workflows

Out-of-order workflow runs: stop stale updates from winning

Separate event, delivery and completion order, then protect the latest record state before an older run writes.

Two changes to one order can start almost together. The newer event may finish first, then a slower run writes the older status over it. Both runs can show success while the final record is wrong.

Short answer

Check the current record version or state before a write that could move it backwards. If arrival order is itself the business rule and the platform supports sequential processing, consider that setting after checking queue and waiting-run limits. Serializing one workflow does not protect against another workflow or person editing the same destination.

Distinguish the four orders

Evidence What it shows What it does not prove
Event time When the source recorded a change Delivery order
Delivery time When the automation platform received it Start or finish order
Run start and finish Observed processing order Which business state should prevail
Current version or state What the authorized source recognizes now Whether every external action completed

This distinction is a Pulse editorial synthesis. Record a stable order or record ID, event ID and source-generated sequence or version when available. Clocks on separate systems can disagree; do not choose a winner from local timestamps alone.

Match the guard to the effect

Situation Candidate guard Limit to test
Actions require arrival order and only this flow writes the destination The platform’s documented sequential processing Queues and incomplete runs can delay later work
Several flows or people edit the same record Conditional write against the current version, if supported by the destination A separate read then write can still race
An old event must not reverse a later transition Fetch authorized state and apply an explicit transition rule The rule must match the specific object
No trustworthy version or state evidence Pause and reconcile Do not infer order from approximate time or value

The matrix is a decision model, not an integrity guarantee. Prefer a destination’s documented atomic version check when available. A check followed by an independent write may be outdated before the write occurs. Test the business rule and rollback path before changing a production flow.

Platform behavior matters

Make documents “Process data in order”. Its webhooks are processed in parallel by default; with the setting enabled, a run waits for the previous one. An incomplete execution can hold subsequent work. This helps when delivery sequence is the correct rule, but adds waiting time. Check the specific scenario settings.

Power Automate documents trigger concurrency and waiting-run limits. Its guidance highlights configuration and capacity consequences of enabling trigger concurrency control. Do not toggle it merely because an order looks wrong; first confirm the cause and how excess arrivals will be handled.

Hypothetical example: order 42 moves from version 1, “received,” to version 2, “approved.” The version 2 run finishes first. Without a guard, version 1 can restore “received.” A destination rule accepting only a version newer than the current one would reject that stale write. This is a proposed test case, not a measured Pulse result.

If timestamps are wrong without overlapping runs, check the time-zone guide. If the same event arrives again, use duplicate-webhook controls. If a record is absent, reconcile missing records before replaying.

Sources and limits

Make supports the sequential-processing description. Microsoft Learn supports the Power Automate concurrency and waiting limits. Version checking is an architecture proposal that requires actual destination support and a defined state rule. Pulse has not applied or tested these settings in a customer operation.

Sources