3 ms·
A Production Outage Was Caused Using Kubernetes Pod Priorities
- reilly3000 7y agoAfter spending a few months getting neck deep into K8s I’m resolved to steer clear of it until 2020-2021 while the smart people work out all of the sharp edges. Picking a tooling stack is complicated, and safe deployments still seem to be a matter of technique rather than technology. I have no doubt the ecosystem will reach a steady state, but until then Fargate ‘Just Works’ for me.
- lowdose 7y agoI have the same hindsight knowledge after using Tensorflow & Polymer 1.0 in production. It almost seems like Google is trying to systemize so many functionality at once it is bound to fail. Every trade-off is essentially a bet in tech and if I would have had this insight at the decision moment I would have picked PyTorch & React instead. Lesson learned dont let your opinion of a business model cloud judgement on choices in the stack. Facebook has made some remarkable innovative open source projects that just work for the first mvp launch. React went against all advise of keeping styling & content separated. PyTorch compiled the graph on the fly while tensorflow required a very specific procedure to define the graph on forehand, in TF 2.0 they caught up with PyTorch.
- reilly3000 7y agoI think Google abstracts around a level of complexity that only applies to the world’s largest companies. Facebook has a clearer sense of the user since that is their primary focus, whether that be the end user or developers.