right recovery
For a broken portal release, roll back the application. For a configuration problem, restore the relevant saved configuration. For a failed disk or operating system, use the machine recovery process before GitHub can reconnect.
Keep this page available
The wiki is hosted independently so you can still read recovery instructions when the local server is offline.
Read the outcome and the timestamp
The portal shows image and database backups separately. A running timer does not mean its last backup succeeded. Check the last outcome, the age of the backup, and the age of the status report.
Application restore drills
Backups and completed recovery tests have separate results. The local server restores saved data and application files in a disposable environment with no production network access. Successful evidence expires after eight days or when the deployed application changes. A failed test stays failed.
Worker recovery
Six NutsNews workers are tested one at a time from encrypted cloud backups. The test restores their saved work and broker, checks readiness, and confirms that a completed message is not processed again. Production workers are not activated or restarted. This does not measure production throughput or prove a full pipeline cutover. Failed, missing, expired or changed-revision evidence remains unverified.
ShowAlgo recovery
ShowAlgo recovery uses the actual Firebase deployment and its hosting rules. Every deployed file must match Firebase’s checksum before an encrypted backup can pass. The downloaded backup is served privately to verify its files, page routes and cache headers. The live website is not redeployed.
Raspberry backup readiness
A daily encrypted backup streams the Pi root filesystem and boot files to the local server, then to the existing cloud destination. This avoids using the nearly full SD card for temporary backup files. A verified downloaded archive proves backup readiness only; boot and full application restoration remain unverified. The disconnected receiver stays inactive.
Production worker recovery
The Cloudflare drill restores three active production shards, their saved feed state and the backend database into a disposable copy. It checks saved public news and duplicate handling without AI calls or external actions. A passing drill does not prove production is collecting fresh news. Long transfers refresh deployment observations without extending the lifetime of old test results.