IP Addressing Range - CIDR
This page provides low-level information on Kubernetes requirements for IP addressing (CIDR-Classless Inter-Domain Routing).
NOTE
Most platforms manage IP allocation for you with a minimum of config options.
CIDR is an exponential block of static IP addresses that is used for routing.
IP Address Pool for Nodes and Pods
- A single “routable” IP for the load balancer and console/api access
- VPC “cloud” IPs for the Kubernetes control plane and worker nodes
- Private IP addresses that are internal to workers/pods (e.g. 192.168.x.x, 172.10.x.x)
These IP addresses are normally set automatically by the cloud provider.
Cloud Provider CIDR
GKE optionally allows full control over the CIDR range allocated by a GKE cluster in order to optimize and limit the IP address allocation for worker nodes and pods within Kubernetes.
Amazon EKS customers previously could only create clusters in VPCs that were addressed with RFC-1918 for private IPs. EKS now supports configuring additional VPC CIDR blocks.
By default Azure AKS clusters are configured to use Kubenet, which manages the virtual network and subnet automatically, but also allows for finer-grained IP control with Azure Container Networking Interface.
General Guidance
The following is intended to provide guidance around the configuration of IP addresses required for a typical Cortex Fabric installation. Specific situations may require adjustments to these guidelines, however this should suffice as a starting point.
Base Platform requirements
Cortex Fabric
- Number of Services: 25
- Number of Pods: 50
- Number of IPs: 50
- Number of Nodes: 3
- Public IPs: IP + DNS for Fabric install https://api.* and https://private-regsitry.*
Application requirements
In addition to the base requirements you must consider your application requirements.
The following are only rough estimates.
Assumptions
Each environment may host a few medium and large applications running on top of Cortex.
Medium Example (3 medium applications)
Agents | Skills | Services | Pods |
---|---|---|---|
2 | 6-8 | 32 | 300 (100 per app) |
Large Example (3 large applications)
Agents | Skills | Services | Pods | IPs |
---|---|---|---|---|
4 | 10 | 64 | 300 (100 per app) | 192 |
Minimums and Maximums
- Minumum Nodes: 3 (for base platform)
- Maximum Pods per Node: 100
Non-routable IP addresses by category
- Pods: /18 (16,384 addresses)
- Nodes: /22 (1024 addresses)
- Services: /20 (4096 addresses)
- Master: /28 (16 addresses)