Select Language:
If you’re experiencing issues with your AWS Database Migration Service (DMS), especially with high CDCChangesDiskTarget values and messages about transaction caching, this might be due to the target database struggling to keep up with the changes being sent. Essentially, DMS caches changes on the replication instance when the target can’t process them quickly enough, which prevents storage from filling up but can cause delays.
You might notice the system pauses reading from the source because the target is slow to write. This can happen even if the replication instance has plenty of memory, indicating that the problem lies with the target database’s processing speed rather than the DMS server.
Here’s what you can do to fix this problem:
First, check if there are any issues with indexing on your target database. Slow writes are often caused by missing or inefficient indexes. Also, look for database locks, as lock contention can prevent DMS from writing data properly.
If you find tables with many transactions or long-running DML operations, consider moving those to a separate task to reduce the load. Adjusting the memory settings—such as increasing MemoryLimitTotal and MemoryKeepTime—may also help improve performance. Additionally, see if you can apply changes in batches more effectively, which can help the target process data faster.
Since CPU usage is very low, it indicates that DMS isn’t waiting on processing power but rather is held up waiting for the target database to catch up. This confirms that the bottleneck is likely at the target side.
To address this, focus on optimizing your target database performance. Improving indexes, reducing locks, or scaling up your database resources can make a big difference. These steps help ensure that DMS can continuously write data without delays.
If you need more detailed guidance, Amazon’s documentation offers great troubleshooting tips for latency and performance issues with DMS. They provide practical steps for diagnosing and solving these kinds of problems.




