3 ms·

I used to be that sysadmin, writing config to set that all up. It was far more labor intensive than today where as a dev I can write a single manifest and have

by haiku2077 1y ago

I used to be that sysadmin, writing config to set that all up. It was far more labor intensive than today where as a dev I can write a single manifest and have the cluster take care of everything for me, including stuff like configuring a load balancer with probes and managing TLS certificates.

darkwater 1y ago
Nobody is denying that. But GP was saying that now with k8s developers don't need to know about the network. My rebuttal is that devs never had to do that. Now maybe even Ops people can ignore some of that part because many more things are automated or work out of the box. But the inner complexity of SDNs inside k8s in my opinion is higher than managing your typical star topology + L4 routing + L7 proxies you had to manage yourself back in the days.
- motorest 1y ago
  > But GP was saying that now with k8s developers don't need to know about the network. My rebuttal is that devs never had to do that. The only developers who never had to know about the network are those who do not work with networks.
- otterley 1y ago
  I think a phone call analogy is apt here. Callers don’t have to understand the network. But they do have to understand that there is a network; they need to know to whom to address a call (i.e., what number to dial); and they need to know what to do when the call doesn’t go through.
- dilyevsky 1y ago
  Devs never had to do that because Moore's Law was still working, the internet was relatively small, and so most never had to run their software on more than one machine outside of some scientific use-cases. Different story now.
p_l 1y ago
Which is why you often had to wait weeks for any change. Hell, in some places, Ops are pushing k8s partially because it makes DNS and TLS something that can be easily and reliably provided in minimal amount of time, so you (as a dev) don't have a request for DNS update wait 5 weeks while Ops are fighting fire all the time.

darkwater 1y ago

Nobody is denying that. But GP was saying that now with k8s developers don't need to know about the network. My rebuttal is that devs never had to do that. Now maybe even Ops people can ignore some of that part because many more things are automated or work out of the box. But the inner complexity of SDNs inside k8s in my opinion is higher than managing your typical star topology + L4 routing + L7 proxies you had to manage yourself back in the days.

motorest 1y ago
> But GP was saying that now with k8s developers don't need to know about the network. My rebuttal is that devs never had to do that. The only developers who never had to know about the network are those who do not work with networks.
otterley 1y ago
I think a phone call analogy is apt here. Callers don’t have to understand the network. But they do have to understand that there is a network; they need to know to whom to address a call (i.e., what number to dial); and they need to know what to do when the call doesn’t go through.
dilyevsky 1y ago
Devs never had to do that because Moore's Law was still working, the internet was relatively small, and so most never had to run their software on more than one machine outside of some scientific use-cases. Different story now.

p_l 1y ago

Which is why you often had to wait weeks for any change. Hell, in some places, Ops are pushing k8s partially because it makes DNS and TLS something that can be easily and reliably provided in minimal amount of time, so you (as a dev) don't have a request for DNS update wait 5 weeks while Ops are fighting fire all the time.