In order to create effective high memory CloudWatch Alarms for DMS Instances, an effort has been made to observe the behaviour when a DMS Replication Tasks is "Stopped" with and Out of Memory Error.
Observed Behavior.
- Set DMS Instance machine class to something small.
- Start many "Full Load and Ongoing" Migration Tasks simultaneously.
- Eventually, one or more tasks will transition to a stopped state with the "Last Error" being "Out of Memory."
This indicates that there must be a high memory threshold that the instances use to monitor memory. Replication Tasks seem to be stopped with an "Out of Memory" error, while the instance continues and is available. This makes sense in an effort to not crash the dms instance, some tasks are taking out and placed in a "stopped" state.
I have two questions regarding the observance (please correct if observation is not correct).
- Is there any information about which task(s) are chosen as the victim when a memory threshold is reached?
- In order to create a low available memory alert, I would like to stay below the dms instances low memory threshold, is there any information relating to this threshold?
Thanks for the reply