Table of Contents
🌏 中文版
DigitalOcean App Platform is a managed PaaS that builds, deploys, and scales Git repositories or container images. It sits between Droplets and Kubernetes: the platform manages OS, routing, TLS, deployment, and container lifecycle; teams retain application, data, dependency, and capacity responsibilities.
Component types are process contracts
An App can contain a public HTTP Service, an ingress-free resident Worker, a scheduled or deployment Job that exits, a Static Site, and serverless Functions. Do not disguise queue consumers as Services or run migrations in every web replica.
Components may use separate source directories or images and bind variables from databases and peers. Internal routing provides reachability, not dependency readiness. Clients still need timeouts, retries, connection budgets, and graceful shutdown.
An App Spec is complete state, not a partial patch
The YAML App Spec declares components, sources, commands, environments, domains, ingress, alerts, databases, regions, and scaling. Version it and download the current state before CLI or API edits: an update spec fully defines the App, so omitted configuration can disappear.
Never commit plaintext secret values. Use separate Apps, credentials, and databases across environments. Distinguish build-time from runtime variables and prevent production secrets from entering artifacts.
Health and autoscaling require application cooperation
Health checks provide readiness and remove unhealthy instances from traffic; liveness probes can restart Services and Workers. Readiness asks whether new work is safe, while liveness detects a stuck process. Neither should scan an entire database on every probe.
Autoscaling supports fixed horizontal replicas and CPU or HTTP-request metrics; vertical instance sizing remains a fixed manual choice. Any exceeded metric can scale up, while all must fall below targets to scale down. Queue depth, database saturation, and third-party limits remain outside that controller, so workers need backpressure and external signals.
Treat runtime filesystems as ephemeral. Put durable objects in Spaces and relational state in Managed Databases. Workloads requiring writable local volumes, network appliances, or host control often fit Droplets or DOKS better.
App Platform suits teams already using DigitalOcean databases or Spaces that want web, worker, job, and static deployments with less operations. Compare Railway or Render for developer workflow, Koyeb or Fly.io for placement, and hyperscalers for IAM or compliance depth. Test bad-release rollback, readiness versus liveness, job replay, scaling connection storms, and database restoration.
References
Loading...