Rollback

Restore state to last safe checkpoint

Restore state to a previous checkpoint when things go wrong.

State Snapshots

from splinter.coordination import SharedState

state = SharedState(initial_data={"topic": "AI"})

# Make changes
state.set("research.findings", ["finding1", "finding2"])

# Take snapshot
snapshot = state.snapshot()

# Make more changes
state.set("research.findings", ["bad data"])

# Rollback
state.restore(snapshot)
print(state.get("research.findings"))  # ["finding1", "finding2"]

With Checkpoints

from splinter.coordination import CheckpointManager, FileCheckpointStorage

manager = CheckpointManager(
    storage=FileCheckpointStorage("./checkpoints"),
)

# Automatic checkpoint after each agent
manager.create_checkpoint(
    workflow_id="wf-123",
    step=5,
    agent_id="researcher",
    status=AgentStatus.COMPLETED,
    state=current_state,
    metrics=current_metrics,
)

# On failure, rollback to last good checkpoint
checkpoint = manager.get_latest_checkpoint("wf-123")
state.restore(checkpoint.state_snapshot)

Rollback Strategies

Last Good Checkpoint

# Get the most recent successful checkpoint
checkpoint = manager.get_latest_checkpoint("wf-123")
if checkpoint.status == AgentStatus.COMPLETED:
    state.restore(checkpoint.state_snapshot)

Specific Step

# Rollback to specific step
checkpoint = manager.get_checkpoint("wf-123", step=3)
state.restore(checkpoint.state_snapshot)

State History

# Use built-in state history
state = SharedState()

# Make changes
state.set("key", "value1")
state.set("key", "value2")
state.set("key", "value3")

# View history
history = state.get_history("key")

# Restore from history
state.restore(history[0].snapshot)  # Back to value1

Automatic Rollback

from splinter.coordination import ResumableWorkflow

resumable = ResumableWorkflow(
    workflow_id="wf-123",
    checkpoint_manager=manager,
)

# On error, automatically rollback to last checkpoint
resume_step = resumable.try_resume()
if resumable.is_resumed:
    print(f"Rolled back to step {resume_step}")