> > If it were to be included in HBase in any fashion, I think the best way > that could be done would be to modify the Thrift server to have an optional > HTTP port it could listen on for health checks. That way it would be easy > for a non-thrift client such as Nagios or a load balancer to still query the > Thrift server for general health. >
Don't do that. If for some reason the main connection port can't be contacted, but the "health check" port can, then Nagios doesn't alert but service is down. Example: someone firewalls off Thrift port but not HTTP healthcheck port. The server daemon should not be able to tell what's connecting to it is an actual client trying to do real work or a monitoring service (other than that monitoring services tend to ask Really Easy Questions). Easiest way to accomplish this is create a bridge daemon between the protocol the Nagios/loadBalancer speaks and the protocol the server daemon speaks. In this case, HTTP-->Thrift. -- timeless(ness)
