Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Expose the option for Kuadrant to be installed in HA mode #798

Open
7 tasks
maleck13 opened this issue Aug 6, 2024 · 4 comments
Open
7 tasks

Expose the option for Kuadrant to be installed in HA mode #798

maleck13 opened this issue Aug 6, 2024 · 4 comments

Comments

@maleck13
Copy link
Collaborator

maleck13 commented Aug 6, 2024

What

When you install Kuadrant, it defaults to a single instance of Authorino and Limitador. In order to be resilient to failure, some installations may want multiple instances of Authorino and Limitador deployed as these components are in the critical path for requests. As Authorino and Limitador support having multiple instances deployed on the same cluster, we should expose options for this deployment topology to be used via Kuadrant

At a high level the key things to expose would be the number of replicas of each that you want and how you want them distributed.
Kubernetes supports distributing instances of a pod via topology constraints
https://kubernetes.io/docs/concepts/scheduling-eviction/topology-spread-constraints/#topology-spread-constraint-examples

use cases:

Support a multi-az k8s cluster where I want to have a gateway instance per AZ that routes to an Authorino and Limitador in the same AZ I want to spread out instances of Authorino and Limitador across AZs rather than allowing them to be potentially scheduled to the same AZ or even the same node. Allow me to have more than one instance of these components per AZ / per cluster for redundancy and to improve resiliency against node failure and AZ failure.

One concept for Kuadrant CRD:

deployment:
   limitador | ratelimiting:
     replicas: 3
     topologyKey: zone
   authorino | auth:
     replicas: 3
     topologyKey: zone

We may also want to consider a simpler level of configuration:

deployment:
   mode: HA
   topologyKey: zone | node

In Authorino and Limitador we would need to add (just an example):

topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: zone
    whenUnsatisfiable: ScheduleAnyway
    labelSelector:
      matchLabels:
        app: limitador

Done

  • Decide how to expose this deployment topology via the Kuadrant CR
  • Understand any required changes to the Authorino and Limitador operators
  • Implement any needed changes to the Authorino and Limitador operators and CRDs
  • Implement the chosen Kuadrant CR and Kuadrant Operator changes
  • Add e2e test that deploys Kuadrant in this shape and tests it is as expected
  • Document the options for this deployment type in the Kuadrant CR
  • Document our recommended topologies in a multi-az and single az setup
@maleck13
Copy link
Collaborator Author

maleck13 commented Aug 6, 2024

@didierofrivia @guicassolato @alexsnaps FYI interested in thoughts / counter approaches

@maleck13
Copy link
Collaborator Author

maleck13 commented Aug 6, 2024

Open Question:

With this configuration, would we also force a redisConfig to be set?

@alexsnaps
Copy link
Member

I'm a bit confused by the line of thinking here... Does that mean a "HA deployment" of the gateways as well? Limitador would need to share counters as well, one way would be a shared redis... but how is that HA then? Or do we consider that "out of scope"? Or would disk persistence be enough? If so how "HA is HA"? The CRDT in memory based counters could be used to share across on the replicas... but that's out of scope for this release.

@Boomatang
Copy link
Contributor

Is the "simpler level of configuration" the correct why to approach this, as it would more inline with what you suggested with in the RFC: Observability API PR and we are wanting to create a unified API.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Status: No status
Development

No branches or pull requests

3 participants