Remediation

Review the workload before changing capacity. An INCOMPLIANT result identifies a tuning opportunity; it does not establish a specific safe reduction.

1. Review representative and peak-input runs over the last 14 days. Compare processing duration, input volume, CPU and memory peaks, disk pressure, failures, and completion deadlines. For UNDETERMINED results, first establish the required task activity and positive utilization measurements.
2. Identify the cause of low utilization. Address I/O bottlenecks, insufficient parallelism, or skew when those limit useful processing. Low JVM heap usage alone does not establish that a smaller worker has enough total memory or disk space.
3. Record the existing job configuration, including worker type/count, Glue version, arguments, and run-level overrides. Preserve a rollback copy.
4. In the AWS Glue console, open the job's Job details. For supported workloads, evaluate Automatically scale the number of workers. If already enabled, inspect actual DPU usage and executor demand before changing Maximum number of workers. Otherwise, test a smaller worker type or worker count where the workload supports it.
5. Benchmark one change at a time with representative and peak input data. Compare total DPU consumption, runtime, output correctness, retries, and failures. A smaller allocation that runs longer may not save money. Prioritize changes with a material demonstrated benefit.
6. Deploy an improvement that meets delivery requirements. Monitor subsequent runs and restore the previous configuration if cost, reliability, or completion times regress.