3 ms·

The problem is the layering of the abstractions. Having PaaS on top of Kubernetes in a way that you move down to a more powerful primitive is going to be the mo

The problem is the layering of the abstractions. Having PaaS on top of Kubernetes in a way that you move down to a more powerful primitive is going to be the most scalable. What happened in the public cloud space was a disparate set of services which made the choices really costly. Azure went in heavy on PaaS while AWS was focused on VMs. The easy adoption and migration was on AWS. K8s seems like a reasonable abstraction that reduces the cost of these decisions because people seem to agree that K8s is a reasonable base. Being able to slide on a compatible PaaS layer shouldn't negate that K8s investment, and make it easier to adopt for some organizations, or parts of those organizations. But, I wouldn't argue that we are the point where that layer is mature.

debarshri 5y ago
I do not agree with you. PaaS abstraction becomes super leaky on top kubernetes as you start solving real world problems. That is one the first mistakes newly formed platform teams in a company makes, abstracting infrastructure. Happy flow of PaaS on kubernetes is great. It is magical to certain extent. But negative flows is a really mess. The customer no clue why certain things do not work and you as a engineer had no clue that this would have happened. Secondly, the amount breaking changes you have deal with as a result of abstraction is crazy. People pushing code does not care. It is basically moving complexity from one place to another.
westurner 5y ago
dokku-scheduler-kubernetes https://github.com/dokku/dokku-scheduler-kubernetes#functionality https://github.com/dokku/dokku-scheduler-kubernetes#function... > The following functionality has been implemented: Deployment and Service annotations, Domain proxy support via the Nginx Ingress Controller, Environment variables, Letsencrypt SSL Certificate integration via CertManager, Pod Disruption Budgets, Resource limits and reservations (reservations == kubernetes requests), Zero-downtime deploys via Deployment healthchecks, Traffic to non-web containers (via a configurable list)

debarshri 5y ago

I do not agree with you. PaaS abstraction becomes super leaky on top kubernetes as you start solving real world problems. That is one the first mistakes newly formed platform teams in a company makes, abstracting infrastructure. Happy flow of PaaS on kubernetes is great. It is magical to certain extent. But negative flows is a really mess. The customer no clue why certain things do not work and you as a engineer had no clue that this would have happened. Secondly, the amount breaking changes you have deal with as a result of abstraction is crazy. People pushing code does not care. It is basically moving complexity from one place to another.

westurner 5y ago

dokku-scheduler-kubernetes https://github.com/dokku/dokku-scheduler-kubernetes#functionality https://github.com/dokku/dokku-scheduler-kubernetes#function... > The following functionality has been implemented: Deployment and Service annotations, Domain proxy support via the Nginx Ingress Controller, Environment variables, Letsencrypt SSL Certificate integration via CertManager, Pod Disruption Budgets, Resource limits and reservations (reservations == kubernetes requests), Zero-downtime deploys via Deployment healthchecks, Traffic to non-web containers (via a configurable list)