When CM agent found a datanode drive failed, it will try to create a new data directory under / file system. Is there a way to stop CM aget from doing that?
Here is an example of the problem:
1: A datanode has the following data dir /hadoop1 /hadoop2 /hadoop3 on three sepereate disks.
2: /hadoop1 is offline because of disk failure
3: When restart cm agent and datanode service, it will find /hadoop1 is missing from kernel /proc/mounts, the agent will try recreate /hadoop1 but mount it on / filesystem. The / file system will soon be filled up.