- trace: for tracing messages from start to finish, including unroutable/discards [1]
- debug: state changes that we only need to know when debugging, client conns opening and closing [1]
- info: harmless, but interesting not-so-common events. e.g. outbound connection changes, manually triggered flushes, etc. (this used to be
notice
) - warn: minor issues (network timeouts etc)
- error: recoverable errors
- fatal: errors and problems that result in shutdown
- panic: not used
logging related to instances of objects:
{conn,dest,..} <spec>
where spec is typically the addr or listening port
[1] these metrics are potentially high volume and resource intensive