Hadware configuration changes since ver 2.*

(Shai ) #1

hi ,

i am planning an upgrade for a cluster to version 2.1.1.

i understood that there are changes at the mechanism for allocation of shards , does it change the recommandation to configure the disks as raid 0 ?


(Ali Beyad) #2

Hi Shai,

The recommended approach if you have multiple disks is to use multiple path.data locations in your path.data setting. The difference you were referring to is that in 1.x, if you had multiple data paths, you could have a single shard located on multiple disks, and so if one disk went down, you would lose the entire shard. In 2.x, a single shard will always go to the same disk, see: https://www.elastic.co/guide/en/elasticsearch/reference/current/breaking_20_multiple_literal_data_path_literal_striping.html

So multiple path.data striping in 2.x is recommended in general, because Raid0 suffers from the same problem of a single shard could go on multiple disks. Raid0 will be faster performance wise, but multiple path.data striping in 2.x will be safer from the perspective of losing data.

Hope this helps.

(system) #3