Shrink an Amazon EBS root volume with minimal instance downtime

Although my migration attempt cross different regions failed,
I successfully shrink the os volume from 8G to 4G. Here is how:

1. snapshot the running os volume
2. create a new volume use the snapshot, and attach to /dev/big
3. create an empty volume with smaller size, and attach to /dev/small
4. operate on the big volume

   e2fsck -f /dev/big
   
   resize2fs -M -p /dev/big
   # the command above shrink the volume

5. mirror the content from the big to the small volume

   dd if=/dev/big ibs=16M of=/dev/small obs=16M count=128
   # count needs to be adjusted; mine was 415315 4k block, which was about 101 16M block
   
   resize2fs -p /dev/small
   # the command above expand the volume to full space
   
   e2fsck -f /dev/small

6. stop the running instance, replace the os volume with the small and all set