Website operations are stable by checklist: patches, capacity, failure logs and recovery manual

It's not that there's no fault in stability, it's that important work doesn't depend on memory.

The most dangerous phase after the web site is often not the first deployment, but a few months of calm: certificates remind people to look, disks to grow slowly, backup scripts to fail long ago, and separation accounts to remain valid. Maintenance lists place these low-frequency but high-impact events to a fixed rhythm, and running manuals ensure that no orders are found in chat records when failure occurs. The list is short and verifiable. Write “check server security” cannot be performed; write “see if installation security updates are needed, records need to be restarted, and confirm the date of the last patch” is the result.

Daily: see only signals that affect the current service

Automatic external check every day HTTPS Core health endpoint, 5 xx An anomaly, disk criticality, repeated reboot of the service and the latest backup age. Manually only handles alarms and critical feedback. SSH Look. top It's not exactly surveillance, but it's easy to create the illusion that "I saw it yesterday, so today's okay."

Weekly: trends in updates, logs and capacity

View system security updates and application dependency bulletins, evaluate the window before arranging the window. Check login failure, abnormal 5 within a week xx, OOM, service restart and database error. Compares the growth rate of disks, databases, uploads and backups, and when thresholds are expected to be reached. Checks the integrity of a recent backup and confirms that a remote copy has arrived.

Monthly: accounts, certificates, costs and recovery preparedness

Review Manager,SSH key Check the domain name, certificate, server and storage renewal date; automated certificates are usually renewed, but the rest of the time is monitored. Reviews the issuance and malfunctions of the month: which warnings have no operational value, which problems have been identified by the user first, and whether the rollback has reached its target.

Quarterly: one real resumption exercise

Recovering the database and uploading from back-up settings, deploying applications, proxy and environment configurations on file and recording time. If a key source or software version of the manual is missing, make it up on site.RTO It is beyond the scope of operational acceptance. An assessment of the operating system, database, and the support cycle during the main operation is also made.

The malfunctioning manual only writes what is needed first.

First page: service and domain names list, architecture chart, key contact, status page or notification mode, stop condition, most recently posted location.502/504, certificate error, disk full, database not available, suspect key leaking. Each chapter uses the same structure: phenomena, confirmation commands, data protection actions, actions to restore the service, upgrade the processing conditions, authentication, evidence preservation. The command uses an example variable instead of a real password.

The fault logs are used to reset the system. They're not used to get people to take the fall.

The time line records the first impact, discovery, mitigation, recovery and confirmation; the root causes distinguish between direct triggers and systemic conditions; the line must have a responsible person and a deadline. `Be more careful after' is not the action item, as is the addition of a disk alert, correction release check or rehearsal recovery. Automatic scripts also need to be automated. dry-run Clear scope and failure to cease cannot be expanded or re-opened for reasons of economy.

What should the series end up with?

From the structure chart,systemd/Caddy/Compose, keys and backups, to issue, monitor, error and manual, each item should be wired to evidence and restore the path. Maintenance documents indicate the final authentication date and configuration changes are synchronized. The single server will still fail, but it is no longer a black box that can only be maintained by the original author ' s memory. First time you're cleaning up, you can get the last one. 502, 504 and HTTPS Queued error Step inverts: If you need an account number, log, command and contact person at each level, write what you want in the manual.