Red Hat OpenShift Pipelines release notes

TP: Technology Preview

GA: General Availability

The custom tasks have the following enhancements:

The finally task has the following enhancements:

Support for multiple secrets of the type dockercfg or dockerconfigjson is added for authentication at runtime.

Functionality to support sparse-checkout with the git-clone task is added. This enables you to clone only a subset of the repository as your local copy, and helps you to restrict the size of the cloned repositories.

You can create pipeline runs in a pending state without actually starting them. In clusters that are under heavy load, this allows Operators to have control over the start time of the pipeline runs.

Ensure that you set the SYSTEM_NAMESPACE environment variable manually for the controller; this was previously set by default.

A non-root user is now added to the build-base image of pipelines so that git-init can clone repositories as a non-root user.

Support to validate dependencies between resolved resources before a pipeline run starts is added. All result variables in the pipeline must be valid, and optional workspaces from a pipeline can only be passed to tasks expecting it for the pipeline to start running.

The controller and webhook runs as a non-root group, and their superfluous capabilities have been removed to make them more secure.

You can use the tkn pr logs command to see the log streams for retried task runs.

You can use the --clustertask option in the tkn tr delete command to delete all the task runs associated with a particular cluster task.

Support for using Knative service with the EventListener resource is added by introducing a new customResource field.

An error message is displayed when an event payload does not use the JSON format.

The source control interceptors such as GitLab, BitBucket, and GitHub, now use the new InterceptorRequest or InterceptorResponse type interface.

A new CEL function marshalJSON is implemented so that you can encode a JSON object or an array to a string.

An HTTP handler for serving the CEL and the source control core interceptors is added. It packages four core interceptors into a single HTTP server that is deployed in the tekton-pipelines namespace. The EventListener object forwards events over the HTTP server to the interceptor. Each interceptor is available at a different path. For example, the CEL interceptor is available on the /cel path.

The pipelines-scc Security Context Constraint (SCC) is used with the default pipeline service account for pipelines. This new service account is similar to anyuid, but with a minor difference as defined in the YAML for SCC of OpenShift Container Platform 4.7:

fsGroup:
  type: MustRunAs

The build-gcs sub-type in the pipeline resource storage, and the gcs-fetcher image, are not supported.

In the taskRun field of cluster tasks, the label tekton.dev/task is removed.

For webhooks, the value v1beta1 corresponding to the field admissionReviewVersions is removed.

The creds-init helper image for building and deploying is removed.

In the triggers spec and binding, the deprecated field template.name is removed in favor of template.ref. You should update all eventListener definitions to use the ref field.

For EventListener custom resources/objects, the fields PodTemplate and ServiceType are deprecated in favor of Resource.

The deprecated spec style embedded bindings is removed.

The spec field is removed from the triggerSpecBinding.

The event ID representation is changed from a five-character random string to a UUID.

In the Developer perspective, the pipeline metrics and triggers features are available only on OpenShift Container Platform 4.7.6 or later versions.

On IBM Power Systems, IBM Z, and LinuxONE, the tkn hub command is not supported.

When you run Maven and Jib Maven cluster tasks on an IBM Power Systems (ppc64le), IBM Z, and LinuxONE (s390x) clusters, set the MAVEN_IMAGE parameter value to maven:3.6.3-adoptopenjdk-11.

Triggers throw error resulting from bad handling of the JSON format, if you have the following configuration in the trigger binding:

params:
  - name: github_json
    value: $(body)

When upgrading from Pipelines 1.3.x to 1.4.x, you must recreate the routes.

Previously, the tekton.dev/task label was removed from the task runs of cluster tasks, and the tekton.dev/clusterTask label was introduced. The problems resulting from that change is resolved by fixing the clustertask describe and delete commands. In addition, the lastrun function for tasks is modified, to fix the issue of the tekton.dev/task label being applied to the task runs of both tasks and cluster tasks in older versions of pipelines.

When doing an interactive tkn pipeline start pipelinename, a PipelineResource is created interactively. The tkn p start command prints the resource status if the resource status is not nil.

