WaitForNotification should detect lost connections about pgx

[jackc]

I added working code in the commit 251d54d to branch lost-conn. I haven't merged it in yet as I'm not sure how I feel about it. On the plus side, it should definitely work. Keep-alive and Exec'ing a nop in the WaitForNotification loop did not detect lost connections (in a reasonable amount of time) on my linux workstation when physically pulling the network cable. This goes back to the previous discussion about TCP keep-alive being highly variable across platforms. This new approach uses read deadlines and should behave correctly and consistently across platforms.

It works as is, but I'm curious if what is really needed is package wide lost connection / timeout management. Because it is still possible in Query/Exec to block until keep-alive finally did its thing. I'm not sure exactly what that would look like, so maybe this should just be merged in for the time being. It doesn't expose any public methods/values so it can be changed freely later...

[jackc]

Oh, I also updated https://github.com/jackc/pgx_notify_lost_connection to use this new version and it does show it working.

[flowchartsman]

What are your concerns about this code?

[jackc]

No functional concerns, just wasn't sure how it felt stylistically. Wasn't sure if I liked having to adjust the SetReadDeadline in the middle like that. I'll probably merge it in a day or two, just wanted to let it sit a little and come back later with fresh eyes.

[jackc]

Rebased and merged in 5edd660.