A latency-sensitive Lambda function behind a synchronous API shows high p99 latency caused by cold starts during traffic bursts. The team must reduce the number and duration of cold starts using Lambda configuration, and they accept paying for warmed capacity. Which TWO configuration changes directly reduce cold starts for this function? (Select TWO.)
- AConfigure provisioned concurrency on the function alias the API points to so a set number of execution environments are initialised and kept warm. Correct
- BIncrease the function timeout from 30 seconds to 300 seconds so each invocation has more time to finish.
- CRaise the memory allocation, which proportionally increases the CPU available during the initialisation phase and shortens cold-start duration. Correct
- DSet a higher reserved concurrency limit on the function so more environments are available during the burst.
- EMove large dependencies into a Lambda layer so the deployment package is smaller and downloads more quickly.
Why A is correct: Provisioned concurrency pre-initialises execution environments before invocations arrive, so requests skip the init phase that causes cold starts.
Why B is wrong: Tempting because timeout is a runtime setting, but timeout governs how long a running invocation may take and does nothing to remove the init delay of a cold start.
Why C is correct: Memory is the proportional CPU lever in Lambda; more CPU during init runs the runtime and handler setup faster, cutting cold-start duration.
Why D is wrong: Reserved concurrency only caps and guarantees a share of concurrency; it does not pre-warm any environment, so the first invocation in each new environment still cold-starts.
Why E is wrong: A layer is still downloaded and unpacked into the same environment at init, so total code to initialise is unchanged and cold-start time is not meaningfully reduced.