Kafka Event Poller Autoscaling Lag Indicator

warning

scalingUpdated Feb 23, 2026

Lambda's on-demand Kafka event pollers scale based on offset lag evaluation every minute, but the autoscaling process takes up to three minutes to complete. High offset lag combined with low event poller counts indicates insufficient polling capacity before autoscaling can respond.

Sources

Apache Kafka event poller scaling modes in Lambda - AWS Lambdadocs.aws.amazon.com

Technologies:

Apache KafkaThe root cause of this issue originates in Apache Kafka

kafka.consumer.max_lag

kafka.consumer.bytes_consumed

How to detect:

Track Kafka OffsetLag or MaxOffsetLag rising while Lambda's provisioned or on-demand event pollers remain constant or low. In provisioned mode, monitor ProvisionedPollers and EventPollerThroughputInBytes against throughput capacity limits (5 MB/sec per poller).

Recommended action:

Switch to provisioned mode for Kafka event source mappings with predictable high throughput. Set MinimumPollers based on baseline throughput (divide expected MB/sec by 5). For spiky workloads, configure MaximumPollers to handle surges and set up CloudWatch alarms on OffsetLag with 2-minute evaluation periods to account for autoscaling delay.