Tiered Storage RAM_DISK

[#1]

I am following:

https://docs.hortonworks.com/HDPDocuments/HDP2/HDP-2.6.1/bk_hdfs-administration/content/configure_me...

I've been able to set it up for SSD but can't get RAM_DISK to work. I can create the ramdisk at the linux level, attach a storage policy and write to it outside of Hadoop. It appears that HDP set’s up its dir structure on the ramdisk but all write attempts end up on regular [DISK]. When I attempt to specify the dfs.datanode.data.dir using the URI (i.e [RAM_DISK]file:///ramdisk/hdfs) as detailed in the doc mentioned above,the datanode fails to start, so I use [RAM_DISK]/ramdisk/hdfs.

The document mentioned above seems to have some inconsistencies:

a) Says to use dfs.data.dir not dfs.datanode.data.dir

b) Says to set dfs.checksum.type to NULL.

When the datanode is starting a warning message pops up that says the checksum type is invalid and is reverting to CRC.

Anyone with some experience to share on TIERED STORAGE?

[#2]

dfs.data.dir is deprecated by dfs.datanode.data.dir in new version of Hadoop. Both should work in HDP 2.6.1. Please check and ensure entries are comma separated without any extra space.

bq. "When I attempt to specify the dfs.datanode.data.dir using the URI (i.e [RAM_DISK]file:///ramdisk/hdfs) as detailed in the doc mentioned above,the datanode fails to start."

@Constantin Stanca, can you post the datanode log when datanode failed to start?