Choosing a database connection pool size A connection pool should fit both the application’s concurrency and the database’s usable capacity. Each worker process may own a separate pool, so the…

Choosing a health check that reflects user-visible failure A health check should answer a defined operational question. A process can respond while login, database writes or an essential customer workflow is…

Designing a recovery test for application files and data A backup is useful only when it can restore the components needed for a working application. Test a recovery in a separate environment before an…

Documenting the production runtime and recovery procedure A production runbook records the steps needed to operate and recover the actual application. It should be specific enough that an authorised maintainer…

Keeping temporary uploads from filling application storage Upload workflows can use temporary space before a file reaches its final storage. Failed processing and abandoned requests can leave data that consumes…

Measuring application latency by dependency A slow request may spend most of its time waiting for a database, remote API, storage operation or queue. Measure the components before changing the…

Planning a backward-compatible database deployment A backward-compatible schema change lets old and new application processes coexist during a release. It also gives a code rollback a better chance of…

Preparing a hosting handover for a development team A hosting handover should give the incoming team the access and operational context needed to maintain the site without sharing one unrestricted account…

Protecting shared session and cache services Redis, Memcached or another shared store can hold more than disposable page data. Sessions, locks, queues and application state need appropriate access,…

Scheduling application jobs without overlapping runs A scheduled job may still be running when its next interval arrives. Without coordination, overlapping copies can duplicate work or compete for the same…

Serving large file downloads without tying up workers A large download can occupy an application worker for much longer than the authorization decision takes. A supported web-server or storage-service…

Setting retention limits for application logs Application logs need enough history for support and incident review without growing until they stop the website. Retention should reflect operational…

Using a queue for slow exports and notifications A queue is useful when work should continue independently of a web request and can be tracked to completion. It also introduces storage, worker and…