You're monitoring Kubernetes Wrong
•Download as PPTX, PDF•
1 like•986 views
This document discusses monitoring Kubernetes clusters and provides best practices. It notes that monitoring has become more complex with distributed architectures like microservices and containers. Key challenges discussed are getting monitoring data from containers, making sense of large and varied data, troubleshooting distributed systems, and ensuring monitoring works for people as systems become more distributed. The document demonstrates how Sysdig provides container-level monitoring data and tools for segmenting and troubleshooting that data. It also introduces Sysdig Teams for access control and customization based on organizational teams.
Recommended
More Related Content
What's hot
What's hot (20)
Viewers also liked
Viewers also liked (15)
Similar to You're monitoring Kubernetes Wrong
Similar to You're monitoring Kubernetes Wrong (20)
More from Sysdig
Recently uploaded
Recently uploaded (20)
You're monitoring Kubernetes Wrong
- 1. You're Monitoring Kubernetes Wrong Loris Degioanni, CEO, Sysdig
- 2. Data Application Health Checks System/ Process Infrastructure Custom Metrics JVM Data Big Honkin’ Data Engine Unicorns and Rainbows Dashboards
- 4. Loris Degioanni Founder and CEO of sysdig Past: WinPcap, Wireshark http://www.sysdig.org @lorisdegio Me
- 5. Today • Monitoring Kubernetes • Few best practices • Some ideas • Some proposals
- 6. Using • Sysdig • https://github.com/draios/sysdig • Sysdig Cloud • https://sysdig.com/
- 7. In a Vendor Neutral Way
- 8. We Are in the Middle of a Big Change PCs/Servers Virtual Machines Containers
- 9. We Are in the Middle of a Big Change PCs/Servers Virtual Machines Containers Unit: machine Orchestration: none Architecture: monolithic Unit: machine Orchestration: external Architecture: monolithic Unit: app/service Orchestration: native Architecture: service-based
- 10. Things Are More Complex Now Cache Webserver Database
- 11. Computing Node Computing Node Computing Node Service1 Service2 Service3 Computing Node Computing Node Computing Node Things Are More Complex Now
- 12. Computing Node Computing Node Computing Node Computing Node Computing Node Computing Node Service1 Service2 Service3 Things Are More Complex
- 13. This Time Is More Than Servers and Processes
- 14. How Does This Impact Monitoring?
- 15. 4 Things That Are Harder Now 1. Getting the data 2. Making sense of the data 3. Troubleshooting 4. People
- 16. Getting The Data • What makes containers great… • Simple • Small • Isolated • Few dependencies • … Also makes them more opaque
- 17. Getting The Data: Pods OS Container1 Docker Container2 Docker
- 18. PodPod Getting The Data: Pods OS Container1 Docker Container2 Docker
- 19. PodPod Getting The Data: Pods OS Container1 Docker Monitoring Docker Container2 Docker Monitoring Docker
- 20. Getting The Data: Sysdig OS Container1 Docker Container2 Docker Container3 DockerAppApp
- 21. Getting The Data: Sysdig OS Container1 Docker Container2 Docker Container3 DockerAppApp kernel module
- 22. Getting The Data: Sysdig OS Container1 Docker Container2 Docker Container3 DockerAppApp sysdig Docker Capture and analysis
- 23. Getting The Data: Sysdig OS Container1 Docker Container2 Docker Container3 DockerAppApp sysdig Docker Capture and analysis
- 24. Getting/Making Sense of the Data: Guidelines • You should not be involved in monitoring instrumentation • You should not be involved in producing anything that is not a custom metric • Every metric should be tagged • Tagging must be integrated with Kubernetes • You should not be involved in tagging metrics • You should collect everything, with no filters
- 25. Demo – monitoring and segmenting data with sysdig monitoring
- 27. Demo – troubleshooting with open source sysdig
- 28. People • Microservices are about people • Monitoring microservices should be about people too
- 31. Sysdig Teams: Service-based access control • Designed to reflect teams of people • Integrated with Kubernetes • Support isolation and customization • Self service • Centralized
- 32. Use Cases • “dev vs prod” split • Team per Microservice • PaaS • Restricted environments
- 33. Links • Sysdig • https://github.com/draios/sysdig • Sysdig Cloud • https://sysdig.com/ • Teams • https://sysdig.com/blog/introducing-sysdig-teams/
Editor's Notes
- Kubernetes monitoring is the focus of this presentation Sysdig has worked with hundreds of customers deploying and monitoring kubernetes Here’s a little bit of what we’ve learned
- Monitoring is easy right? Just grab a bunch of data, put it in the washing machine, and happiness unicorns and rainbows emerge.
- But in reality it looks a little more like this
- Hi I’m loris, creator and founder of sysdig Started as an opensource tool (http://www.sysdig.org) Formerly one of the co-creators of wireshark
- Sure I love Sysdig, but much of what I’ll show you here is applicable to monitoring any kubernetes environment.
- We’ve seen series of generational changes to deployment architectures
- In each generation, we’ve seen more complexity in application architecture in order to support a faster development process, that can deliver more functionality faster to the end user.
- Here’s my simple example, your classic three tier application
- In the microservice world you’ve got: Multiple containers per service Distributed physically across many nodes And the nodes might not be physically located in the same place Driven by an orchestration system that can be making changes at any point in time
- Networking and communications in this world is an order of magnitude more complex Naturally that makes the data even more essential to understanding the system
- ….and it’s not just tech, it’s also the people behind it You’re actually changing the way dev teams are working They are mapping organizations to their services Smaller teams dedicated to individual services And different expectations about how their tools will support them
- Now Let’s talk about the impact of docker containers, microservices and kubernetes on monitoring
- Pods are how kubernetes describes containers that require a shared fate
- Pods are used to create a “sidecar” for monitoring agents. This is useful, but still complex: Networking requirements Resource utilization (agent / pod) More uncertainty for application performance
- The other method uses transparent instrumentation at the kernel/host layer
- The pros here are greater simplicity around instrumentation, networking, and resource utilization
- The con here is that, in order for this to work, the agent must be deployed in priviledged mode.
- If there is one thing you remember from this presentation, it should be this slide. This is the way we believe monitoring data should be collected, filtered, and managed in a kubernetes world.
- Let’s go deeper into troubleshooting The complexity of orchestrated environments means that troubleshooting is even hard than before. In addition, the slimmed nature of containers means you won’t have most of your tools like tcpdump or htop. Even if you did, they wouldn’t have the context of your environment, containers, and orchestrators to solve your issues. That’s where sysdig’s open source CLI comes into play.
- This isn’t all about technology. What excites me about the move to microservices is that it also can impact the way people work. We’ve seen that in terms of developer teams reorganizing around services. We think your monitoring should do the same.