Struct WorkflowRuntimeBuilder 

pub struct WorkflowRuntimeBuilder<R, H> { /* private fields */ }

Builder for WorkflowRuntime.

Construct via WorkflowRuntime::builder.

Implementations

impl<R: StepRunner, H: TerminalHook> WorkflowRuntimeBuilder<R, H>

pub fn queue_name(self, name: impl Into<String>) -> Self

The Taquba queue name that step jobs are enqueued onto. Defaults to "workflow-steps". Multiple runtimes can share a Queue handle by using distinct queue names.

pub fn memo_prefix(self, prefix: impl Into<String>) -> Self

The object-store path prefix of the Delivery::memo entries. Defaults to "{queue_name}-memo", so runtimes with distinct queue names on one object store have distinct memo namespaces.

pub fn max_concurrent_steps(self, n: usize) -> Self

Maximum number of steps processed concurrently in WorkflowRuntime::run. Defaults to 16.

pub fn poll_interval(self, interval: Duration) -> Self

Maximum time the worker loop waits on an empty queue before re-checking. Defaults to 250ms. The retention sweeps run at this interval.

pub fn memo_retention(self, retention: Duration) -> Self

Enable memo retention with the given window. When set, the runtime writes a terminal marker for every run that reaches a terminal state, and the in-process sweeper clears that run’s memo entries and terminal record retention after termination. When unset (default), no marker is written and memo entries and terminal records are retained indefinitely.

pub fn step_output_replay(self) -> Self

Enable content-addressed replay of runner-returned step outcomes.

When enabled, the runtime writes every StepOutcome the runner returns, including Fail and Cancel, to object storage before applying it. Step errors (StepError) are not recorded, so retries still invoke the runner. The replay key is scoped to (run_id, step_number, SHA-256(step payload)). If the same step is delivered again after a crash before ack, the stored outcome is replayed without invoking the runner again. The record includes the effects staged through Delivery::effects, so a replayed outcome applies them as well. A replayed StepOutcome::Continue with a Trigger::After delay reduces the delay by the time already elapsed since the outcome was stored, preserving the original schedule.

This is disabled by default because it adds one object-store read per step delivery (the replay lookup) plus one write per recorded outcome, and makes that write part of step settlement.

pub fn clock(self, clock: Arc<dyn Clock>) -> Self

Override the Clock the runtime reads its timestamps from. Defaults to the same clock the Queue was opened with (via Queue::clock).

pub fn group_retention(self, retention: Duration) -> Self

Remove a RunGroup’s state (its manifest, member records and the memo entries and terminal records of its members) retention after a RunGroup::results consumer observed the last member’s termination, through a sweep when the worker starts and at every poll interval after that. When unset (default), no group terminal marker is written. A group whose results are never consumed is retained until RunGroup::forget in either case.

pub fn build(self) -> WorkflowRuntime<R, H>

where H: 'static,

Finalize the builder.