or requests (push/pull) › “Connective tissue for AWS services” › Scaling is automatic to event rate › Only pay for what is used – don’t pay for idle › Billed in 100ms increments (against allocated memory)
Can’t run a webserver › Can offload process-intensive tasks – provision less › Doubling memory allocated doubles CPU share › Generous permanent free tier
Container may be re-used (no control though) › Can take advantage of (possible) re-use › e.g. caching in /tmp › Running processes are frozen, then thawed if re-used
› Eliminated the need for an EC2 instance › Reduced the annual cost from ~$750 to ~$2 › 99.73% cost reduction › No longer have to spend time maintaining infrastructure
generates PUT events that Lambda function acts upon › To avoid, use separate input & output buckets/tables/etc › Or use flags to indicate already processed, e.g. S3 custom metadata › This will still involve twice the Lambda invocations