3 ms·
It is so obvious that it will be a failure. I know like 4 big Google projects and all seem to be failures (Kubernetes being the most successful, but still in h
by erikb 9y ago
It is so obvious that it will be a failure.
I know like 4 big Google projects and all seem to be failures (Kubernetes being the most successful, but still in hype so nobody realizes yet how mediocre it is). Right now it seems a good time to start working on a Google killer.
- plicense 9y agoMind naming the 4 big failed projects?
- dragonwriter 9y ago> I know like 4 big Google projects and all seem to be failures > Right now it seems a good time to start working on a Google killer. If all Google had was the handful of failures you claim to be (rather selectively) aware of, there'd be no need for a Google killer.
- Crash0v3rid3 9y agoCan you clarify on what is mediocre about Kubernetes?
- erikb 9y agoCould you clarify what is not? The only advantage over rolling out docker images by yourself is the scheduler. And if I consider what we wasted amounts of time we already in my team I'd say in that time we could have written our own docker container that can schedule containers on (or from) other hosts. You can easily see that a solution is mediocre if you see that it doesn't attempt to solve any of the hard problems (in this case networking, network debugging, storage, performance) while requiring lots of overhead to do a trivial thing (e.g. here are my static html files, no go and host these). It works okay on AWS/GoogleCloud but at least on AWS you already have all the features of k8s on IaaS level. On everything else it runs shitty and you spend more human resources on getting it to work / keeping it working than when you do the scheduling yourself.