Previously, the tekton.dev/task=name label was removed from the task runs created from cluster tasks. This fix modifies the tkn clustertask start command with the --last flag to check for the tekton.dev/task=name label in the created task runs.

When a task uses an inline task specification, the corresponding task run now gets embedded in the pipeline when you run the tkn pipeline describe command, and the task name is returned as embedded.

The tkn version command is fixed to display the version of the installed Tekton CLI tool, without a configured kubeConfiguration namespace or access to a cluster.

If an argument is unexpected or more than one arguments are used, the tkn completion command gives an error.

Previously, pipeline runs with the finally tasks nested in a pipeline specification would lose those finally tasks, when converted to the v1alpha1 version and restored back to the v1beta1 version. This error occurring during conversion is fixed to avoid potential data loss. Pipeline runs with the finally tasks nested in a pipeline specification is now serialized and stored on the alpha version, only to be deserialized later.

Previously, there was an error in the pod generation when a service account had the secrets field as {}. The task runs failed with CouldntGetTask because the GET request with an empty secret name returned an error, indicating that the resource name may not be empty. This issue is fixed by avoiding an empty secret name in the kubeclient GET request.

Pipelines with the v1beta1 API versions can now be requested along with the v1alpha1 version, without losing the finally tasks. Applying the returned v1alpha1 version will store the resource as v1beta1, with the finally section restored to its original state.

Previously, an unset selfLink field in the controller caused an error in the Kubernetes v1.20 clusters. As a temporary fix, the CloudEvent source field is set to a value that matches the current source URI, without the value of the auto-populated selfLink field.

Previously, a secret name with dots such as gcr.io led to a task run creation failure. This happened because of the secret name being used internally as part of a volume mount name. The volume mount name conforms to the RFC1123 DNS label and disallows dots as part of the name. This issue is fixed by replacing the dot with a dash that results in a readable name.

Context variables are now validated in the finally tasks.

Previously, when the task run reconciler was passed a task run that did not have a previous status update containing the name of the pod it created, the task run reconciler listed the pods associated with the task run. The task run reconciler used the labels of the task run, which were propagated to the pod, to find the pod. Changing these labels while the task run was running, caused the code to not find the existing pod. As a result, duplicate pods were created. This issue is fixed by changing the task run reconciler to only use the tekton.dev/taskRun Tekton-controlled label when finding the pod.

Previously, when a pipeline accepted an optional workspace and passed it to a pipeline task, the pipeline run reconciler stopped with an error if the workspace was not provided, even if a missing workspace binding is a valid state for an optional workspace. This issue is fixed by ensuring that the pipeline run reconciler does not fail to create a task run, even if an optional workspace is not provided.

The sorted order of step statuses matches the order of step containers.

Previously, the task run status was set to unknown when a pod encountered the CreateContainerConfigError reason, which meant that the task and the pipeline ran until the pod timed out. This issue is fixed by setting the task run status to false, so that the task is set as failed when the pod encounters the CreateContainerConfigError reason.

Previously, pipeline results were resolved on the first reconciliation, after a pipeline run was completed. This could fail the resolution resulting in the Succeeded condition of the pipeline run being overwritten. As a result, the final status information was lost, potentially confusing any services watching the pipeline run conditions. This issue is fixed by moving the resolution of pipeline results to the end of a reconciliation, when the pipeline run is put into a Succeeded or True condition.

Execution status variable is now validated. This avoids validating task results while validating context variables to access execution status.

Previously, a pipeline result that contained an invalid variable would be added to the pipeline run with the literal expression of the variable intact. Therefore, it was difficult to assess whether the results were populated correctly. This issue is fixed by filtering out the pipeline run results that reference failed task runs. Now, a pipeline result that contains an invalid variable will not be emitted by the pipeline run at all.

The tkn eventlistener describe command is fixed to avoid crashing without a template. It also displays the details about trigger references.

Upgrades from Pipelines 1.3.x and earlier versions to Pipelines 1.4.0 breaks event listeners because of the unavailability of template.name. In Pipelines 1.4.1, the template.name has been restored to avoid breaking event listeners in triggers.

