This project is a component of the Operator Framework, an open source toolkit to manage Kubernetes native applications, called Operators, in an effective, automated, and scalable way. Read more in the introduction blog post.
OLM extends Kubernetes to provide a declarative way to install, manage, and upgrade operators and their dependencies in a cluster.
It also enforces some constraints on the components it manages in order to ensure a good user experience.
This project enables users to do the following:
- Define applications as a single Kubernetes resource that encapsulates requirements and metadata
- Install applications automatically with dependency resolution or manually with nothing but
kubectl - Upgrade applications automatically with different approval policies
This project does not:
- git
- go version v1.12+.
- docker version 17.03+.
- kubectl version v1.11.3+.
- Access to a Kubernetes v1.11.3+ cluster.
Install OLM on a Kubernetes or OpenShift cluster by following the installation guide.
For a complete end-to-end example of how OLM fits into the Operator Framework, see the Operator Framework Getting Started Guide.
Use the OpenShift admin console (compatible with upstream Kubernetes) to interact with and visualize the resources managed by OLM. Create subscriptions, approve install plans, identify Operator-managed resources, and more.
Ensure kubectl is pointing at a cluster and run:
$ make run-console-localThen visit http://localhost:9000 to view the console.
Subscription detail view:
An Operator is an application-specific controller that extends the Kubernetes API to create, configure, manage, and operate instances of complex applications on behalf of a user.
OLM requires that applications be managed by an operator, but that doesn't mean that each application must write one from scratch. Depending on the level of control required you may:
- Package up an existing set of resources for OLM with helm-app-operator-kit without writing a single line of go.
- Use the operator-sdk to quickly build an operator from scratch.
The primary vehicle for describing operator requirements with OLM is a ClusterServiceVersion. Once you have an application packaged for OLM, you can deploy it with OLM by creating its ClusterServiceVersion in a namespace with a supporting OperatorGroup.
ClusterServiceVersions can be collected into CatalogSources which will allow automated installation and dependency resolution via an InstallPlan, and can be kept up-to-date with a Subscription.
Learn more about the components used by OLM by reading about the architecture and philosophy.
OLM standardizes interactions with operators by requiring that the interface to an operator be via the Kubernetes API. Because we expect users to define the interfaces to their applications, OLM currently uses CRDs to define the Kubernetes API interactions.
Examples: EtcdCluster CRD, EtcdBackup CRD
OLM introduces the notion of “descriptors” of both spec and status fields in kubernetes API responses. Descriptors are intended to indicate various properties of a field in order to make decisions about their content. For example, this can drive connecting two operators together (e.g. connecting the connection string from a mysql instance to a consuming application) and be used to drive rich interactions in a UI.
See an example of a ClusterServiceVersion with descriptors
To minimize the effort required to run an application on kubernetes, OLM handles dependency discovery and resolution of applications running on OLM.
This is achieved through additional metadata on the application definition. Each operator must define:
- The CRDs that it is responsible for managing.
- e.g., the etcd operator manages
EtcdCluster.
- e.g., the etcd operator manages
- The CRDs that it depends on.
- e.g., the vault operator depends on
EtcdCluster, because Vault is backed by etcd.
- e.g., the vault operator depends on
Basic dependency resolution is then possible by finding, for each “required” CRD, the corresponding operator that manages it and installing it as well. Dependency resolution can be further constrained by the way a user interacts with catalogs.
Dependency resolution is driven through the (Group, Version, Kind) of CRDs. This means that no updates can occur to a given CRD (of a particular Group, Version, Kind) unless they are completely backward compatible.
There is no way to express a dependency on a particular version of an operator (e.g. etcd-operator v0.9.0) or application instance (e.g. etcd v3.2.1). This encourages application authors to depend on the interface and not the implementation.
OLM has the concept of catalogs, which are repositories of application definitions and CRDs.
Catalogs contain a set of Packages, which map “channels” to a particular application definition. Channels allow package authors to write different upgrade paths for different users (e.g. alpha vs. stable).
Example: etcd package
Users can subscribe to channels and have their operators automatically updated when new versions are released.
Here's an example of a subscription:
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
name: etcd
namespace: olm
spec:
channel: singlenamespace-alpha
name: etcd
source: operatorhubio-catalog
sourceNamespace: olmThis will keep the etcd ClusterServiceVersion up to date as new versions become available in the catalog.
Catalogs are served internally over a grpc interface to OLM from operator-registry pods. Catalog data such as bundles are documented there as well.
To explore any operator samples using the OLM, see the https://operatorhub.io/ and its resources in Community Operators.
See the proposal docs and issues for ongoing or planned work.
Operator Lifecycle Manager is under Apache 2.0 license. See the LICENSE file for details.