Per past conversations with @EBernhardson and @dcausse , it may be possible to keep fast searches (autocomplete) fast even when slower searches are overloading a host by:

• Hosting autocomplete indices on a separate cluster
• Setting a latency target in OpenSearch. When this value is set, the cluster measures individual server latency and if the latency is too high, it stops using that host for queries. We can't set this now, because the expected latencies between different types of searches (such as autocomplete vs morelike, etc) are too different.

Creating this ticket to:

• Explore the feasibility of hosting autocomplete indices in a different cluster
• Make a decision whether or not to move forward