Configuration reference

Configuration is environment-variable driven. The Compose files provide runnable defaults; production values must be reviewed explicitly.

Kubernetes manifests keep the same environment-variable contract. Kustomize overlays merge the ConfigMaps and Secrets that provide those values.

All Go processes accept ENV as the runtime mode label. It defaults to development, but credentials, TLS, origins, storage, and public URLs still require their own explicit production values.

HTTP server

Host and timeouts use SERVER_HOST, SERVER_PORT, request/read/header/idle timeout values, request body limit, session and CSRF expiry, host URL, allowed origins, same-site mode, SERVER_CSRF_HMAC_SECRET, and CRYPTO_SECRET.

Startup requires persistent, distinct values for CRYPTO_SECRET and SERVER_CSRF_HMAC_SECRET in any environment. Startup rejects empty values and reuse of one value for both purposes. Both Compose profiles require both variables explicitly; the Kubernetes setup workflow validates chronoverse-server-security.

gRPC

Servers use GRPC_HOST, GRPC_PORT, request timeout, and GRPC_TLS_*. Clients use per-service host, port, TLS enablement, CA path, plus shared client certificate and key paths.

Infrastructure

• PostgreSQL: PgBouncer endpoint, connection pool, dial timeout, dedicated application credentials, database, and TLS paths. Kubernetes defines explicit per-workload POSTGRES_MIN_CONNS/POSTGRES_MAX_CONNS budgets; zero is a valid idle minimum.
• ClickHouse: hosts, credentials, pool limits, dial timeout, and TLS paths.
• Redis: pool sizes, timeouts, memory policy, password, database, and TLS paths. Production compose sets REDIS_MAX_MEMORY=${REDIS_MAX_MEMORY:-768mb} for every Redis client process so startup order cannot reset the server to the lower code default.
• Meilisearch: URI, master key, and TLS client settings.
• Kafka: brokers, consumer group, topic partitions, replication factor, and TLS paths.

Workers

Scheduler settings control poll interval, context timeout, and batch size. Jobs service settings include runtime heartbeat windows for claim and recovery. Jobs and notifications services use COMMAND_IDEMPOTENCY_EVENT_RETENTION, which defaults to 336h, cannot be below 168h, and must cover the longest Kafka, published-outbox, or manual redrive window. Runtime agent settings separate the local health endpoint (RUNTIME_AGENT_DOCKER_HEALTH_ENDPOINT) from either an explicit advertised endpoint or an IPv4/IPv6-safe RUNTIME_AGENT_DOCKER_ADVERTISE_HOST/PORT; identity, heartbeat interval, and max concurrency remain independently configurable, and RUNTIME_AGENT_ID must remain stable for the same runtime node across restarts. Workflow worker settings control digest resolution and image-pull locks. Execution worker settings control concurrency, runtime-local image-pull locks, per-workload Docker container memory/CPU/PID limits, leases, retries, recovery, and log publication. Processor settings control batching and cleanup. Outbox settings control enabled topic groups, claims, retries, cleanup, and retention; OUTBOX_RELAY_IDEMPOTENCY_CLEANUP_MAX_BATCHES bounds the number of expired-ledger batches removed per cycle.

Domain-service-specific limits include workflow and job list fetch limits, workflow cleanup settings, and NOTIFICATIONS_SERVICE_CONFIG_FETCH_LIMIT for notification listing.

The Kubernetes observability ConfigMap exports metrics every 2 seconds and batches spans/logs on a 1-second schedule. These intervals avoid turning a 500ms export loop into material idle CPU usage on small service replicas.

Detailed source

The complete current variable inventory remains in docs/configuration.md and the typed files under internal/config.