AWS MGN incorrectly replicating physical disks instead of MD logical volumes in RAID1 setup

When working with AWS Application Migration Service (MGN) and software RAID configurations, the behavior you're experiencing is expected. MGN replicates at the block level of physical disks rather than at the logical volume level.

To properly handle your RAID1 setup with MGN, you should:

1. During agent installation, manually identify only the necessary disks for replication. Since you have a RAID1 configuration, you only need to replicate one of the physical disks from each RAID array, as they contain identical data.

2. When installing the AWS Replication Agent, use the option to manually select disks for replication. You can specify just one of the physical disks (e.g., only nvme0n1) since it contains all the data needed.

3. Note that the root disk is always replicated regardless of your selection, so you cannot exclude it.

4. Be aware that when manually selecting disks, you must use exact disk paths and avoid using apostrophes, brackets, or non-existent disk paths.

5. After installation, verify that only the intended disks are being replicated by checking the "Disk settings" section in the MGN console.

If you've already installed the agent and need to change which disks are being replicated, you'll need to reinstall the agent. When you reinstall, you'll notice the "Total replicated storage" value changes, and a progress bar will appear indicating that MGN is rescanning the original volumes.

For your RAID1 setup, replicating both physical disks is redundant since they contain identical data. By selecting only one disk from each RAID pair, you can reduce the replication volume and avoid potential issues during the migration process.

After migration, you may need to reconfigure the RAID setup on the target instance, as the migrated instance will only have the replicated physical disks without the RAID configuration.
Sources
Agent related - Application Migration Service
Installing the AWS Replication Agent on Linux servers - Application Migration Service

The suggested solution is not working as expected in practice. Issue with Manual Disk Selection: As shown in the installation logs below, when I manually specify only one disk during agent installation, MGN still identifies and replicates BOTH physical disks:

Choose the disks you want to replicate: /dev/nvme1n1

Result:

• Identified volume for replication: /dev/nvme1n1 of size 895 GiB
• Identified volume for replication: /dev/nvme0n1 of size 895 GiB ← Still added automatically

Choose the disks you want to replicate: /dev/nvme0n1

Result:

• Identified volume for replication: /dev/nvme0n1 of size 895 GiB
• Identified volume for replication: /dev/nvme1n1 of size 895 GiB ← Still added automatically

Current Behavior:

• Manual selection is ignored
• Both disks are always replicated regardless of input
• Total replication shows ~1.3TB instead of expected 895GB
• This defeats the purpose of manual selection

Questions:

• Why does MGN ignore the manual disk selection and add both disks anyway?
• Is there a specific parameter or flag needed to enforce single-disk replication with RAID1?
• Are there known limitations with mdadm software RAID that prevent selective replication?
• Is this behavior specific to NVMe drives or RAID1 configurations?

System Details:

Both disks are components of mdadm RAID1 arrays (md2, md3) RAID arrays are healthy and synchronized [UU]

The manual selection feature appears to not function correctly with our RAID1 setup. Could you provide additional troubleshooting steps or alternative approaches? Request: Please escalate to the MGN engineering team if this is a known limitation or bug with software RAID configurations.