Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
172 changes: 172 additions & 0 deletions enhancements/cluster-constraints-properties.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,172 @@
---
title: cluster constraint and property
authors:
- "@dinhxuanvu"
reviewers:
- "@kevinrizza"
- "@joelandford"
approvers:
- "@kevinrizza"
creation-date: 2021-10-08
last-updated: 2020-10-11
status: provisional
---

# cluster-constraint-and-property

## Release Signoff Checklist

- [ ] Enhancement is `implementable`
- [ ] Design details are appropriately documented from clear requirements
- [ ] Test plan is defined
- [ ] Graduation criteria for dev preview, tech preview, GA

## Summary

The dependency resolution in OLM needs to work with cluster constraints in order to effectively install operators that will work properly in the cluster at runtime. Plus, OLM needs to support cluster properties which reflects the runtime information that the cluster may provide and allows operators to depend on those properties if needed to ensure proper installation and operation.
Copy link
Member

@njhale njhale Oct 20, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

At this point, we don't know what a "cluster constraint" is. Could we define it up front?


## Motivation

### Existing dependency resolution

OLM currently only resolves constraints that are provided by the operators and the resolution outcome at runtime is solely based on the constraints being satisfied by other operators who provide the required properties. It is not possible at the moment for OLM to consider cluster information or properties during resolution or apply cluster constraints to the available operators to prevent or allow certain operators to be installed successfully.

### Cluster information

Operators can depend on certain information such as usable API(s) that are available in the cluster. It is important to surface those information so operators can determine at runtime if they will be able to install and work properly in the cluster. The cluster information/properties can be depended upon as a constraint for a particular operator. This dependency relationship will ensure only applicable operators can be installed in the cluster and prevent possible failed installation and potentially nonworking condition.

### Cluster restriction

The cluster itself can impose certain restriction to which operators can be installed in the cluster via the properties associated with the operators. For example, the cluster imposes a minumum kubernetes version all for installable operators. A constraint can be constructed for a specific cluster restriction and then will be resolved during dependency resolution to ensure only operators with corresponding property with a satisfiable value.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it is more a "wording" thing but it is not the cluster itself that "imposes a minimum kubernetes version". It is the operator author that requires it for the operator to work.
I think it is important in the context of "cluster constraints" to have a clear separation:

  • The cluster has properties
  • Operators may have constraints on these properties


### Goals

- Provide the mechanism to provide cluster properties and constraints
- Expose cluster properties and constraints to users

### Non-Goals
- Determine which cluster properties/constraints are required on the cluster
- Provide mechanism to retrieve and/or set up cluster properties and constraints automatically
- Provide mechanism to have separate namespaced and cluster-scoped properties/constraints

## Proposal

A new constraint API provides an effective away for cluster administrators to create custom resources (CR) to contain cluster information such as cluster constraints and properties.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think it is the role of cluster-administrators to create CR to surface cluster information. This can't be done consistently across clusters owned by different companies having totally different requirements, environments, operational processes, etc.
Operator authors that are unaware of these specific clusters need:

  • either a known finite set of properties they can leverage for their constraints
  • or a mechanism to run generic predefined queries to retrieve the information needed by the constraints they specify. Examples:
  • HW architecture
  • number of nodes
  • presence of an API
  • HA cluster or not


OLM resolver will retrieve the cluster constraints and properties from the CR(s) and the information can be used during resolution.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am not sure whether this is what you have in mind but I would expect to have constraints and properties captured by different CRDs


The cluster constraints and properties are presented as a virtual cluster-scoped operator that is installed in the namespace. For all incoming operators that are intended to run the namespace, they must satisfy all constraints that are associated with this cluster-scoped operator. At the same time, the installable operators can depend on the properties that are provided by this virtual operator.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am not sure of the interest of having something namespaced. Have you concrete examples in mind? It feels like it goes against "the operator descoping" idea


### User Stories

#### Cluster constraints

As a cluster admin, I would like to specify constraints that are specified to the cluster and they that will be applied to all operators. As a result, only operators that satisfy these constraints to be installable in the cluster.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As per my comment above. As a cluster-admin I would like primarily that operators get only installed on clusters where they can work on without me needing to do extra things for it. Do you have a concrete example of a scenario where a cluster-admin would want to define what I would call "reverse constraints"? Today they can leverage catalogs to make operators available or not on specific clusters.


