VM based deployments to Kubernetes
Starting your journey with Captain Kube
Bhavin Gandhi
The beginning
Image Credits: phippy.io (CC-BY 4.0)
- Application with minimum business logic
- Less number of users
- Small team of engineers
Infrastructure
- Deployed as few Virtual Machines (VMs)
The growth
- More users
- Demand for new features
- More number of engineers
- Need of moving faster with new features
What about the infrastructure?
Should the way we deploy things change?
Continuous Delivery with VMs
- Building VM images for application updates
- Using tools like Ansible, Puppet
Shortcomings
Image Credits: phippy.io (CC-BY 4.0)
Time consuming
- VMs take time to boot
- Building VM images
- OS updates
Auto scaling
- No metrics about actual consumption of application
- Scaling based on VMs memory/CPU usage
Introducing Kubernetes
“Kubernetes” is the Greek word for a ship’s captain.
The Kubernetes project focuses on building a robust platform for running thousands of containers in production.
What Kubernetes offers
- Networking, scheduling, load balancing, autoscaling
- Declarative approach to deployments
- Abstraction over underlying infrastructure
- And many more.
How Kubernetes helps
- No need to manage individual VMs
- Common/Standard industry patterns for deploying things
- No vendor lock in
Migrating to Kubernetes
Containerizing applications
Image Credits: phippy.io (CC-BY 4.0)
Containers?
Isolated processes running with their dependencies
Best practices
- One process per container
- Try to keep the images lightweight
- No direct lift and shift from VM images (Packer)
- Logging to stdout and stderr
Ready to sail with Captain Kube?
Image Credits: phippy.io (CC-BY 4.0)
Challenges running in Kubernetes
- Resource limits
- Interprocess communication
Don't miss this
- Pod IPs are ephemeral
- Health check end points for applications
- Resource requests and limits for each application
- Use community Helm charts and operators
- Namespaces per application or teams
Migration strategies
- Lift and shift all the applications
- Migrate gradually
Migrating gradually
Start with stateless applications followed by stateful applications and then move databases
- One application at a time
- One API end point at a time
Should I run my monoliths?
YES! YES! YES!
Make sure you containerize them correctly
Cluster management
Deploying the cluster
- Using managed cluster solutions
- GKE, EKS, AKS, …
- Self managed clusters
- Rancher, kops, Kubespray, …
Backup and restore
Use tools like Velero
Cluster updates
One minor release every 3 months
The way ahead
- Checkout tools from CNCF landscape
- Prometheus for metrics collection
- Argo for CI/CD
- nginx-ingress to expose services publicly
Questions
bhavin192[at]geeksocket.in
References
These slides are released publicly under
Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)