In Pipelines 1.4.1, the ConsoleQuickStart custom resource has been updated to align with OpenShift Container Platform 4.7 capabilities and behavior.

Tekton Pipelines 0.19.0

Tekton tkn CLI 0.15.0

Tekton Triggers 0.10.2

cluster tasks based on Tekton Catalog 0.19.0

IBM Power Systems on OpenShift Container Platform 4.7

IBM Z and LinuxONE on OpenShift Container Platform 4.7

Pipeline templates that use PipelineResources CRDs are now deprecated and will be removed in a future release.

The template.name field is deprecated in favor of the template.ref field and will be removed in a future release.

The -c shorthand for the --check command has been removed. In addition, global tkn flags are added to the version command.

CEL overlays add fields to a new top-level extensions function, instead of modifying the incoming event body. TriggerBinding resources can access values within this new extensions function using the $(extensions.<key>) syntax. Update your binding to use the $(extensions.<key>) syntax instead of the $(body.<overlay-key>) syntax.

The escaping parameters behavior by replacing " with \" is now removed. If you need to retain the old escaping parameters behavior add the tekton.dev/old-escape-quotes: true" annotation to your TriggerTemplate specification.

You can embed TriggerBinding resources by using the name and value fields inside a trigger or an event listener. However, you cannot specify both name and ref fields for a single binding. Use the ref field to refer to a TriggerBinding resource and the name field for embedded bindings.

An interceptor cannot attempt to reference a secret outside the namespace of an EventListener resource. You must include secrets in the namespace of the `EventListener`resource.

In Triggers 0.9.0 and later, if a body or header based TriggerBinding parameter is missing or malformed in an event payload, the default values are used instead of displaying an error.

Tasks and pipelines created with WhenExpression objects using Tekton Pipelines 0.16.x must be reapplied to fix their JSON annotations.

When a pipeline accepts an optional workspace and gives it to a task, the pipeline run stalls if the workspace is not provided.

To use the Buildah cluster task in a disconnected environment, ensure that the Dockerfile uses an internal image stream as the base image, and then use it in the same manner as any s2i cluster task.

Extensions added by a CEL Interceptor are passed on to webhook interceptors by adding the Extensions field within the event body.

The activity timeout for log readers is now configurable using the LogOptions field. However, the default behavior of timeout in 10 seconds is retained.

The log command ignores the --follow flag when a task run or pipeline run is complete, and reads available logs instead of live logs.

References to the following Tekton resources: EventListener, TriggerBinding, ClusterTriggerBinding, Condition, and TriggerTemplate are now standardized and made consistent across all user-facing messages in tkn commands.

Previously, if you started a canceled task run or pipeline run with the --use-taskrun <canceled-task-run-name>, --use-pipelinerun <canceled-pipeline-run-name> or --last flags, the new run would be canceled. This bug is now fixed.

The tkn pr desc command is now enhanced to ensure that it does not fail in case of pipeline runs with conditions.

When you delete a task run using the tkn tr delete command with the --task option, and a cluster task exists with the same name, the task runs for the cluster task also get deleted. As a workaround, filter the task runs by using the TaskRefKind field.

The tkn triggertemplate describe command would display only part of the apiVersion value in the output. For example, only triggers.tekton.dev was displayed instead of triggers.tekton.dev/v1alpha1. This bug is now fixed.

The webhook, under certain conditions, would fail to acquire a lease and not function correctly. This bug is now fixed.

Pipelines with when expressions created in v0.16.3 can now be run in v0.17.1 and later. After an upgrade, you do not need to reapply pipeline definitions created in previous versions because both the uppercase and lowercase first letters for the annotations are now supported.

By default, the leader-election-ha field is now enabled for high availability. When the disable-ha controller flag is set to true, it disables high availability support.

Issues with duplicate cloud events are now fixed. Cloud events are now sent only when a condition changes the state, reason, or message.

