tcp_connection_duration should be measured in seconds

[olix0r]

We currently measure connection durations in milliseconds. It's probably more appropriate to measure connection durations in terms of seconds. In either case, we should use a histogram that makes sense for connection durations:

tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="1"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="2"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="3"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="4"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="5"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="10"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="20"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="30"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="40"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="50"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="100"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="200"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="300"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="400"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="500"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="1000"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="2000"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="3000"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="4000"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="5000"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="10000"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="20000"} 0
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="30000"} 1
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="40000"} 1
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="50000"} 1
tcp_connection_duration_ms_bucket{direction="inbound",peer="src",tls="true",errno="ENOTCONN",le="+Inf"} 5
tcp_connection_duration_ms_count{direction="inbound",peer="src",tls="true",errno="ENOTCONN"} 5
tcp_connection_duration_ms_sum{direction="inbound",peer="src",tls="true",errno="ENOTCONN"} 28334607

[zaharidichev]

@olix0r Based on your observations, what seconds - measured buckets would make more senes ?

[psinghal20]

Hi @olix0r, I am new to this project and would love to take up this issue if it is still relevant. Can you help me out with some pointers to tackle this issue?

[psinghal20]

hi @olix0r, I looked through the codebase and found latency::Ms being used for the calculation of connection_duration. I added a similar implementation for latency::Sec using the same bounds. I modified the EosMetrics to use this new type and changed its name to tcp_connection_duration_s. Did I get it correct? If yes, I can raise a PR.