as sql errors and performance are such common data source, we wrote sql error to a file and ingest with regular monitor stanza at this kind of scale.
we saved DB connect inputs for important data in table format.
you can always scale you HF layer and have as many as required with DB Connect on them to sustain all 2000 connections.
as i am not regarding the intervals you are pulling data at, or the volume of the data and the length of the queries, its tough to calculate how many HF youll need...
read here forward and plan according to your requirements and conditions:
https://docs.splunk.com/Documentation/DBX/3.2.0/DeployDBX/Architectureandperformanceconsiderations
hope it helps
... View more