There are some stats that would be very useful for diagnosing certain common network issues. For instance we can identify packet loss by counting retransmissions, congestion is characterised by a persistently nonempty send buffer, and the TCP keepalive config can be determined from the timer state on each connection.
These stats are system-level concerns and not generally exposed to the users of connections, but at least on Linux we could try and extract these data from /proc/net/tcp{,6} (docs) and expose them in the node stats API if available.
Extracted from #67664 (comment)
There are some stats that would be very useful for diagnosing certain common network issues. For instance we can identify packet loss by counting retransmissions, congestion is characterised by a persistently nonempty send buffer, and the TCP keepalive config can be determined from the timer state on each connection.
These stats are system-level concerns and not generally exposed to the users of connections, but at least on Linux we could try and extract these data from
/proc/net/tcp{,6}(docs) and expose them in the node stats API if available.Extracted from #67664 (comment)