When a service account name is missing from a PipelineRun or TaskRun spec, the controller uses the service account name from the config-defaults config map. If the service account name is also missing in the config-defaults config map, the controller now sets it to default in the spec.

Validation for compatibility with the affinity assistant is now supported when the same persistent volume claim is used for multiple workspaces, but with different subpaths.

Tekton Pipelines 0.16.3

Tekton tkn CLI 0.13.1

Tekton Triggers 0.8.1

cluster tasks based on Tekton Catalog 0.16

IBM Power Systems on OpenShift Container Platform 4.6

IBM Z and LinuxONE on OpenShift Container Platform 4.6

$(params) parameters are now removed from the triggertemplate resource and replaced by $(tt.params) to avoid confusion between the resourcetemplate and triggertemplate resource parameters.

The ServiceAccount reference of the optional EventListenerTrigger-based authentication level has changed from an object reference to a ServiceAccountName string. This ensures that the ServiceAccount reference is in the same namespace as the EventListenerTrigger object.

The Conditions custom resource definition (CRD) is now deprecated; use the WhenExpressions CRD instead.

The PipelineRun.Spec.ServiceAccountNames object is being deprecated and replaced by the PipelineRun.Spec.TaskRunSpec[].ServiceAccountName object.

This release of Red Hat OpenShift Pipelines adds support for a disconnected installation. However, some images used by the cluster tasks must be mirrored for them to work in disconnected clusters.

Pipelines in the openshift namespace are not deleted after you uninstall the Red Hat OpenShift Pipelines Operator. Use the oc delete pipelines -n openshift --all command to delete the pipelines.

Uninstalling the Red Hat OpenShift Pipelines Operator does not remove the event listeners.

When you run a multi-arch container image task without command specification on an IBM Power Systems (ppc64le) or IBM Z (s390x) cluster, the TaskRun resource fails with the following error:

Error executing command: fork/exec /bin/bash: exec format error

$ skopeo inspect --raw <image_name>| jq '.manifests[] | select(.platform.architecture == "<architecture>") | .digest'

A simple syntax validation to check the CEL filter, overlays in the Webhook validator, and the expressions in the interceptor has now been added.

Triggers no longer overwrite annotations set on the underlying deployment and service objects.

Previously, an event listener would stop accepting events. This fix adds an idle timeout of 120 seconds for the EventListener sink to resolve this issue.

Previously, canceling a pipeline run with a Failed(Canceled) state gave a success message. This has been fixed to display an error instead.

The tkn eventlistener list command now provides the status of the listed event listeners, thus enabling you to easily identify the available ones.

Consistent error messages are now displayed for the triggers list and triggers describe commands when triggers are not installed or when a resource cannot be found.

Previously, a large number of idle connections would build up during cloud event delivery. The DisableKeepAlives: true parameter was added to the cloudeventclient config to fix this issue. Thus, a new connection is set up for every cloud event.

Previously, the creds-init code would write empty files to the disk even if credentials of a given type were not provided. This fix modifies the creds-init code to write files for only those credentials that have actually been mounted from correctly annotated secrets.

Tekton Pipelines 0.14.3

Tekton tkn CLI 0.11.0

Tekton Triggers 0.6.1

cluster tasks based on Tekton Catalog 0.14

The --namespace or -n flags for all cluster-wide commands, including the clustertask and clustertriggerbinding commands, are deprecated. It will be removed in a future release.

The name field in triggers.bindings within an event listener has been deprecated in favor of the ref field and will be removed in a future release.

Variable interpolation in trigger templates using $(params) has been deprecated in favor of using $(tt.params) to reduce confusion with the pipeline variable interpolation syntax. The $(params.<paramName>) syntax will be removed in a future release.

The tekton.dev/task label is deprecated on cluster tasks.

The TaskRun.Status.ResourceResults.ResourceRef field is deprecated and will be removed.

The tkn pipeline create, tkn task create, and tkn resource create -f subcommands have been removed.

Namespace validation has been removed from tkn commands.

The default timeout of 1h and the -t flag for the tkn ct start command have been removed.

