It's possible for an Elasticsearch cluster spread across multiple availability zones, and with sufficient replicas, to still be vulnerable to problems if an entire availability zone failed.
This could happen if the a replica and primary shard are in the same availability zone, but different instances.
It's possible for an Elasticsearch cluster spread across multiple availability zones, and with sufficient replicas, to still be vulnerable to problems if an entire availability zone failed.
This could happen if the a replica and primary shard are in the same availability zone, but different instances.
Elasticsearch has a built in way to handle this: allocation awareness