Error Handling

Each pipeline has independent error handling. A failure in one pipeline does not affect others.

Error policies

error_handling {
  max_retries = 5
  retry_backoff_ms = 1000
  retry_backoff_max_ms = 30000
  on_persistent_failure = isolate
}

Policy	Behavior
isolate	Mark pipeline as ERROR, continue all others. Source ACK advances past this pipeline.
stop_source	Stop all pipelines on this source. Other sources continue.
stop_all	Halt the engine.

Retry

On a transient error (e.g., HTTP 503), the engine retries the change with exponential backoff:

1. Wait retry_backoff_ms
2. Retry the change
3. If still failing, double the backoff (up to retry_backoff_max_ms)
4. After max_retries, apply the error policy

Dead letter store

Failed changes can be captured in a dead letter store for later inspection and replay:

error_handling {
  dead_letter {
    enabled = true
    store = s3
    config {
      bucket = my-bucket
      prefix = "dead-letter/"
    }
  }
}

CLI commands

# List dead letters for a pipeline
laredo dead-letters pg_main:public.audit_log:http-sync

# Replay dead letters (re-deliver to the target)
laredo dead-letters replay pg_main:public.audit_log:http-sync

# Purge dead letters
laredo dead-letters purge pg_main:public.audit_log:http-sync

TTL / expiry

In-memory targets can have automatic row expiry based on a timestamp column:

ttl {
  mode = field
  field = expires_at
  check_interval = 30s
}

Expired rows are removed from the target on the check interval. The engine fires OnRowExpired through the observer for each removal.