Pro ยท Lesson 10 of 11
Pick compute on purpose. Size it for the log, not for the sales diagram. When a task fails, read the log and the network before you change the mapping.
Sign in to save your place.
A rented truck costs money while it sits in the yard. A yard that hands you trucks still has a minimum and a maximum. Neither truck fixes a locked gate.
A replication instance has a class and a disk. It is billed while it exists, including evenings when the task is idle, until you stop or delete it. Storage that fills will stall the task. Multi-AZ keeps a standby and costs more. Read the current DMS pricing page for the class. Do not reuse a price from this course.
A serverless replication config asks for a minimum and maximum capacity in DMS capacity units. AWS has changed how idle capacity is billed. The current serverless pricing and user-guide pages win over any sentence that says it always scales to zero or never does. Schema conversion projects are also called serverless and they are not this capacity range. Do not mix the two bills.
| Compute | You manage | You look up |
|---|---|---|
| Replication instance | Class, storage, Multi-AZ, and when to delete it | The hourly price of that class |
| Serverless replication | Minimum and maximum capacity | How idle capacity is billed this month |
| Schema conversion project | The migration project, not an instance class | S3 and related charges the guide lists |
Test the endpoint again. Read the task log in CloudWatch. Check source log retention and disk. Check target disk and connections. Look at CDC latency and instance free memory or serverless capacity. Only then change table mappings.
Swap files on a replication instance are a sign the class is too small for the LOB or the parallelism you chose. A bigger instance is a test, not a personality.
Private subnets, security groups that name the DMS client, secrets instead of pasted passwords, and IAM roles that can read those secrets and write logs. Encryption in transit is the endpoint SSL mode the database requires. Encryption at rest is the storage setting on the databases and on DMS storage, documented per feature. Turn on what those pages describe. Do not assume a new account already did.
Step 1 of 3An instance you forgot to delete keeps billing.
Step 2 of 3Serverless stays inside the range. Confirm the idle rule on the current price page.
Step 3 of 3Connection, task log, then the source disk that holds the change log.
Use this the next time a task is not green. Guessing a bigger instance is not step one.
Endpoint test: pass/fail
Task log line:
CDC latency:
Source log disk free:
Target connections:
Compute: instance class or serverless min/max
Change I will NOT make until the log is read:
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.