Blog / Web management /
A website recovery checklist after a failed deployment
The deployment fails. The site half-works. Three people start fixing it at once: one restores a file, another changes a setting, someone republishes the latest build. Everyone’s busy, and nobody can say what state the site is in.
I treat recovery as a sequence: establish the impact, choose an authorised mitigation, restore a known state, then verify the customer task. The steps depend on the platform, but the decision record and ownership should be clear enough that competing fixes don’t cause a second incident.
Establish the current impact
Identify what is unavailable or incorrect and which users or workflows are affected. A broken purchase route, missing content and a minor display defect need different immediate responses.
Capture a safe reproducible example with the public URL, time and observed behaviour. Keep sensitive information out of shared screenshots and logs.
Name the person coordinating recovery and the person authorised to make consequential decisions. The coordinator does not need to perform every action, but should know what is being changed and in which order.
Pause competing work
Stop unrelated deployments and relevant edits through the agreed process. Keep an urgent exception route for corrections that cannot wait.
Record actions already taken. A recovery plan based on the original failed version can become wrong if several partial changes have happened since the first report.
Silence in the chat doesn’t mean nobody is working. Ask the owners to confirm the current state and their next action before anyone applies another overlapping fix.
Identify the last known useful state
Locate the relevant version, configuration and data references. Confirm what each recovery point includes and excludes.
A previous application or theme version may not restore database changes, external settings or new customer records. Map those layers before deciding that a rollback is a single operation.
Keep access available to authorised staff and use the approved recovery procedure. I don’t run an untested command copied from another environment. It adds more doubt than it clears.
Choose rollback, repair or containment deliberately
A rollback can be appropriate when a known prior state can safely restore the task. A targeted repair may be needed where data or dependencies make a simple return unsuitable.
Containment can switch off a failing feature to keep essential service running, but think through what the customer loses. Don’t hide required information or show a false success message just to make the page load.
Record the decision and reason. If uncertainty remains about the root cause, distinguish restoring service from proving the explanation. Those can happen at different stages.
Protect data created after the release
Consider orders, submissions, comments or other writes received during the incident. A broad restore can overwrite legitimate new activity if the recovery plan ignores it.
Involve the relevant technical and business owners in deciding how those records are preserved or reconciled. Do not improvise deletions to make a backup easier to apply.
Keep private data in its approved systems. The incident record generally needs references and handling decisions, not another copy of customer details.
Verify the restored public journey
Repeat the failing task and important normal routes. Check the actual public output rather than only the deployment dashboard or administration screen.
Inspect dependent services such as forms, notifications, feeds or payment handoffs where the recovery affected them. A homepage returning successfully does not establish full operational recovery.
Record tested conditions and remaining limitations. Give unresolved issues a next action and owner instead of declaring the whole site fixed from one successful page load.
Separate recovery from the later investigation
Communicate verified status to the relevant teams and preserve the release and incident timeline. Avoid guessed restoration times or an unsupported cause announced too early.
Once the immediate service is stable, investigate why the failure occurred and which preparation or check should change. Keep the finding linked to the recovery record.
Check the difference between restored and reconciled
A public task can start working again before every consequence is sorted. A delayed notification, a partial import or an interrupted background job may still need review once the main site is back. Give those explicit owners and keep them out of the recovery announcement, so the business knows which customer actions are verified and which records still need reconciling.
Look at what actually happened around the failed deployment. A retry may have done part of a job twice, and a rollback may have returned the code without reversing an external action. Ask the system owner what a safe retry means. Don’t repeat a consequential process blindly to make a dashboard look finished. And keep the status specific enough for support to be accurate: “the website is available” can be true while one feature is paused.
Recovery is complete when the service and the necessary records are in an understood state, not when the first page loads again.
Related reading
- What belongs in a website change record
- A practical SSL renewal check for a business owner
- A calm process for triaging plugin and app conflicts
Pick the most important task on your site and decide what recovery state you’d use after a failed deployment. Which data or external setting would sit outside that rollback, and who would deal with it?