Durable execution
Durable execution is a way of running a workflow in which each completed step is committed, so the run can resume from that point after a restart, a failure, or a long wait instead of starting over.
Why business workflows need it
A business process rarely finishes in one sitting. An invoice waits for an approver who is on leave. A connection to the ERP drops halfway through a batch. The server running the workflow is restarted for an update. Without durable execution, each of those events either loses the work or forces someone to work out by hand which steps already happened.
Starting over is not free either. Re-running a finished step can repeat a model call you already paid for, read a record that has since changed, or send a second write to a system that accepted the first.
How Runtara applies it
Every Runtara workflow run resumes from its last committed step.
- Completed steps are kept when a later step fails, so a failure costs only the step that failed.
- A failed step is retried with its saved inputs and state, rather than recomputed from scratch.
- Waiting for approval does not consume active workflow compute. A run held for a person can wait for days.
In the illustrative invoice run INV-0847, a variance is held for approval. The run continues 23 days later from its committed state, after a runtime restart, with zero steps repeated. The extraction, the purchase order lookup, and the match are not run again.
What it does not do
Durable execution makes the workflow’s own progress safe to resume. A step that writes to an external system still has to be designed so that a retry is safe, for example by checking whether the record already exists before creating it. Runtara records each call and its response in the run, which makes that check straightforward to build and to audit afterwards.