Configurations for a large cluster

So here is the overall situation.
We have 8 GB of RAM on 4 nodes of the Keycloak cluster. JVM is configured to use 4 GB of RAM. We have about 80K login requests and 40K refresh token requests every day. About 20K requests for token refresh are failing with the following error.

2021-10-19 08:15:37,133 WARN [] (default task-3174) type=REFRESH_TOKEN_ERROR, realmId=prd, clientId=sampleclient, userId=sampleuserid, ipAddress=, error=invalid_token, grant_type=refresh_token, refresh_token_type=Offline, refresh_token_id=samplerefreshtokenid, client_auth_method=client-secret

NOTE : realmid, clientid, userid, ipaddress and refreshtokenid is changed in the above log statement so that its apt for public posting.