AWSSenior
Lambda function is slow, times out or loops
Lambda timeouts trace to too little memory and CPU, large payloads, a slow downstream service, or a recursive loop between the function and its trigger.
What you see
- The function runs far longer than on your laptop
- Invocations hit the configured timeout
- A function keeps being invoked without end
What it means
When your code runs much longer in Lambda than locally, it is often constrained by the memory or CPU it has. In Lambda, memory and CPU scale together.
Timeouts also come from large payloads, slow downstream services, and loops. A function that is invoked by an event, then changes the same resource that triggers it, invokes itself again.
How to confirm the cause
- 1Check the duration against the timeout
# CloudWatch: look at Duration (p99) and Errors for the function, and the REPORT line in its logs
Compare the longest runs with the configured timeout.
- 2Check the payload size
# The maximum size of a Lambda event payload is 1 MB.
Large payloads mean more work per invocation.
- 3Check the function's downstream calls
# Look for calls to databases, third-party APIs or other services in the function code.
Slow or unavailable downstream services show up as timeouts.
Causes and fixes
Not enough memory (and so CPU)
- HOW TO TELL
- The same code is much slower than locally, and memory use is near the limit.
- ROOT CAUSE
- Memory sets CPU. A small allocation means a small slice of CPU.
- FIX
- Configure the function with more memory, which also raises CPU.
A large payload outruns the timeout
- HOW TO TELL
- Small events succeed; larger ones time out at a predictable item count.
- ROOT CAUSE
- The docs' example: one record per 100 ms with a 3 second timeout works up to about 29 items and fails above 30.
- FIX
- Set the timeout for the largest expected payload, test with upper-bound data, and validate payload sizes.
A slow or unavailable downstream service
- HOW TO TELL
- Timeouts line up with another system's slowness or outage.
- ROOT CAUSE
- Lambda scales fast, and a non-serverless dependency can fall over under the spike. The function then waits.
- FIX
- Handle errors and retries in code, cache results, put a queue such as SQS in front of the dependency, and use a connection proxy such as RDS Proxy for databases.
A recursive loop with the trigger
- HOW TO TELL
- Invocation counts keep climbing, and the function writes to the same bucket or table that invokes it.
- ROOT CAUSE
- Each write creates an event that invokes the function again. The docs describe this for S3, SQS and DynamoDB sources.
- FIX
- Write to a different resource, or a different prefix or suffix than the trigger. Use event filtering, and use recursive loop detection to find these.
How to stop it coming back
- Alarm on
Durationp99 and on throttles. - Load test with maximum-size payloads.
- Never write to the same resource that triggers the function without a filter.
Keep going
- Runbook: Lambda logs missing from CloudWatch
- Runbook: Application Load Balancer 502, 503 and 504 errors
- Cheat sheet: AWS CLI
- Lesson: the cloud topic in Learn
Sources
AWS documentation (docs.aws.amazon.com), read 18 Nov 2026.