Zhivko Todorov
ALL CASE STUDIES

CASE 145 · SABER · 2025

LAMBDACONCURRENCYRESERVEDTHROTTLING

Critical-path Lambdas that throttle other workloads, not the customer.

A B2B platform had 80+ Lambda functions sharing the regional concurrency limit. A spike in batch-processing Lambdas had once throttled the customer-facing checkout flow. We added reserved concurrency for the critical path and provisioned concurrency for the latency-sensitive piece.

INDUSTRY

B2B platform

DOMAIN

RELIABILITY

DELIVERED

2025

STACK

AWS LAMBDA·RESERVED CONCURRENCY·PROVISIONED CONCURRENCY·CLOUDWATCH METRICS·EVENTBRIDGE

RESULTS

What changed, by the numbers.

CHECKOUT THROTTLE EVENTS

0

90 DAYS POST-FIX

CRITICAL-PATH FUNCTIONS

12

RESERVED CONCURRENCY

COLD-START p99

< 50ms

WITH PROVISIONED

BATCH-THROTTLE TRADE

ACCEPTED

BATCH RUNS SLOWER

HOW IT WENT

The post-incident analysis was straightforward. A misbehaving batch job had spawned 800 concurrent Lambda invocations, hitting the account concurrency limit. The checkout-flow Lambda — which was small and well-behaved — got throttled because there was no concurrency reserved for it.

Reserved concurrency on the 12 critical-path functions guaranteed them a floor; provisioned concurrency on the latency-sensitive front-door function eliminated cold-start hits during traffic spikes. The batch jobs got a reserved concurrency cap to keep them well-behaved.

Checkout throttle events dropped to zero in the 90 days post-fix. Cold-start p99 on the customer-facing function is under 50ms. The batch jobs run slower at peak, which the team accepted as a deliberate trade-off — batch is async, customer-facing isn’t.

READY WHEN YOU ARE

Let's get your AWS bill (and architecture) in order.

The discovery call is free. You walk away with at least one concrete idea — even if we never work together.

Or email directly →