Add Node Affinity for TaskRuns that share PVC workspace

TaskRuns within a PipelineRun may share files using a workspace volume.
The typical case is files from a git-clone operation. Tasks in a CI-pipeline often
perform operations on the filesystem, e.g. generate files or analyze files,
so the workspace abstraction is very useful.

The Kubernetes way of using file volumes is by using [PersistentVolumeClaims](https://kubernetes.io/docs/concepts/storage/persistent-volumes/#persistentvolumeclaims).
PersistentVolumeClaims use PersistentVolumes with different [access modes](https://kubernetes.io/docs/concepts/storage/persistent-volumes/#access-modes).
The most commonly available PV access mode is ReadWriteOnce, volumes with this
access mode can only be mounted on one Node at a time.

When using parallel Tasks in a Pipeline, the pods for the TaskRuns is
scheduled to any Node, most likely not to the same Node in a cluster.
Since volumes with the commonly available ReadWriteOnce access mode cannot
be use by multiple nodes at a time, these "parallel" pods is forced to
execute sequentially, since the volume only is available on one node at a time.
This may make that your TaskRuns time out.

Clusters are often _regional_, e.g. they are deployed across 3 Availability
Zones, but Persistent Volumes are often _zonal_, e.g. they are only available
for the Nodes within a single zone. Some cloud providers offer regional PVs,
but sometimes regional PVs is only replicated to one additional zone, e.g. not
all 3 zones within a region. This works fine for most typical stateful application,
but Tekton uses storage in a different way - it is designed so that multiple pods
access the same volume, in a sequece or parallel.

This makes it difficult to design a Pipeline that starts with parallel tasks using
its own PVC and then have a common tasks that mount the volume from the earlier
tasks - since - what happens if those tasks were scheduled to different zones -
the common task can not mount the PVCs that now is located in different zones, so
the PipelineRun is deadlocked.

There are a few technical solutions that offer parallel executions of Tasks
even when sharing PVC workspace:

- Using PVC access mode ReadWriteMany. But this access mode is not widely available,
  and is typically a NFS server or another not so "cloud native" solution.

- An alternative is to use a storage that is tied to a specific node, e.g. local volume
  and then configure so pods are scheduled to this node, but this is not commonly
  available and it has drawbacks, e.g. the pod may need to consume and mount a whole
  disk e.g. several hundreds GB.

Consequently, it would be good to find a way so that TaskRun pods that share
workspace are scheduled to the same Node - and thereby make it easy to use parallel
tasks with workspace - while executing concurrently - on widely available Kubernetes
cluster and storage configurations.

A few alternative solutions have been considered, as documented in tektoncd#2586.
However, they all have major drawbacks, e.g. major API and contract changes.

This commit introduces an "Affinity Assistant" - a minimal placeholder-pod,
so that it is possible to use [Kubernetes inter-pod affinity](https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/#inter-pod-affinity-and-anti-affinity) for TaskRun pods that need to be scheduled to the same Node.

This solution has several benefits: it does not introduce any API changes,
it does not break or change any existing Tekton concepts and it is
implemented with very few changes. Additionally it can be disabled with a feature-flag.

**How it works:** When a PipelineRun is initiated, an "Affinity Assistant" is
created for each PVC workspace volume. TaskRun pods that share workspace
volume is configured with podAffinity to the "Affinity Assisant" pod that
was created for the volume. The "Affinity Assistant" lives until the
PipelineRun is completed, or deleted. "Affinity Assistant" pods are
configured with podAntiAffinity to repel other "Affinity Assistants" -
in a Best Effort fashion.

The Affinity Assistant is _singleton_ workload, since it acts as a
placeholder pod and TaskRun pods with affinity must be scheduled to the
same Node. It is implemented with [QoS class Guaranteed](https://kubernetes.io/docs/tasks/configure-pod-container/quality-service-pod/#create-a-pod-that-gets-assigned-a-qos-class-of-guaranteed) but with minimal resource requests -
since it does not provide any work other than beeing a placeholder.

Singleton workloads can be implemented in multiple ways, and they differ
in behavior when the Node becomes unreachable:

- as a Pod - the Pod is not managed, so it will not be recreated.
- as a Deployment - the Pod will be recreated and puts Availability before
  the singleton property
- as a StatefulSet - the Pod will be recreated but puds the singleton
  property before Availability

Therefor the Affinity Assistant is implemented as a StatefulSet.

Essentialy this commit provides an effortless way to use a functional
task parallelism with any Kubernetes cluster that has any PVC based
storage.

