Struct AtomicWriter 

pub struct AtomicWriter<S: ContextStore> { /* private fields */ }

Encapsulates the atomic write-before-complete discipline.

The protocol:

1. Write output to ContextStore
2. If write succeeds → call the completion callback
3. If write fails → do NOT call completion; the Step will be retried
4. If write succeeds but completion fails → on retry, write returns AlreadyExists → the Step knows the output is already durable and can safely re-attempt completion

Implementations

impl<S: ContextStore> AtomicWriter<S>

pub fn new(store: Arc<S>) -> Self

pub fn write_and_complete<F, T>( &self, flow_run_id: FlowRunId, node_id: NodeId, value: &Value, complete_fn: F, ) -> Result<T, AtomicWriteError>

where F: FnOnce() -> Result<T, Box<dyn Error + Send + Sync>>,

Execute the write-before-complete protocol.

complete_fn is called only after the write succeeds. If the write returns AlreadyExists, this is treated as success (idempotent retry) and complete_fn is still called.

Returns the result of complete_fn, or the write error if the write failed for a non-idempotent reason.