Skip to content
Flow Like logo Flow Like Docs

Database

The API stores platform state in PostgreSQL or CockroachDB. Choose the database topology independently of API and execution replica counts.

Internal database

Section titled “Internal database”

database.type=internal deploys one CockroachDB Pod:

database:
  type: internal
  internal:
    replicas: 1
    persistence:
      storageClass: ""
      size: 10Gi

It runs start-single-node --insecure for evaluation and development. Increasing replicas does not form a CockroachDB cluster; the chart requires one internal node. Use an externally operated service for production.

External database

Section titled “External database”
database:
  type: external
  external:
    provider: postgresql
    existingSecret: flow-like-database
  pool:
    maxConnections: 10
    minConnections: 1

The Secret contains a complete DATABASE_URL for both API and migration Job. Set provider=cockroachdb only for a CockroachDB target. Supply the URL through the setup workflow or your existing Secret manager; keep it out of ordinary Helm values.

Use the database operator’s required TLS settings. Private endpoints may need networkPolicy.controlPlaneExtraEgress.

Connection budget

Section titled “Connection budget”

database.pool.maxConnections and minConnections apply per API process, with defaults of ten and one. Multiply the maximum by the greatest API replica count, include rollout surge and other database clients, and leave capacity for administration and migrations.

A minimum of zero lets idle API pools release all connections. The maximum must be positive and at least the minimum. More API replicas can exhaust the database even when individual Pods are lightly loaded.

Schema application

Section titled “Schema application”

database.migration.enabled=true creates a release-specific migration Job. It waits for the database, selects the PostgreSQL or CockroachDB schema, runs pre-push SQL and applies the Prisma schema. API init containers wait for the release’s Job to finish before starting.

The Job is an ordinary release resource, not a pre-install or post-install hook. Its completed result remains available for replacement API Pods until the next upgrade.

The migration command currently uses prisma db push --accept-data-loss. Review schema changes and back up important data before upgrading. If an external migration process owns schema changes, set database.migration.enabled=false and apply the required schema before rolling out the API.

The older workstation scripts/migrate-db.sh has a separate development configuration path. Its --docker mode expects a Compose service absent from this directory. Prefer the chart Job for Helm installations.

Verify and recover

Section titled “Verify and recover”
Terminal window
kubectl get jobs -n flow-like -l app.kubernetes.io/component=db-migration
kubectl logs -n flow-like -l app.kubernetes.io/component=db-migration
kubectl rollout status deployment/flow-like-api -n flow-like
kubectl port-forward service/flow-like-api 8083:8080 -n flow-like

GET /health/ready checks required database tables and columns plus the execution-state store within two seconds. A failed migration or unavailable dependency keeps the API out of Service endpoints.

A Helm rollback does not roll back SQL schema or restore data. Restore database, object storage, service credentials and execution records consistently, then reconcile accepted runs before resuming dispatch.