Description

Currently, autoscaling is only concerned with disk space, so in case if an index if configured to auto_expand_replicas and one of the shards hits the disk watermark on a node with limited disk space, we will continue autoscaling the cluster until the cluster gets rebalanced and the disk watermark is removed.

We should more mindful of that condition and consider not to autoscale if one replica shard can't be allocated and let rebalancing do the job.