Solves tektoncd#2586
/kind feature

Author: jlpettersson

config/200-clusterrole.yaml

@@ -52,7 +52,7 @@ rules:
     # Unclear if this access is actually required.  Simply a hold-over from the previous
     # incarnation of the controller's ClusterRole.
   - apiGroups: ["apps"]
-    resources: ["deployments"]
+    resources: ["deployments", "statefulsets"]
     verbs: ["get", "list", "create", "update", "delete", "patch", "watch"]
   - apiGroups: ["apps"]
     resources: ["deployments/finalizers"]

docs/install.md

@@ -268,6 +268,18 @@ file lists the keys you can customize along with their default values.
 
 To customize the behavior of the Pipelines Controller, modify the ConfigMap `feature-flags` as follows:
 
+- `disable-affinity-assistant` - set this flag to disable the [Affinity Assistant](./workspaces.md#affinity-assistant-and-specifying-workspace-order-in-a-pipeline)
+  that is used to provide Node Affinity for `TaskRun` pods that share workspace volume. The Assistant pods are
+  incompatible with NodeSelector and other affinity rules configured for `TaskRun` pods.
+
+  **Note:** Affinity Assistant use [Inter-pod affinity and anti-affinity](https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/#inter-pod-affinity-and-anti-affinity)
+  that require substantial amount of processing which can slow down scheduling in large clusters
+  significantly. We do not recommend using them in clusters larger than several hundred nodes
+
+  **Note:** Pod anti-affinity requires nodes to be consistently labelled, in other words every
+  node in the cluster must have an appropriate label matching `topologyKey`. If some or all nodes
+  are missing the specified `topologyKey` label, it can lead to unintended behavior.
+
 - `disable-home-env-overwrite` - set this flag to `true` to prevent Tekton
 from overriding the `$HOME` environment variable for the containers executing your `Steps`.
 The default is `false`. For more information, see the [associated issue](https://github.com/tektoncd/pipeline/issues/2013).

docs/labels.md

@@ -58,6 +58,8 @@ The following labels are added to resources automatically:
   reference a `ClusterTask` will also receive `tekton.dev/task`.
 - `tekton.dev/taskRun` is added to `Pods`, and contains the name of the
   `TaskRun` that created the `Pod`.
+- `app.kubernetes.io/instance` and `app.kubernetes.io/component` is added to 
+  Affinity Assistant `StatefulSets` and `Pods`. These are used for Pod Affinity for TaskRuns.
 
