Workflow Checkpointing

State is snapshotted to CheckpointStorage after each superstep; execution can resume from a chosen checkpoint

Workflow Checkpointing State is snapshotted to CheckpointStorage after each superstep; execution can resume from a chosen checkpoint 01 / Workflow supersteps 02 / Checkpoint storage 03 / Outcomes Running · superstep N · Workflow supersteps · entry 01 Running superstep N entry Superstep · compute delta · Workflow supersteps · work 02 Superstep compute delta work Snapshot · save + check · Workflow supersteps · gate 03 Snapshot save + check gate Completed · final state · Workflow supersteps · terminal 04 Completed final state terminal Checkpoint · thread_id @ step · Checkpoint storage · durable Checkpoint thread_id @ step durable persist next superstep resume(id) Legend active state waiting terminal success failure / exit

Snapshot Per Step

  • • State is saved after every superstep boundary
  • • Each checkpoint is keyed by thread_id and step
  • • Progress is never lost between supersteps

The Superstep Loop

  • • A superstep computes the next state delta
  • • Snapshotting gates the advance to the next step
  • • The loop exits to Completed when no work remains

Resume Anywhere

  • • Saved checkpoints can be listed by id
  • • Execution resumes from any chosen checkpoint
  • • Recovery restarts the workflow from durable state