A field lab for event-driven systems
A retry is
still work.
See how failures change the work your system must do. Explore a workload, follow a migration, and take the assumptions with you.
FIG. 01 / ATTEMPT LOADSynthetic replay
One thousand events.
30% failure chance per attempt · 2 retries
1,000Initial
300Retry 1
90Retry 2
Total attempts1,390
Exhausted events27
01 / MODEL THE WORK02 / CHOOSE THE BOUNDARY03 / VERIFY THE OUTCOME
The engineering notebook
Start with a decision.
01 / MIGRATION
Express to Lambda, one route at a time
Carry one HTTP route across a deployment boundary. Protect its semantics, isolate slow work and leave yourself a credible rollback.
Read field note 02 / CAPACITYRead a workload before scaling it
Use attempts, occupancy and downstream capacity to make a testable choice about concurrency.
Read field note 03 / RELIABILITYRetries, duplicates and safe replay
Identify the layer that retries work, preserve business identity and build a usable path for events that run out of attempts.
Read field noteHow to read the numbers
Keep the assumptions visible.
We separate a model’s output from a provider’s behavior. Sources, scope and review dates travel with each guide.