Dropbox уходит из амазона

http://www.wired.com/2016/03/epic-story-dropboxs-exodus-amazon-cloud-empire/

• Мало деталей, рассуждения на тему почему большие компании уходят из облаков.
• Немного коснулись Rust, что они с Go переписали часть сервисов на Rust
• В комментах на HN:
  • heap size is the issue. How little memory and CPU is used to manage 1GB of data. Perfect hash tables, extensive bit-packing. Lot of custom inline cache-friendly data structures. Data on stack to take the pressure off the VM.
  • CPU optimisation is easier to do in Rust than in Go
  • GC pauses is not an issue.
  • powerful abstractions, no null, no segfaults, no leaks, yet C-like performance and control over memory and you can use that whole C/C++ bag of optimization tricks.
  • very close to Rust team! Rust developers often go to Dropbox offices
  • Big ask - improve compile time
  • pain to work around unstable API (dust hasn’t settled yet in Rust core team)
  • Comment from Rust Core team: “If you use Rust in production, we want to hear from you! We’re very interested in supporting production users.”

Kubernetes-1.2.0

http://blog.kubernetes.io/2016/03/Kubernetes-1.2-even-more-performance-upgrades-plus-easier-application-deployment-and-management-.html

• scale: 30k pods per cluster
• deployments: versioning, multiple simultaneous rollouts, aggregating status across pods, http://kubernetes.io/docs/user-guide/deployments/
• Dynamic Configuration (via ConfigMap API): приложения могут вытаскивать конфигурацию во время старта, вместо запаковывания в образ во время создания образа. More info: https://github.com/kubernetes/kubernetes/blob/master/docs/design/configmap.md
• multi-zone scheduling in AWS
• graceful node shutdown (node drain): http://kubernetes.io/docs/user-guide/kubectl/kubectl_drain/
• Custom Metrics for Autoscaling
• new GUI
• Community: 400 pull requests per week
• Обновили документацию
• Bugs Bounty!!! For every PR: https://github.com/kubernetes/kubernetes.github.io/issues?q=is%3Aissue+is%3Aopen+label%3ABounty:
  • $100 to every accepted diagram (for concepts)
  • $25 for each merged pull request that fully processes a page and makes sure it is linking out to helpful information.
  • $25 for anyone who submits a PR that ads {% raw %}{% endraw %} in the codeblocks
• Next version: 1.3 is scheduled to 24 Jun
• RedHat Kubernetes 1.3 features:
  • Better Seccomp support for docker/container (restrict the actions/syscalls available within the container)
  • kubelete evicts pods when low on memory (QoS tier)
  • Backpressure support in several controllers
  • k8s on docker-1.10
  • dynamic provisioning on volumes
  • PersistentVolume quota
  • PetSet (stateful services)
  • Scale out to 1000 nodes of half-max-pod-density with 250 pods per server
  • apiserver performance (adding protobuf serialization)
  • assist with disk accounting
  • cgroups QoS hierarchy
• Node SIG 1.3 feature
  • introduce out-of-disk policy
  • disk account for device mapper
  • handle “out-of-memory”
• Other SIG 1.3 feature:
  • etcd3