 ## Examples
 

docs/tasks.md

@@ -363,7 +363,8 @@ steps:
 ### Specifying `Workspaces`
 
 [`Workspaces`](workspaces.md#using-workspaces-in-tasks) allow you to specify
-one or more volumes that your `Task` requires during execution. For example:
+one or more volumes that your `Task` requires during execution. It is recommended that `Tasks` uses **at most**
+one writeable `Workspace`. For example:
 
 ```yaml
 spec:

docs/workspaces.md

@@ -15,7 +15,7 @@ weight: 5
     - [Mapping `Workspaces` in `Tasks` to `TaskRuns`](#mapping-workspaces-in-tasks-to-taskruns)
     - [Examples of `TaskRun` definitions using `Workspaces`](#examples-of-taskrun-definitions-using-workspaces)
   - [Using `Workspaces` in `Pipelines`](#using-workspaces-in-pipelines)
-    - [Specifying `Workspace` order in a `Pipeline`](#specifying-workspace-order-in-a-pipeline)
+    - [Affinity Assistant and specifying `Workspace` order in a `Pipeline`](#affinity-assistant-and-specifying-workspace-order-in-a-pipeline)
     - [Specifying `Workspaces` in `PipelineRuns`](#specifying-workspaces-in-pipelineruns)
     - [Example `PipelineRun` definition using `Workspaces`](#example-pipelinerun-definitions-using-workspaces)
   - [Specifying `VolumeSources` in `Workspaces`](#specifying-volumesources-in-workspaces)
@@ -89,7 +89,8 @@ To configure one or more `Workspaces` in a `Task`, add a `workspaces` list with
 
 Note the following:
 
-- A `Task` definition can include as many `Workspaces` as it needs. 
+- A `Task` definition can include as many `Workspaces` as it needs. It is recommended that `Tasks` use
+  **at most** one _writable_ `Workspace`.
 - A `readOnly` `Workspace` will have its volume mounted as read-only. Attempting to write
   to a `readOnly` `Workspace` will result in errors and failed `TaskRuns`.
 - `mountPath` can be either absolute or relative. Absolute paths start with `/` and relative paths
@@ -244,26 +245,27 @@ Include a `subPath` in the workspace binding to mount different parts of the sam
 
 The `subPath` specified in a `Pipeline` will be appended to any `subPath` specified as part of the `PipelineRun` workspace declaration. So a `PipelineRun` declaring a Workspace with `subPath` of `/foo` for a `Pipeline` who binds it to a `Task` with `subPath` of `/bar` will end up mounting the `Volume`'s `/foo/bar` directory.
 
-#### Specifying `Workspace` order in a `Pipeline`
+#### Affinity Assistant and specifying `Workspace` order in a `Pipeline`
 
 Sharing a `Workspace` between `Tasks` requires you to define the order in which those `Tasks`
-will be accessing that `Workspace` since different classes of storage have different limits
-for concurrent reads and writes. For example, a `PersistentVolumeClaim` with
-[access mode](https://kubernetes.io/docs/concepts/storage/persistent-volumes/#access-modes)
-`ReadWriteOnce` only allow `Tasks` on the same node writing to it at once.
-
-Using parallel `Tasks` in a `Pipeline` will work with `PersistentVolumeClaims` configured with
-[access mode](https://kubernetes.io/docs/concepts/storage/persistent-volumes/#access-modes)
-`ReadWriteMany` or `ReadOnlyMany` but you must ensure that those are available for your storage class.
-When using `PersistentVolumeClaims` with access mode `ReadWriteOnce` for parallel `Tasks`, you can configure a
-workspace with it's own `PersistentVolumeClaim` for each parallel `Task`.
-
-Use the `runAfter` field in your `Pipeline` definition to define when a `Task` should be executed. For more
-information, see the [`runAfter` documentation](pipelines.md#runAfter).
-
-**Warning:** You *must* ensure that this order is compatible with the configured access modes for your `PersistentVolumeClaim`.
-Parallel `Tasks` using the same `PersistentVolumeClaim` with access mode `ReadWriteOnce`, may execute on
-different nodes and be forced to execute sequentially which may cause `Tasks` to time out.
+write to or read from that `Workspace`. Use the `runAfter` field in your `Pipeline` definition
+to define when a `Task` should be executed. For more information, see the [`runAfter` documentation](pipelines.md#runAfter).
+
+When a `PersistentVolumeClaim` is used as volume source for a `Workspace` in a `PipelineRun`,
+an Affinity Assistant will be created. The Affinity Assistant acts as a placeholder for `TaskRun` pods
+sharing the same `Workspace`. All `TaskRun` pods within the `PipelineRun` that share the `Workspace`
+will be scheduled to the same Node as the Affinity Assistant pod. This means that Affinity Assistant is incompatible
+with e.g. NodeSelectors or other affinity rules configured for the `TaskRun` pods. The Affinity Assistant
+is deleted when the `PipelineRun` is completed. The Affinity Assistant can be disabled by setting the
+[disable-affinity-assistant](install.md#customizing-basic-execution-parameters) feature gate.
+
+**Note:** Affinity Assistant use [Inter-pod affinity and anti-affinity](https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/#inter-pod-affinity-and-anti-affinity)
+that require substantial amount of processing which can slow down scheduling in large clusters
+significantly. We do not recommend using them in clusters larger than several hundred nodes
+
+**Note:** Pod anti-affinity requires nodes to be consistently labelled, in other words every
+node in the cluster must have an appropriate label matching `topologyKey`. If some or all nodes
+are missing the specified `topologyKey` label, it can lead to unintended behavior.
 
 #### Specifying `Workspaces` in `PipelineRuns`
 

examples/v1beta1/pipelineruns/pipeline-run-with-parallel-tasks-using-pvc.yaml

@@ -0,0 +1,202 @@
+# This example shows how both sequiential and parallel Tasks can share data
