In case of I have 30TB storage(6 spinning 5TB disks with stripe(raid 0)) and 64GB RAM x 4 nodes,
RAM is going insufficient but storage just used about 10%.
Indices have no replica.
To improve availability I have two options,
set indices replica to 1.
change disk raid to 1+0 ( 1mirror)
For 1, I worry about the lack of memory for the new replica shard.
(May replica shard will consume RAM amount of just like primary shard does, right?)
For 2, I heard mirroring is not recommend for es(they say just use replica!)
hello jihun,
I do not have much to say about RAM usage. I guess there is no option to change the hardware? That is a lot of disk compared to the amount of RAM.
As for disk layout, RAID 10 will make the filesystem more reliable but does little to improve ES availability. If one node goes down, the shards allocated to that node are gone.
For any application where data loss is not tolerated you really should have at least 1 replica.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.