feat: add logging of ssh connections to agent #8096
Conversation
Signed-off-by: Spike Curtis <spike@coder.com>
Debug is the lowest level, and should be reserved for very detailed logs that let us trace execution flow. I'd like to get away from running with debug logging on by default. I.e. our assumption should be that we don't get debug logs from production scenarios unless we specifically ask a customer to turn them on for a specific process. The volume of SSH sessions coming and going on a single agent should be low. For many use cases, it's only a few a day, and they're a frequent subject of support requests, so we should be logging major lifecycle events (starts, stops, errors) at Info or higher. |
| func (s *Server) Serve(l net.Listener) (retErr error) { | ||
| s.logger.Info(context.Background(), "started serving listener", slog.F("listen_addr", l.Addr())) | ||
| defer func() { | ||
| s.logger.Info(context.Background(), "stopped serving listener", |
There was a problem hiding this comment.
nit: maybe we should also log the total serving time for debugging/support purposes?
There was a problem hiding this comment.
logs have timestamps, so I'm not generally worried about being able to tell how much time elapsed.
Listeners get restarted only on agent reconnections to coderd, which should be pretty obvious from the logs.
There was a problem hiding this comment.
Ok, I was thinking only about reducing the investigation time, even if the information is redundant. Feel free to leave it as is.
Signed-off-by: Spike Curtis <spike@coder.com>
…-ssh-connection-logs
First PR of #7961
Logs SSH connection beginning and end, but I need to modify coder/ssh to log the reason for disconnections.