#### Cluster properties

As a cluster admin, I would like to provide cluster properties that are relevant to cluster availability and capabilities so that operator authors can depend on those properties if needed to ensure their operators are installable and functioning properly in the cluster.

#### Bundle properties and constraints

As an operator author, I would like to include constraints that rely on cluster properties so my operators in my bundles so that they can only be installed on suitable clusters.

I also would like to provide essential properties that are used in cluster constraints in my bundle so that they can be screened at runtime if the clusters have certain requirements that may impact my operators' operation.

### Design Details

#### Cluster information API

The cluster constraint API is constructed using CustomResourceDefinition (CRD):

```yaml=
apiVersion: apiextensions.k8s.io/v1
kind: CustomResourceDefinition
metadata:
name: constraints.operators.coreos.com
spec:
group: operators.coreos.com
versions:
- name: v1
served: true
storage: true
schema:
openAPIV3Schema:
type: object
properties:
spec:
type: object
properties:
constraints:
type: string
properties:
type: string
scope: Cluster
names:
plural: constraints
singular: constraint
kind: constraint
shortNames:
- constraint
```

The CRD is cluster-scoped as the CR(s) are representing cluster information. The cluster administrators have RBAC permission to create CustomResource (CR) that will contain cluster constraints and properties. For example:

```yaml=
apiVersion: operators.coreos.com/v2
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this actually part of deppy or rukpak?

kind: Constraint
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we make this PropertySet or something? Isn't a constraint just a specifial form of property?

metadata:
name: cluster-constraints
spec:
constraints: '[{"type":"olm.constraint","value":{"evaluator":{"id":"cel"},"rule":"properties.exists(p, p.type == \"certified\")","message":"require to have certified property","action":{"id":"require"}}}]'
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why a string and not a typed array?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

+1

properties: '[{"type":"olm.k8s","value":{"version":"1.21"}},{"type":"olm.label","value":{"label":"secured"}}]'
```

The CR's spec contains two fields: `constraints` and `properties`. The `constraints` field is a list of `olm.constraint` type (See [1](https://github.com/operator-framework/enhancements/pull/91)). Each constraint is specified using the following syntax:

```json=
{
"type":"olm.constraint",
"value":{
"evaluator":{
"id":"cel"
},
"rule": "properties.exists(p, p.type == \"certified\")",
"message": "require to have certified property",
"action":{
"id":"require"
}
}
}
```

The `properties` field is a list of properties that are specific to the cluster at the runtime. Each property is specified using the following syntax:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Will be able to work with ranges? If yes, could add an example?
Also, what happens when is not possible to install because of the constrain? could we:
a) Ensure that for any predefined type we will have a good and nice msg to output in order to make it easier for the consumer to know why the operator was not installed?
b) If I properly understood, in the EP: #91 we are able to create generic rules. What message we will show in this case? WDYT about the operator author also providing the msg?


```json=
{
"type":"olm.k8sversion",
"value":{
"version": "1.20.1"
}
}
}
```

A acceptable property is required to have `type` and `value` fields. The value for `type` can be any string. The value for `value` field can be a single data type or a struct that contains nested data type.

The API will come with data verification to ensure the data provided in `constraints` and `properties` fields are syntactically acceptable.

#### Bundle property and constraint

At the moment, OLM allows the bundle to include properties via `properties.yaml` in `/metadata` directory of the bundle image. This continues to be the mechanism for operator authors to supply essential properties for their operators in the bundle.


For constraints, the `dependencies.yaml` in `/metadata` directory continues to the place for operator authors to include constraints which rely on cluster properties by using the [generic constraint API](https://github.com/operator-framework/enhancements/pull/91).

#### Cluster property/constraint observability

The new constraint API is intended for cluster administrator use. Only admins should have RBAC to create, delete and update the constraint CR(s). However, regular users should have read RBAC to access the information on the CR(s) in order to see what cluster properties that they can rely upen while building their operators.

### Risks and Mitigations

## Alternatives