The s2i cluster task has been deprecated.

Conditions do not support workspaces.

The --workspace option and the interactive mode is not supported for the tkn clustertask start command.

Support of backward compatibility for $(params.<paramName>) syntax forces you to use trigger templates with pipeline specific params as the trigger s webhook is unable to differentiate trigger params from pipelines params.

Pipeline metrics report incorrect values when you run a promQL query for tekton_taskrun_count and tekton_taskrun_duration_seconds_count.

pipeline runs and task runs continue to be in the Running and Running(Pending) states respectively even when a non existing PVC name is given to a workspace.

Previously, the tkn task delete <name> --trs command would delete both the task and cluster task if the name of the task and cluster task were the same. With this fix, the command deletes only the task runs that are created by the task <name>.

Previously the tkn pr delete -p <name> --keep 2 command would disregard the -p flag when used with the --keep flag and would delete all the pipeline runs except the latest two. With this fix, the command deletes only the pipeline runs that are created by the pipeline <name>, except for the latest two.

The tkn triggertemplate describe output now displays resource templates in a table format instead of YAML format.

Previously the buildah cluster task failed when a new user was added to a container. With this fix, the issue has been resolved.

Tekton Pipelines 0.11.3

Tekton tkn CLI 0.9.0

Tekton Triggers 0.4.0

cluster tasks based on Tekton Catalog 0.11

The environment variable $HOME, and variable workingDir in the Steps specification are deprecated and might be changed in a future release. Currently in a Step container, the HOME and workingDir variables are overwritten to /tekton/home and /workspace variables, respectively.

The following commands are deprecated and might be removed in the future release: tkn pipeline create, tkn task create.

The -f flag with the tkn resource create command is now deprecated. It might be removed in the future release.

The -t flag and the --timeout flag (with seconds format) for the tkn clustertask create command are now deprecated. Only duration timeout format is now supported, for example 1h30s. These deprecated flags might be removed in the future release.

If you are upgrading from an older version of Red Hat OpenShift Pipelines, you must delete your existing deployments before upgrading to Red Hat OpenShift Pipelines version 1.0. To delete an existing deployment, you must first delete Custom Resources and then uninstall the Red Hat OpenShift Pipelines Operator. For more details, see the uninstalling Red Hat OpenShift Pipelines section.

Submitting the same v1alpha1 tasks more than once results in an error. Use the oc replace command instead of oc apply when re-submitting a v1alpha1 task.

The buildah cluster task does not work when a new user is added to a container.

useradd: /etc/passwd.8: lock file already used
useradd: cannot lock /etc/passwd; try again later.

Previously, the DeploymentConfig task triggered a new deployment build even when an image build was already in progress. This caused the deployment of the pipeline to fail. With this fix, the deploy task command is now replaced with the oc rollout status command which waits for the in-progress deployment to finish.

Support for APP_NAME parameter is now added in pipeline templates.

Previously, the pipeline template for Java s2i failed to look up the image in the registry. With this fix, the image is looked up using the existing image pipeline resources instead of the user provided IMAGE_NAME parameter.

All the OpenShift Pipelines images are now based on the Red Hat Universal Base Images (UBI).

Previously, when the pipeline was installed in a namespace other than tekton-pipelines, the tkn version command displayed the pipeline version as unknown. With this fix, the tkn version command now displays the correct pipeline version in any namespace.

The -c flag is no longer supported for the tkn version command.

Non-admin users can now list the cluster trigger bindings.

The event listener CompareSecret function is now fixed for the CEL Interceptor.

The list, describe, and start subcommands for tasks and cluster tasks now correctly display the output in case a task and cluster task have the same name.

Previously, the OpenShift Pipelines Operator modified the privileged security context constraints (SCCs), which caused an error during cluster upgrade. This error is now fixed.

In the tekton-pipelines namespace, the timeouts of all task runs and pipeline runs are now set to the value of default-timeout-minutes field using the config map.

Previously, the pipelines section in the web console was not displayed for non-admin users. This issue is now resolved.