Snap + Kubernetes

by andrzej-k (Intel)

Snap overview

Why:

• Make telemetry data available - to improve scheduling, allow for meaningful data analytics, flexible auto-scaling, to have insight into apps performance, ...

How:

• Fully open source project - https://github.com/intelsdi-x/snap/
• Plugin architecture for metrics collection, processing and publishing
• Tasks define what to collect, how to process and where to publish
• Tribe helps with manageability - group nodes and create agreements

​

Snap overview

Snap + Kubernetes

Why:

• custom/service metrics can be added easily (with no code changes)
• wealth of metrics that can be collected (incl. those collected by cAdvisor)
• ability to push metrics to different endpoints
• processing metrics in-place
• zero downtime metrics update

Try it:

• on Google Cloud Platform - https://github.com/intelsdi-x/kubesnap

​

Snap + Kubernetes

How:

• Snap is a daemon-set - addon in Kubernetes
• Docker collector (for “core” metrics) and Heapster publisher plugins are pre-loaded
• All snap daemons are forming one tribe
• Heapster supports snap as data source
• Other snap plugins can be used to collect custom/service metrics, process the data and push to different endpoints (for example: databases, message queues, etc.)
• Snap is transparent to Heapster clients (for example: HPA)

​

Snap + Kubernetes

Demo

Highlights:

• Snap running as daemon-set in Tribe mode
• How to load a new plugin
• What is a metrics catalog
• How snap task looks like
• How custom metrics are defined
• What happens when new version of the plugin is loaded
• How to collect app-level metrics

More info

Here:

• Snap and Kubernetes: together at last
• The benefits of using Snap with Kubernetes