Which Opensearch instance type to choose for a new webapplication with little data?

Question

Amazon recommends running an opensearch domain that contains 3 regular nodes and 3 master nodes distributed accross 3 AZ zones. The lowest instance type that is still suitable for a production environment is the `t3 medium.search` instance type.

I've run this setup for about 1.5 days, using `t3 small.search` instead of `medium`. When i looked at the bill afterwards i could see that running such an instance for merely 1.5 days already costs 9 dollars. That's way too expensive for me. According to the [amazon cost calculator](https://calculator.aws/#/addService/OpenSearchService) the monthly cost for this setup would have been well over 350 dollars.

My web application will use the open search server only for serving autocomplete suggestions and finding documents whose coordinates reside within a certain geographical area. When the webapplicatoin is launched the open search server will start out with only 5 indexes containing only a small number of documents, no more then 200 mb in total. Of these only one index is used to preform geospatial queries on.  I don't think i need a t3 medium instance for this.

So my question is: With what kind of open search instance can i start out with? The setup needs to be economical because it will take a while before my web application starts making money. I was thinking about setting up a `t2 micro.search` domain service with 2 micro master instances and 2 micro worker instances. That would cost me about 50 dollars a month in total. Could this be a good setup to start with? If so then i would like to know how i can setup a domain that uses `t2 micro.search` instances. When i go to the domain creation page in my aws console i'm not able to select `t2 micro.search` from the instance type list. The smallest i can select is `t3 small.search` but thats already too expensive for me because i want to run nodes in atleast two availability zones. I could opt for running only one `t3 small.search` master node and one worker node, which would cost 50 dollars a month as well, but then the domain service is no longer highly available. If the availability zone it sits in crashes then i can't serve autocomplete suggestions anymore, nor can i return documents based on their coordinates.

I'd love to hear your opinions on this.

Thank you

Answer

While the instance size will depend on your queries and data usage, that's something you'll be able to increase later on.

If you need to use T2.Micro instances for the Data Nodes, these are [restricted to only Elasticsearch 1.5 and 2.3 engine versions.](https://docs.aws.amazon.com/opensearch-service/latest/developerguide/supported-instance-types.html)
* Click on the **Engine options**
* Select 2.3 or 1.5 under the Elasticsearch section
* Under the instance type, you may have to toggle **Include previous generation instance types** to refresh the options

While this may be able to get you the cheapest instances available, do keep in mind in the future you'll have to upgrade the cluster and it could lead to some development time used.

Another factor we could take advantage of is the purchase of Reserved Instances (RI) to lower your cost, by committing to a 1 or 3 year utilization, but the smallest instance would be a t3.medium.search instance. This would at least put you in the newer version of OpenSearch as well.

Which Opensearch instance type to choose for a new webapplication with little data?

Relevant content