Amazon SQS insights

For Lambda-SQS integrations, tracking ESM-specific metrics like InvokedEventCount, FailedInvokeEventCount, and FilteredOutEventCount reveals whether filtering, throttling, or batch configuration is limiting throughput.

Event Source Mapping Starvation

Lambda event source mappings for SQS show high FailedInvokeEventCount or DroppedEventCount, indicating Lambda cannot process messages due to throttling, errors, or configuration issues.

Terraform-Managed Lambda Concurrency Exhaustion Leading to S3 Throttling Cascade

When Lambda functions managed by Terraform reach reserved concurrency limits while processing S3 events, backpressure causes S3 request metrics to spike and queue depth to grow in dependent SQS queues, creating a cascading failure pattern across infrastructure components.

Terraform-Provisioned Consumer Lag Detected via SQS Queue Depth and Message Age

SQS queue depth (ApproximateNumberOfMessages) growing alongside ApproximateAgeOfOldestMessage indicates consumers provisioned via Terraform infrastructure-as-code are failing to keep pace with producers, often due to misconfigured CPU/memory limits in ECS task definitions or Lambda concurrency caps.

Messages accumulate faster than consumers can process them, causing queue depth to grow and oldest message age to increase. This indicates consumer throughput is insufficient for producer load.

Messages repeatedly fail processing and accumulate in the dead-letter queue, indicating systematic processing failures that require investigation and remediation.

Messages accumulating in a dead-letter queue (DLQ) indicate systematic processing failures that are being masked by retry logic. Even a single message in the DLQ represents a failure requiring investigation.

ApproximateAgeOfOldestMessage growing beyond expected processing SLAs indicates consumers cannot keep pace with message arrival rate, creating a backlog that will continue to worsen without intervention.

High NumberOfEmptyReceives indicates consumers are polling the queue too frequently and finding no messages, wasting compute resources and incurring unnecessary API costs.

High ApproximateNumberOfMessagesNotVisible indicates messages are being received but not deleted, suggesting consumer processing is slow, stuck, or failing to complete successfully.

Mismatch between NumberOfMessagesReceived and NumberOfMessagesDeleted indicates messages are being consumed but not successfully processed and removed, causing duplicate processing or queue buildup.

High ApproximateNumberOfMessagesNotVisible without corresponding deletes indicates consumers are retrieving messages but failing to process them, often due to Lambda throttling, consumer crashes, or visibility timeout misconfiguration.

High NumberOfEmptyReceives indicates consumers are over-polling the queue, wasting compute resources and incurring unnecessary costs without retrieving messages. This often results from aggressive short-polling configurations.

Disparity between NumberOfMessagesReceived and NumberOfMessagesDeleted indicates messages are being retrieved but not successfully processed and removed, leading to reprocessing cycles and potential duplicate work.

Dead Letter Errors Indicate Systematic Invocation Failures

DeadLetterErrors metric shows events failing to reach DLQ after exhausting retries, indicating systematic failures beyond transient errors. These represent data loss scenarios requiring immediate investigation.