Dms Diagnostics

AWS DMS Diagnostics

When to use

Any AWS DMS investigation where the console alone is insufficient — task failures, CDC lag, endpoint connectivity, replication instance issues, schema conversion, data validation, performance tuning, or security configuration.

Investigation workflow

Step 1 — Collect and triage

aws dms describe-replication-tasks --filters Name=replication-task-id,Values=<task-id>
aws dms describe-replication-instances --filters Name=replication-instance-id,Values=<instance-id>
aws dms describe-endpoints --filters Name=endpoint-id,Values=<endpoint-id>
aws dms describe-table-statistics --replication-task-arn <task-arn>

Step 2 — Domain deep dive

aws dms describe-replication-task-assessment-results --replication-task-arn <task-arn>
aws dms test-connection --replication-instance-arn <instance-arn> --endpoint-arn <endpoint-arn>
aws cloudwatch get-metric-statistics --namespace AWS/DMS --metric-name CDCLatencySource --dimensions Name=ReplicationInstanceIdentifier,Value=<instance-id> --start-time <start> --end-time <end> --period 300 --statistics Average
aws dms describe-event-subscriptions

Step 3 — Detailed investigation

aws dms describe-replication-task-individual-assessments --filters Name=replication-task-assessment-run-arn,Values=<run-arn>
aws cloudtrail lookup-events --lookup-attributes AttributeKey=EventSource,AttributeValue=dms.amazonaws.com --max-results 20
aws cloudwatch get-metric-statistics --namespace AWS/DMS --metric-name FreeStorageSpace --dimensions Name=ReplicationInstanceIdentifier,Value=<instance-id> --start-time <start> --end-time <end> --period 300 --statistics Average

Read references/dms-guardrails.md before concluding on any DMS issue.

Tool quick reference

Gotchas: AWS DMS

• CDC requires supplemental logging on the source. Without it, DMS cannot capture changes. Oracle needs supplemental logging at database or table level. MySQL needs binlog_format=ROW. PostgreSQL needs wal_level=logical.
• LOB columns can drastically slow migration. Full LOB mode reads each LOB individually. Limited LOB mode truncates at max size. Inline LOB mode is fastest but has size limits. Choose the right mode for your data.
• Replication instance storage fills up during large migrations. DMS uses local storage for sorting, caching, and transaction logs. Monitor FreeStorageSpace. Increase storage or instance size proactively.
• Table mappings control what gets migrated. Selection rules choose tables. Transformation rules rename or modify. Incorrect wildcards can include/exclude wrong tables. Test mappings before full migration.
• Data validation runs after full load and during CDC. It compares source and target row by row. Validation adds overhead. Mismatches may be due to ongoing changes, not actual errors.
• Multi-AZ replication instances provide HA but not performance. The standby is for failover only. It does not handle read traffic. Multi-AZ doubles the cost.
• VPC peering or VPN may be needed for on-premises sources. The replication instance must reach both source and target. Security groups, NACLs, and route tables must all allow traffic.
• Task settings JSON controls detailed behavior. BatchApplyEnabled, ParallelLoadThreads, MaxFullLoadSubTasks all affect performance. Default settings are conservative.
• Schema conversion and DMS are separate tools. AWS SCT converts schema; DMS migrates data. Use SCT first for heterogeneous migrations.

Anti-hallucination rules

1. Always cite specific task ARNs, endpoint IDs, or API responses as evidence.
2. CDC requires source-specific configuration. Never assume it works out of the box.
3. LOB handling mode significantly impacts performance. Never ignore LOB settings.
4. Replication instance class affects throughput. Never assume default is sufficient.
5. Data validation mismatches during CDC may be transient. Never declare failure without re-checking.
6. Spend no more than 2 minutes on any single hypothesis. Pivot if inconclusive.

26 runbooks