Operations

Parley is one Go binary with the frontend compiled into it, plus a Postgres database. There is no second service, no Node runtime in production, no cache to warm and no queue to drain.

That buys a very small operational surface. Parley runs happily on one replica or several — fanout, presence and the room-code throttle all go through Postgres — so the shape of the deployment is your call. If you read one page here before deploying, read Scaling and limits.

ArchitectureWhat the process is, how a request and a WebSocket flow through it, and what the health endpoints really check.DeploymentDocker Compose, a bare binary, and the settings that matter in both.One serverCompose, a domain and a reverse proxy — the whole team on one box, without a cluster.KubernetesThe chart and the plain manifest, the guards that fail at render time, and what a rollout does.Reverse proxyCaddy and nginx, including the WebSocket timeout that catches everyone.ObservabilityWhat Parley emits, what it does not, and where to get the rest.Scaling and limitsWhat extra replicas buy you, the connection pool, and every timeout.Backups and recoverypg_dump, restore, and what recovery actually costs you.UpgradingMigrations, why rolling back is unsafe, and upgrading Postgres itself.Air-gapped and private-CA installsMirroring the image and chart offline, and trusting an enterprise CA for the identity provider and Postgres.RunbookSymptom in, cause and fix out.