Chapter 2. Understanding the Operator Lifecycle Manager (OLM)

In OpenShift Container Platform 4.3, the Operator Lifecycle Manager (OLM) helps users install, update, and manage the lifecycle of all Operators and their associated services running across their clusters. It is part of the Operator Framework, an open source toolkit designed to manage Kubernetes native applications (Operators) in an effective, automated, and scalable way.

Figure 2.1. Operator Lifecycle Manager workflow

The OLM runs by default in OpenShift Container Platform 4.3, which aids cluster administrators in installing, upgrading, and granting access to Operators running on their cluster. The OpenShift Container Platform web console provides management screens for cluster administrators to install Operators, as well as grant specific projects access to use the catalog of Operators available on the cluster.

For developers, a self-service experience allows provisioning and configuring instances of databases, monitoring, and big data services without having to be subject matter experts, because the Operator has that knowledge baked into it.

A ClusterServiceVersion (CSV) is a YAML manifest created from Operator metadata that assists the Operator Lifecycle Manager (OLM) in running the Operator in a cluster.

A CSV is the metadata that accompanies an Operator container image, used to populate user interfaces with information like its logo, description, and version. It is also a source of technical information needed to run the Operator, like the RBAC rules it requires and which Custom Resources (CRs) it manages or depends on.

A CSV is composed of:

In the Operator Lifecycle Manager (OLM) ecosystem, the following resources are used to resolve Operator installations and upgrades:

Operator metadata, defined in CSVs, can be stored in a collection called a CatalogSource. OLM uses CatalogSources, which use the Operator Registry API, to query for available Operators as well as upgrades for installed Operators.

Figure 2.2. CatalogSource overview

Within a CatalogSource, Operators are organized into packages and streams of updates called channels, which should be a familiar update pattern from OpenShift Container Platform or other software on a continuous release cycle like web browsers.

Figure 2.3. Packages and channels in a CatalogSource

A user indicates a particular package and channel in a particular CatalogSource in a Subscription, for example an etcd package and its alpha channel. If a Subscription is made to a package that has not yet been installed in the namespace, the latest Operator for that package is installed.

Each CSV has a replaces parameter that indicates which Operator it replaces. This builds a graph of CSVs that can be queried by OLM, and updates can be shared between channels. Channels can be thought of as entry points into the graph of updates:

Figure 2.4. OLM’s graph of available channel updates

For example:

packageName: example
channels:
- name: alpha
  currentCSV: example.v0.1.2
- name: beta
  currentCSV: example.v0.1.3
defaultChannel: alpha

For OLM to successfully query for updates, given a CatalogSource, package, channel, and CSV, a catalog must be able to return, unambiguously and deterministically, a single CSV that replaces the input CSV.

2.1.3.2. Skipping upgrades

OLM’s basic path for upgrades is:

• A CatalogSource is updated with one or more updates to an Operator.
• OLM traverses every version of the Operator until reaching the latest version the CatalogSource contains.

However, sometimes this is not a safe operation to perform. There will be cases where a published version of an Operator should never be installed on a cluster if it has not already, for example because a version introduces a serious vulnerability.

In those cases, OLM must consider two cluster states and provide an update graph that supports both:

• The "bad" intermediate Operator has been seen by the cluster and installed.
• The "bad" intermediate Operator has not yet been installed onto the cluster.

By shipping a new catalog and adding a skipped release, OLM is ensured that it can always get a single unique update regardless of the cluster state and whether it has seen the bad update yet.

For example:

CSV with skipped release

apiVersion: operators.coreos.com/v1alpha1
kind: ClusterServiceVersion
metadata:
  name: etcdoperator.v0.9.2
  namespace: placeholder
  annotations:
spec:
    displayName: etcd
    description: Etcd Operator
    replaces: etcdoperator.v0.9.0
    skips:
    - etcdoperator.v0.9.1

Consider the following example Old CatalogSource and New CatalogSource:

Figure 2.5. Skipping updates

This graph maintains that:

• Any Operator found in Old CatalogSource has a single replacement in New CatalogSource.
• Any Operator found in New CatalogSource has a single replacement in New CatalogSource.
• If the bad update has not yet been installed, it will never be.

olm.skipRange: <semver_range>

apiVersion: operators.coreos.com/v1alpha1
kind: ClusterServiceVersion
metadata:
    name: elasticsearch-operator.v4.1.2
    namespace: <namespace>
    annotations:
        olm.skipRange: '>=4.1.0 <4.1.2'

2.1.3.4. Z-stream support

A z-stream, or patch release, must replace all previous z-stream releases for the same minor version. OLM does not care about major, minor, or patch versions, it just needs to build the correct graph in a catalog.

In other words, OLM must be able to take a graph as in Old CatalogSource and, similar to before, generate a graph as in New CatalogSource:

Figure 2.6. Replacing several Operators

This graph maintains that:

• Any Operator found in Old CatalogSource has a single replacement in New CatalogSource.
• Any Operator found in New CatalogSource has a single replacement in New CatalogSource.
• Any z-stream release in Old CatalogSource will update to the latest z-stream release in New CatalogSource.
• Unavailable releases can be considered "virtual" graph nodes; their content does not need to exist, the registry just needs to respond as if the graph looks like this.

The Operator Lifecycle Manager is composed of two Operators: the OLM Operator and the Catalog Operator.

Each of these Operators is responsible for managing the Custom Resource Definitions (CRDs) that are the basis for the OLM framework:

Each of these Operators is also responsible for creating resources:

The Operator Lifecycle Manager (OLM) exposes certain OLM-specific resources for use by the Prometheus-based OpenShift Container Platform cluster monitoring stack.

