Closed Felixoid closed 4 years ago
I assume that the carbon-clickhouse and the graphite-clickhouse are on the same server as the clickhouse. In this configuration, you can use external load balancers for write (carbon-c-relay) and read (carbonapi)
Implementing a mechanism to recognize shouldn't be a hard task, but I'm not sure about the overhead for a new connection per insert. Maybe, implement a config parameter for it?
Overhead will be very small and optional parameter is not required. PR are welcome
Hello. During huge data migration from the Whisper to ClickHouse I've recognized that the connection to CH is permanent. Because of that, we have an unbalanced loading: only one of 8 CH nodes under an LB receives all data and then redistribute it with a sharding key.
Implementing a mechanism to recognize shouldn't be a hard task, but I'm not sure about the overhead for a new connection per insert. Maybe, implement a config parameter for it?