Skip to main content

Fault Tolerance & Disaster Recovery

Enterprise-only feature

Checkpoints, fault tolerance, and disaster recovery features are only available in Feldera Enterprise Edition.

Feldera protects pipelines against data loss through two complementary mechanisms: checkpoints for fault tolerance and object store sync for disaster recovery.

Fault tolerance​

Feldera periodically writes checkpoints — consistent snapshots of the entire pipeline state, including computation and connectors — to local storage. If a pipeline crashes, it automatically restarts from the most recent checkpoint without operator intervention.

Three modes are available, each building on the previous one:

  • Checkpoint and resume — A stopped pipeline writes a checkpoint on exit and restores from it on restart. This enables graceful restarts and configuration changes without data loss.

  • At-least-once fault tolerance — The pipeline writes checkpoints automatically (every 60 seconds by default). If the pipeline crashes, it will resume from the last checkpoint. Output records produced beyond that checkpoint may be sent again.

  • Exactly-once fault tolerance — Extends at-least-once mode with input and output journaling. The pipeline replays journaled data and suppresses duplicate output, so each output record is produced exactly once.

For configuration, connector requirements, and implementation details, see Checkpoints & Fault Tolerance.

Disaster recovery​

Feldera can synchronize checkpoints to an S3-compatible object store, enabling recovery even when local storage is lost. This turns checkpoints into a durable backup that supports several disaster recovery patterns:

  • Standby pipelines — A secondary pipeline continuously pulls the latest checkpoint from object store. If the primary fails, the standby activates and resumes processing within seconds.

  • Pipeline seeding — A new pipeline can bootstrap its state from another pipeline's checkpoints stored in a read-only bucket, without affecting the source.

  • Cross-region recovery — Checkpoints stored in object storage can be restored on infrastructure in a different region or availability zone.

In the case that state cannot be restored from a local or remote checkpoint, the pipeline will need to backfill the data from its source.

For setup, standby configuration, and S3 permissions, see Checkpoint Sync to Object Store.