AWS SQS patterns for reliable background job processing

Architecture

Producer → SQS Queue (visibility_timeout >= Lambda timeout)
                    ↓ after maxReceiveCount = 3
           Dead Letter Queue → CloudWatch alarm on messages > 0

Terraform setup

resource "aws_sqs_queue" "dlq" {
  name = "${var.service}-dlq"; message_retention_seconds = 1209600
}
resource "aws_sqs_queue" "worker" {
  name = "${var.service}-queue"; visibility_timeout_seconds = 300
  redrive_policy = jsonencode({
    deadLetterTargetArn = aws_sqs_queue.dlq.arn; maxReceiveCount = 3
  })
}
resource "aws_cloudwatch_metric_alarm" "dlq" {
  alarm_name = "${var.service}-dlq-not-empty"
  comparison_operator = "GreaterThanThreshold"; evaluation_periods = 1
  metric_name = "ApproximateNumberOfMessagesVisible"; namespace = "AWS/SQS"
  period = 300; statistic = "Maximum"; threshold = 0; alarm_actions = [var.alert_sns_arn]
  dimensions = { QueueName = aws_sqs_queue.dlq.name }
}

Lambda with partial batch failure

def handler(event: dict, context) -> dict:
    failures = []
    for record in event['Records']:
        try:
            process_message(json.loads(record['body']))
        except Exception as e:
            failures.append({'itemIdentifier': record['messageId']})
    return {'batchItemFailures': failures}

resource "aws_lambda_event_source_mapping" "worker" {
  event_source_arn        = aws_sqs_queue.worker.arn
  function_name           = aws_lambda_function.worker.arn
  batch_size              = 10
  function_response_types = ["ReportBatchItemFailures"]
  scaling_config          { maximum_concurrency = 10 }
}

The most common mistake

visibility_timeout MUST be >= Lambda timeout

If Lambda runs 60s and visibility_timeout = 30s:
  Message becomes visible again while STILL being processed
  Another consumer picks it up → processed twice

Step2Dev includes SQS templates for async processing projects.

👉 step2dev.com