Add latency to RelayConnectionStats

[yukibtc]

No description provided.

[ok300]

LGTM, I can now sort by latency:

[ok300]

Instead of one thread per relay, maybe it would be better to have one global thread (of the pool?) that checks the latency on all active relays at that moment?

User could add / remove relays, some could go offline or become unresponsive, yet the thread could still query and measure every minute.

[yukibtc]

Instead of one thread per relay, maybe it would be better to have one global thread (of the pool?) that checks the latency on all active relays at that moment?

User could add / remove relays, some could go offline or become unresponsive, yet the thread could still query and measure every minute.

The problem is that if there is a connected relay that is unresponsive, that will block the loop. I can set a timeout for get_events_of but this will increase the time needed to calculate the latency.

I don't think there are problems if we use another thread per relay.
Have you noticed about an increment of RAM or CPU utilization?
I just tested the resources utilization with 16 connected relays and the results are: 0% CPU and 8.3 MB RAM

The RAM will increase if some relays are disconnected since the messages/events will be stored in a channel (max 1024), but in general should be very low.

A problem with this implementation could be that the dev will receive events that not requested in client.notifications()/client.handle_notifications(...).

[ok300]

No, I didn't see any performance impact of 1 extra thread per relay. I was only wondering if it can be further optimized, but as you said, an unresponsive relay could complicate things if one thread is checking all relays. So then I agree, 1 thread per relay is probably the best approach.

A problem with this implementation could be that the dev will receive events that not requested in client.notifications()/client.handle_notifications(...).

Very good point. Then I suppose the NIP-11 Relay Info Document is the least intrusive, since the reply doesn't come as an event? In that case, the latency check would belong behind the nip-11 feature flag.

[ok300]

Thanks!

[yukibtc]

I found an alternative way to ping relays, but unfortunately is not supported on WASM (so no latency stat for WASM).
I'll try to search a solution for that