psql

func AllModels ¶

func AllModels() []any

declare all models in the correct order they should be migrated.

func AsPSQL ¶

func AsPSQL(f any) any

func AsPSQLReplica ¶ added in v0.19.916

func AsPSQLReplica(f any) any

func FetchGcpCloudSqlPassword ¶ added in v0.19.866

func FetchGcpCloudSqlPassword(ctx context.Context, _ database) (string, error)

FetchGcpCloudSqlPassword fetches a GCP OAuth2 access token for use as the password in Cloud SQL IAM database authentication. The token is obtained via the default credential chain (Workload Identity on GKE).

func FetchIamTokenPassword ¶

func FetchIamTokenPassword(ctx context.Context, cfg database) (string, error)

FetchIamTokenPassword fetches an iam token which can be used as a password using the default aws credentials provider

func New ¶

func New(v *validator.Validate,
	l zapgorm2.Logger,
	metricsWriter metrics.Writer,
	lc fx.Lifecycle,
	cfg *internal.Config,
	qc *querycollector.Collector,
) (*gorm.DB, error)

func NewPSQLMigrator ¶

func NewPSQLMigrator(p PSQLParams, lc fx.Lifecycle) *migrations.Migrator

func NewPrimaryListenerConn ¶ added in v0.19.1006

func NewPrimaryListenerConn(ctx context.Context, cfg *internal.Config) (*pgx.Conn, error)

NewPrimaryListenerConn opens a single dedicated, non-pooled connection to the PRIMARY database, suitable for LISTEN/NOTIFY. The caller owns the connection and must Close it.

This deliberately bypasses the pgxpool: a LISTEN binds to a session and pins a connection for its lifetime, which is the wrong thing to take from a shared pool. Because RDS IAM auth tokens rotate (~15m) and pooled connections are recycled on a 5m lifetime, callers should periodically tear this connection down and re-open via this function rather than holding it open indefinitely — each reopen re-runs the IAM token fetch below.

func NewReplica ¶ added in v0.19.916

func NewReplica(v *validator.Validate,
	l zapgorm2.Logger,
	metricsWriter metrics.Writer,
	lc fx.Lifecycle,
	cfg *internal.Config,
	qc *querycollector.Collector,
) (*gorm.DB, error)

NewReplica errors if DBReplicaHost is empty so callers can't silently fall back to the primary.