Pro ยท Lesson 8 of 11

Validation and CloudWatch

Validation compares rows. CloudWatch shows whether the task is keeping up. Neither one proves the application is ready.

Sign in to save your place.

Picture it this way

Validation is counting the boxes on both docks. CloudWatch is the clock on the truck. A matching count does not mean the shop signs were translated.

Data validation

Turn validation on for a task when you need DMS to compare source and target rows. It adds load. It does not replace a query you wrote for a table you care about. Mismatches show up as a validation state, not as a green full-load percentage.

Validation will not notice a wrong procedure, a missing index you meant to add later, or an application still pointed at the source.

Metrics worth naming

CloudWatch metrics include full-load throughput and CDC latency on the source and the target. The useful question is whether latency is growing. A task that is "running" with climbing CDC latency is falling behind.

Task logs go to a CloudWatch log group. When a table fails, the log line is more useful than the task's overall percentage. A replication instance that is out of memory or storage fails in those logs and in instance metrics. Serverless tasks have capacity metrics instead of an instance class graph. Read the page for the compute you chose.

Signal It means It does not mean
Validation failed Some compared rows differ Which business rule is wrong
CDC latency rising The target is falling behind the source That the task has stopped
Full load 100% The load phase finished That CDC has caught up

Is the copy right, and how far behind?

During a migration the source keeps changing, so you check two things: lag and row equality.

  1. Source databaseStill taking writes
  2. CDC streamChanges on their way
  3. Target databaseThe copy being built

Step 1 of 3The application is still writing to the source while DMS copies it, so the target is always catching up.

Source: 1,250 new orders in the last hour

Lab: three numbers you will read before cutover

Leave the values blank until a real task exists. The list is the habit.

CDC source latency:
CDC target latency:
Validation failures:
Log group:
Table that failed, if any:
Application still on the source: yes/no

Concept summary

Knowledge check

Full load shows 100 percent and CDC latency is climbing. Are you ready to cut over?

What does row validation not prove?

Where do you read a table-level error?

Official sources

Checked against the linked documentation in October 2026. A later release can move a flag, a price, or a supported pair. Use the page for the version you installed.