intra-pod latency
What is the expected latency between two services running on Railway?
My expectation within the same cloud / AZ is probably 2-3ms (attributable to Envoy / other k8s overhead). Do those numbers look right to you? It certainly feels significantly higher.
Will private networking bring this down? I assume that right now connection is over public internet, if we are behind a VPC, that should cut down on some SSL etc?
14 Replies
Project ID:
a52f72a9-aa58-47ed-a28d-64a3661edb93a52f72a9-aa58-47ed-a28d-64a3661edb93
i don't know about the latency over the internet.
but for latency in private networking, brody made a tool to measure that
https://utilities.up.railway.app/ping?host=hello-world.railway.internal
I was gonna link that myself, but you beat me to it, thank you!!
and for latency to a public service, theres this
https://utilities.up.railway.app/ping?host=hello-world-public.up.railway.app
ack - looks like 7-8 ms via that utility over public internet
after caching, yes
wdym after caching?
after the dns response is cached
ah
maybe ours is just being kept hot, looks like no cold start
not how it works
railway has no "cold start"
im just confused by the dns cache then
no worries, you can always google terms you dont understand 🙂
these tests may be more what youre looking for
heres an http get request to a service through the internal private network
https://utilities.up.railway.app/request-test?url=http://hello-world.railway.internal:3000&method=get
this is also an http get request, to the same service through its public address
https://utilities.up.railway.app/request-test?url=https://hello-world-public.up.railway.app&method=get
these requests do not use any dns caching and do not keep the connections alive
Thank you - this was actually quite helpful to show the proper url: http (not https, no tls handshake!) and referencing a specific port. Private networking is super helpful for us, thank you!
no problem! that's what the utility is for 🙂