UDP servers and [services.concurrency]

Can I set services.concurrency for UDP servers? For TCP servers it refers to concurrent connections, but is it interpreted as e.g. packets/s for UDP servers?

Given a UDP server might have a lot of UDP traffic and relatively little TCP traffic (e.g., DNS), I’d like to set a scaling threshold based purely/primarily on the UDP traffic.

This is something we haven’t quite figured out how to expose yet. How would you want to scale instances doing UDP? Would you want to use some packets per second metric, or prefer something like system load?

Packets per second would be great. Ideally that would also be configurable to include a time period (e.g., > 500 queries/s over 30 seconds) to ignore short bursts.

Nice. The plan is to enable scaling on any metric we’re tracking, so once we get that going you should be set.

We do a lot of work to enforce connection limits on VMs through our load balancers, but UDP bypasses all that. We haven’t decided what to do about that yet.

1 Like