Limit load on shard allocation

On our ES 2.1 cluster, I notice a consistent decrease in search performances during shard allocation and relocation.

Do you have any advice on how to limit the load caused by shard allocation in order to make sure search time is not affected by this process? I would be happy with a slightly longer allocation time if I could ensure search performances were not affected.

Why is reallocation happening?

Check out

Allocation and relocation happens for different reasons, such as

  • rolling restart (plugin installation, ES upgrade)
  • upscaling, downscaling
    and in none of those search speed should be affected.

I figured that my issue is probably related to the performances of EBS volumes on AWS (I'm not using provisioned IOPS).