A Lambda function has a 10-second timeout and enough invocations each minute for percentile monitoring. Average Duration remains low while slower executions are increasing.
Warn when the 95th percentile of code-processing duration exceeds 8 seconds in one-minute periods. Monitor failures separately with Errors and logs.
Which metric and statistic best match this duration warning?
Compare Duration p95 with 8,000 ms, separately from Average and actual failure monitoring.
Detailed explanation
Duration measures code-processing time in milliseconds; p95 tracks the 95th percentile.
Duration measures code-processing time in milliseconds; p95 tracks the 95th percentile.
Invocation count is not execution time measured in milliseconds.
Invocation count is not execution time measured in milliseconds.
An average can hide slower executions and is not the requested percentile.
An average can hide slower executions and is not the requested percentile.
Actual failure detection does not replace a p95 duration warning that also includes successful slow executions.
Actual failure detection does not replace a p95 duration warning that also includes successful slow executions.
Check official documentation
Compare the following points with the official documentation.
Review Duration units, cold start exclusion, the p95 definition, and function timeout configuration.
Official references
What to confirm
Convert 8 seconds to 8,000 ms and explain the slowest-5% and end-to-end measurement limits.
Key points
- Duration
- p95
- Timeout
Notes
- Environment: AWS SOA-C03 exam guide and the linked AWS documentation
- Reviewed against the linked AWS documentation on 2026-10-04.
Foundation review
Distinguish a percentile from the slowest execution
p95 does not give the maximum duration of the remaining slowest 5%. Some executions can time out while p95 stays below 8 seconds. Use p99 or Maximum, Errors, and logs when needed; align periods and the function/version/alias scope.
Separate code duration from end-to-end latency
Duration measures event-processing code in milliseconds and excludes cold start time. User latency can also involve initialization and waits elsewhere. Duration alone does not guarantee end-to-end performance; inspect logs and instrumented traces for individual processing intervals.