OLM manages the dependency resolution and upgrade lifecycle of running Operators. In many ways, the problems OLM faces are similar to other operating system package managers like yum and rpm.

However, there is one constraint that similar systems do not generally have that OLM does: because Operators are always running, OLM attempts to ensure that you are never left with a set of Operators that do not work with each other.

This means that OLM must never:

OLM upgrades a Custom Resource Definition (CRD) immediately if it is owned by a singular Cluster Service Version (CSV). If a CRD is owned by multiple CSVs, then the CRD is upgraded when it has satisfied all of the following backward compatible conditions:

In the following examples, a provider is an Operator which "owns" a CRD or APIService.

Example: Deprecating dependent APIs

A and B are APIs (e.g., CRDs):

This results in:

This is a case OLM prevents with its upgrade strategy.

Example: Version deadlock

A and B are APIs:

If OLM attempts to update A without simultaneously updating B, or vice-versa, it is unable to progress to new versions of the Operators, even though a new compatible set can be found.

This is another case OLM prevents with its upgrade strategy.

An OperatorGroup is an OLM resource that provides multitenant configuration to OLM-installed Operators. An OperatorGroup selects target namespaces in which to generate required RBAC access for its member Operators.

The set of target namespaces is provided by a comma-delimited string stored in the ClusterServiceVersion’s (CSV) olm.targetNamespaces annotation. This annotation is applied to member Operator’s CSV instances and is projected into their deployments.

An Operator is considered a member of an OperatorGroup if the following conditions are true:

An InstallMode consists of an InstallModeType field and a boolean Supported field. A CSV’s spec can contain a set of InstallModes of four distinct InstallModeTypes:

You can explicitly name the target namespace for an OperatorGroup using the spec.targetNamespaces parameter:

apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: my-group
  namespace: my-namespace
spec:
  targetNamespaces:
  - my-namespace

You can alternatively specify a namespace using a label selector with the spec.selector parameter:

apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: my-group
  namespace: my-namespace
  spec:
    selector:
      cool.io/prod: "true"

If both spec.targetNamespaces and spec.selector are defined, spec.selector is ignored. Alternatively, you can omit both spec.selector and spec.targetNamespaces to specify a global OperatorGroup, which selects all namespaces:

apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: my-group
  namespace: my-namespace

The resolved set of selected namespaces is shown in an OperatorGroup’s status.namespaces parameter. A global OperatorGroup’s status.namespace contains the empty string (""), which signals to a consuming Operator that it should watch all namespaces.

Member CSVs of an OperatorGroup have the following annotations:

Information about what GroupVersionKinds (GVKs) are provided by an OperatorGroup are shown in an olm.providedAPIs annotation. The annotation’s value is a string consisting of <kind>.<version>.<group> delimited with commas. The GVKs of CRDs and APIServices provided by all active member CSVs of an OperatorGroup are included.

Review the following example of an OperatorGroup with a single active member CSV that provides the PackageManifest resource:

apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  annotations:
    olm.providedAPIs: PackageManifest.v1alpha1.packages.apps.redhat.com
  name: olm-operators
  namespace: local
  ...
spec:
  selector: {}
  serviceAccount:
    metadata:
      creationTimestamp: null
  targetNamespaces:
  - local
status:
  lastUpdated: 2019-02-19T16:18:28Z
  namespaces:
  - local

When an OperatorGroup is created, three ClusterRoles are generated. Each contains a single AggregationRule with a ClusterRoleSelector set to match a label, as shown below:

The following RBAC resources are generated when a CSV becomes an active member of an OperatorGroup, as long as the CSV is watching all namespaces with the AllNamespaces InstallMode and is not in a failed state with reason InterOperatorGroupOwnerConflict.

OLM creates copies of all active member CSVs of an OperatorGroup in each of that OperatorGroup’s target namespaces. The purpose of a copied CSV is to tell users of a target namespace that a specific Operator is configured to watch resources created there. Copied CSVs have a status reason Copied and are updated to match the status of their source CSV. The olm.targetNamespaces annotation is stripped from copied CSVs before they are created on the cluster. Omitting the target namespace selection avoids the duplication of target namespaces between tenants. Copied CSVs are deleted when their source CSV no longer exists or the OperatorGroup that their source CSV belongs to no longer targets the copied CSV’s namespace.

An OperatorGroup is static if its spec.staticProvidedAPIs field is set to true. As a result, OLM does not modify the OperatorGroup’s olm.providedAPIs annotation, which means that it can be set in advance. This is useful when a user wants to use an OperatorGroup to prevent resource contention in a set of namespaces but does not have active member CSVs that provide the APIs for those resources.

Below is an example of an OperatorGroup that protects Prometheus resources in all namespaces with the something.cool.io/cluster-monitoring: "true" annotation:

apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: cluster-monitoring
  namespace: cluster-monitoring
  annotations:
    olm.providedAPIs: Alertmanager.v1.monitoring.coreos.com,Prometheus.v1.monitoring.coreos.com,PrometheusRule.v1.monitoring.coreos.com,ServiceMonitor.v1.monitoring.coreos.com
spec:
  staticProvidedAPIs: true
  selector:
    matchLabels:
      something.cool.io/cluster-monitoring: "true"

Two OperatorGroups are said to have intersecting provided APIs if the intersection of their target namespace sets is not an empty set and the intersection of their provided API sets, defined by olm.providedAPIs annotations, is not an empty set.

A potential issue is that OperatorGroups with intersecting provided APIs can compete for the same resources in the set of intersecting namespaces.

Rules for intersection

Each time an active member CSV synchronizes, OLM queries the cluster for the set of intersecting provided APIs between the CSV’s OperatorGroup and all others. OLM then checks if that set is an empty set:

The following actions are performed each time an OperatorGroup synchronizes:

Membership