Re:A Service Mesh for Kubernetes, Part I: Top-Line Service M...
by theptip on Wednesday October 05, 2016 @05:32PM
(#12648096)
kube-proxy (the k8s per-node service agent) is a good generic solution, but has some issues.
The primary issue is that it doesn't support advanced use-cases like circuit breaking or retry budgets (it's just a set of iptables rules).
Also, I believe linkerd uses IPC, which wil be quicker for intra-node communication between processes[1] (i.e. east-west communications).
Not sure how linkerd does it, but it's possible to bypass the kube-proxy and directly watch the set of pod IPs that a service is selecting, so linkerd can avoid hitting the default kube-proxy if it wants to.
The primary issue is that it doesn't support advanced use-cases like circuit breaking or retry budgets (it's just a set of iptables rules).
Also, I believe linkerd uses IPC, which wil be quicker for intra-node communication between processes[1] (i.e. east-west communications).
Not sure how linkerd does it, but it's possible to bypass the kube-proxy and directly watch the set of pod IPs that a service is selecting, so linkerd can avoid hitting the default kube-proxy if it wants to.
[1]: http://bhavin.directi.com/unix-domain-sockets-vs-tcp-sockets...