Skip to content

Getting Started with kops on Spot Ocean

Ocean by Spot simplifies infrastructure management for Kubernetes. With robust, container-driven infrastructure auto-scaling and intelligent right-sizing for container resource requirements, operations can literally "set and forget" the underlying cluster.

Ocean seamlessly integrates with your existing instance groups, as a drop-in replacement for AWS Auto Scaling groups, and allows you to streamline and optimize the entire workflow, from initially creating your cluster to managing and optimizing it on an ongoing basis.

Features

  • Simplify Cluster Management — Ocean's Virtual Node Groups make it easy to run different infrastructure in a single cluster, which can span multiple AWS VPC availability zones and subnets for high-availability.

  • Container-Driven Autoscaling and Vertical Rightsizing — Auto-detect your container infrastructure requirements so the appropriate instance size or type will always be available. Measure real-time CPU/Memory consumption of your Pods for ongoing resource optimization.

  • Cloud-Native Showback — Gain a granular view of your cluster's cost breakdown (compute and storage) for each and every one of the cluster's resources such as Namespaces, Deployments, Daemon Sets, Jobs, and Pods.

  • Optimized Pricing and Utilization — Ocean not only intelligently leverages Spot Instances and reserved capacity to reduce costs, but also eliminates underutilized instances with container-driven autoscaling and advanced bin-packing.

Prerequisites

Make sure you have installed kops and installed kubectl.

Setup your environment

Spot

Generate your credentials here. If you are not a Spot Ocean user, sign up for free here. For further information, please checkout our Spot API guide, available on the Spot Help Center website.

To use environment variables, run:

export SPOTINST_TOKEN=<spotinst_token>
export SPOTINST_ACCOUNT=<spotinst_account>

To use credentials file, run the spotctl configure command:

spotctl configure
? Enter your access token [? for help] **********************************
? Select your default account  [Use arrows to move, ? for more help]
> act-01234567 (prod)
  act-0abcdefg (dev)

Or, manually create an INI formatted file like this:

[default]
token   = <spotinst_token>
account = <spotinst_account>

and place it in:

  • Unix/Linux/macOS:
    ~/.spotinst/credentials
    
  • Windows:
    %UserProfile%\.spotinst\credentials
    

AWS

Make sure to set up a dedicated IAM user, DNS records and cluster state storage. Please refer to setup your environment for further details.

Feature Flags

Flag Description
+Spotinst Enables the use of the Spot integration.
+SpotinstOcean Enables the use of the Spot Ocean integration.
+SpotinstHybrid Toggles between hybrid and full instance group implementations. Allows you to gradually integrate with Spot Ocean by continuing to use instance groups through AWS Auto Scaling groups, except for specific instance groups labeled with a predefined metadata label.
-SpotinstController Toggles the installation of the Spot controller addon off. Please note that the feature flag must be prefixed with a minus (-) sign to set its value to false, which results in disabling the controller.

Creating a Cluster

You can add an Ocean instance group to new or existing clusters. To create a new cluster with a Ocean instance groups, run:

# configure the feature flags
export KOPS_FEATURE_FLAGS="Spotinst,SpotinstOcean"

# create the cluster
kops create cluster --zones=us-west-2a example

Note

It's possible to have a cluster with both Ocean-managed and unmanaged instance groups.

# configure the feature flags
export KOPS_FEATURE_FLAGS="Spotinst,SpotinstOcean,SpotinstHybrid"

# create the instance groups
kops create -f instancegroups.yaml
# instancegroups.yaml
# A cluster with both Ocean-managed and unmanaged instance groups.
---
# Use Ocean in hybrid mode.
apiVersion: kops.k8s.io/v1alpha2
kind: InstanceGroup
metadata:
  labels:
    kops.k8s.io/cluster: "example"
    spotinst.io/hybrid: "true"
  ...
---
# Use AWS Auto Scaling group.
apiVersion: kops.k8s.io/v1alpha2
kind: InstanceGroup
metadata:
  labels:
    kops.k8s.io/cluster: "example"
  ...

Creating an Instance Group

To create a new instance group, run:

# configure the feature flags
export KOPS_FEATURE_FLAGS="Spotinst,SpotinstOcean"

# create the instance group
kops create instancegroup --role=node --name=example

To create a new instance group and have more control over the configuration, a config file can be used.

# instancegroup.yaml
# An instance group with Ocean configuration.
---
apiVersion: kops.k8s.io/v1alpha2
kind: InstanceGroup
metadata:
  labels:
    kops.k8s.io/cluster: "example"
    spotinst.io/spot-percentage: "90"
  ...

Metadata Labels

Label Description Default
spotinst.io/spot-percentage Specify the percentage of Spot instances that should spin up from the target capacity. 100
spotinst.io/utilize-reserved-instances Specify whether reserved instances should be utilized. true
spotinst.io/fallback-to-ondemand Specify whether fallback to on-demand instances should be enabled. true
spotinst.io/draining-timeout Specify a period of time, in seconds, after a node is marked for termination during which on running pods remains active. none
spotinst.io/grace-period Specify a period of time, in seconds, that Ocean should wait before applying instance health checks. none
spotinst.io/ocean-default-launchspec Specify whether to use the InstanceGroup's spec as the default Launch Spec for the Ocean cluster. none
spotinst.io/ocean-instance-types-whitelist Specify whether to whitelist specific instance types. none
spotinst.io/ocean-instance-types-blacklist Specify whether to blacklist specific instance types. none
spotinst.io/ocean-instance-types Specify a list of instance types that should be used by the Ocean Launch Spec. none
spotinst.io/autoscaler-disabled Specify whether the auto scaler should be disabled. false
spotinst.io/autoscaler-default-node-labels Specify whether default node labels should be set for the auto scaler. false
spotinst.io/autoscaler-headroom-cpu-per-unit Specify the number of CPUs to allocate for headroom. CPUs are denoted in millicores, where 1000 millicores = 1 vCPU. none
spotinst.io/autoscaler-headroom-gpu-per-unit Specify the number of GPUs to allocate for headroom. none
spotinst.io/autoscaler-headroom-mem-per-unit Specify the amount of memory (MB) to allocate for headroom. none
spotinst.io/autoscaler-headroom-num-of-units Specify the number of units to retain as headroom, where each unit has the defined CPU and memory. none
spotinst.io/autoscaler-cooldown Specify a period of time, in seconds, that Ocean should wait between scaling actions. 300
spotinst.io/autoscaler-scale-down-max-percentage Specify the maximum scale down percentage. none
spotinst.io/autoscaler-scale-down-evaluation-periods Specify the number of evaluation periods that should accumulate before a scale down action takes place. 5

Documentation

If you're new to Spot and want to get started, please checkout our Getting Started guide, available on the Spot Help Center website.

Getting Help

Please use these community resources for getting help: