Merge pull request #124 from MaciekPytel/ca_0_6_faq

CA 0.6 FAQ update

cluster-autoscaler/FAQ.md

@@ -1,5 +1,12 @@
 # Frequently Asked Questions
 
+# Older versions
+The answers in this FAQ apply to the newest version of Cluster Autoscaler. If
+you're using an older version of CA please refer to corresponding version of
+this document:
+
+[Cluster Autoscaler 0.5.X](https://github.com/kubernetes/autoscaler/blob/cluster-autoscaler-release-0.5/cluster-autoscaler/FAQ.md)
+
 # Table of Contents:
 <!--- TOC BEGIN -->
 * [Basics](#basics)
@@ -11,6 +18,9 @@
   * [Should I use a CPU-usage-based node autoscaler with Kubernetes?](#should-i-use-a-cpuusagebased-node-autoscaler-with-kubernetes)
   * [How is Cluster Autoscaler different from CPU-usage-based node autoscalers?](#how-is-cluster-autoscaler-different-from-cpuusagebased-node-autoscalers)
   * [Is Cluster Autoscaler compatible with CPU-usage-based node autoscalers?](#is-cluster-autoscaler-compatible-with-cpuusagebased-node-autoscalers)
+* [How to?](#how-to)
+  * [I'm running cluster with nodes in multiple zones for HA purposes. Is that supported by Cluster Autoscaler?](#im-running-cluster-with-nodes-in-multiple-zones-for-ha-purposes-is-that-supported-by-cluster-autoscaler)
+  * [How can I monitor Cluster Autoscaler?](#how-can-i-monitor-cluster-autoscaler)
 * [Internals](#internals)
   * [Are all of the mentioned heuristics and timings final?](#are-all-of-the-mentioned-heuristics-and-timings-final)
   * [How does scale up work?](#how-does-scale-up-work)
@@ -94,6 +104,39 @@ They are also not particularly suited to use with Kubernetes in general.
 
 ****************
 
+# How to?
+
+### I'm running cluster with nodes in multiple zones for HA purposes. Is that supported by Cluster Autoscaler?
+CA 0.6 introduced `--balance-similar-node-groups` flag to support this use-case. If you set the flag to true
+CA will automatically identify node groups using the same instance types and
+having the same set of labels (except for automatically added zone labels) and try to
+keep the size of those node groups balanced.
+
+This does not guarantee similar node groups will have exactly the same sizes:
+* Currently the balancing is only done at scale-up. Cluster Autoscaler will
+  still scale-down underutilized nodes regardless of relative size of underlying
+  node groups. We plan to take balancing into account in scale-down in the future.
+* Cluster Autoscaler will only add as many nodes as required to run all existing
+  pods. If the number of nodes is not divisible by number of balanced node
+  groups some groups will get 1 more node than others.
+* Cluster Autoscaler will only balance between node groups that can support the
+  same set of pending pods. If you run pods that can only go to a single node group
+  (for example due to nodeSelector on zone label) CA will only add nodes to
+  this particular node group.
+
+You can opt-out a node group from being automatically balanced with other node
+groups using the same instance type by giving it any custom label.
+
+### How can I monitor Cluster Autoscaler?
+Cluster Autoscaler provides metrics and livenessProbe endpoints. By
+default they're available on port 8085 (configurable with `--address` flag),
+respectively under /metrics and /health-check.
+
+Metrics are provided in Prometheus format and their detailed description is
+available [here](https://github.com/kubernetes/autoscaler/blob/master/cluster-autoscaler/proposals/metrics.md).
+
+****************
+
 # Internals
 
 ### Are all of the mentioned heuristics and timings final?