Default shard allocation alogorithm


(Brandon Metcalf) #1

In elasticsearch 1.5.2, is EvenShardsCountAllocator still the default allocation algorithm? We are seeing shards allocated based on disk utilization for clusters using the AWS Elasticsearch service. I have no idea what AWS is using, but hoping someone might be able to shed some light on this. An example,

shards disk.used disk.avail disk.total disk.percent host ip node
0 122.5mb 6.7gb 6.8gb 1 x.x.x.x Perfection
332 951.3mb 28.5gb 29.5gb 3 x.x.x.x Mekano
332 1.1gb 28.3gb 29.5gb 3 x.x.x.x Slyde
0 122.4mb 6.7gb 6.8gb 1 x.x.x.x Hybrid II
0 122.4mb 6.7gb 6.8gb 1 x.x.x.x The Symbiote
664 1.9gb 27.6gb 29.5gb 6 x.x.x.x Karl Malus

The three data nodes here have an even distribution of disk usage, but not shard count.

Thanks.


(Brandon Metcalf) #2

I was looking at the wrong column and it appears even disk usage is not evenly distributed. But, my question still stands; why would shards or disk usage not be evenly distributed?


(system) #3