+# using a PersistentVolumeClaim as a workspace. The TaskRun pods that share
+# workspace will be scheduled to the same Node in your cluster with an
+# Affinity Assistant (unless it is disabled). The REPORTER task does not
+# use a workspace so it does not get affinity to the Affinity Assistant
+# and can be scheduled to any Node. If multiple concurrent PipelineRuns is
+# executed, their Affinity Assistant pods will repel eachother to different
+# Nodes in a Best Effort fashion.
+#
+# A PipelineRun will pass a message parameter to the Pipeline in this example.
+# The STARTER task will write the message to a file in the workspace. The UPPER
+# and LOWER tasks will execute in parallel and process the message written by
+# the STARTER, and transform it to upper case and lower case. The REPORTER task
+# is will use the Task Result from the UPPER task and print it - it is intented
+# to mimic a Task that sends data to an external service and shows a Task that
+# doesn't use a workspace. The VALIDATOR task will validate the result from
+# UPPER and LOWER.
+#
+# Use the runAfter property in a Pipeline to configure that a task depend on
+# another task. Output can be shared both via Task Result (e.g. like REPORTER task)
+# or via files in a workspace.
+#
+#             -- (upper) -- (reporter)
+#           /                         \
+#  (starter)                           (validator)
+#           \                         /
+#             -- (lower) ------------
+
+apiVersion: tekton.dev/v1beta1
+kind: Pipeline
+metadata:
+  name: parallel-pipeline
+spec:
+  params:
+    - name: message
+      type: string
+
+  workspaces:
+    - name: ws
+
+  tasks:
+    - name: starter          # Tasks that does not declare a runAfter property
+      taskRef:               # will start execution immediately
+        name: persist-param
+      params:
+        - name: message
+          value: $(params.message)
+      workspaces:
+        - name: task-ws
+          workspace: ws
+          subPath: init
+
+    - name: upper
+      runAfter:               # Note the use of runAfter her to declare that this task
+        - starter             # depend on a previous task
+      taskRef:
+        name: to-upper
+      params:
+        - name: input-path
+          value: init/message
+      workspaces:
+        - name: w
+          workspace: ws
+
+    - name: lower
+      runAfter:
+        - starter
+      taskRef:
+        name: to-lower
+      params:
+        - name: input-path
+          value: init/message
+      workspaces:
+        - name: w
+          workspace: ws
+
+    - name: reporter
+      runAfter:
+        - upper
+      taskRef:
+        name: result-reporter
+      params:
+        - name: result-to-report
+          value: $(tasks.upper.results.message)
+
+    - name: validator         # This task validate the output from upper and lower Task
+      runAfter:               # It does not strictly depend on the reporter Task
+        - reporter            # But you may want to skip this task if the reporter Task fail
+        - lower
+      taskRef:
+        name: validator
+      workspaces:
+        - name: files
+          workspace: ws
+---
+apiVersion: tekton.dev/v1beta1
+kind: Task
+metadata:
+  name: persist-param
+spec:
+  params:
+    - name: message
+      type: string
+  results:
+    - name: message
+      description: A result message
+  steps:
+    - name: write
+      image: ubuntu
+      script: echo $(params.message) | tee $(workspaces.task-ws.path)/message $(results.message.path)
+  workspaces:
+    - name: task-ws
+---
+apiVersion: tekton.dev/v1beta1
+kind: Task
+metadata:
+  name: to-upper
+spec:
+  description: |
+    This task read and process a file from the workspace and write the result
+    both to a file in the workspace and as a Task Result.
+  params:
+    - name: input-path
+      type: string
+  results:
+    - name: message
+      description: Input message in upper case
+  steps:
+    - name: to-upper
+      image: ubuntu
+      script: cat $(workspaces.w.path)/$(params.input-path) | tr '[:lower:]' '[:upper:]' | tee $(workspaces.w.path)/upper $(results.message.path)
+  workspaces:
+    - name: w
+---
+apiVersion: tekton.dev/v1beta1
+kind: Task
+metadata:
+  name: to-lower
+spec:
+  description: |
+    This task read and process a file from the workspace and write the result
+    both to a file in the workspace and as a Task Result
+  params:
+    - name: input-path
+      type: string
+  results:
+    - name: message
+      description: Input message in lower case
+  steps:
+    - name: to-lower
+      image: ubuntu
+      script: cat $(workspaces.w.path)/$(params.input-path) | tr '[:upper:]' '[:lower:]' | tee $(workspaces.w.path)/lower $(results.message.path)
+  workspaces:
+    - name: w
+---
+apiVersion: tekton.dev/v1beta1
+kind: Task
+metadata:
+  name: result-reporter
+spec:
+  params:
+    - name: result-to-report
+      type: string
+  steps:
+    - name: report-result
+      image: ubuntu
+      script: echo $(params.result-to-report)
+---
+apiVersion: tekton.dev/v1beta1
+kind: Task
+metadata:
+  name: validator
+spec:
+  steps:
+    - name: validate-upper
+      image: ubuntu
+      script: cat $(workspaces.files.path)/upper | grep HELLO\ TEKTON
+    - name: validate-lower
+      image: ubuntu
+      script: cat $(workspaces.files.path)/lower | grep hello\ tekton
+  workspaces:
+    - name: files
+---
+apiVersion: tekton.dev/v1beta1
+kind: PipelineRun
+metadata:
+  generateName: parallel-pipelinerun-
+spec:
+  params:
+    - name: message
+      value: Hello Tekton
+  pipelineRef:
+    name: parallel-pipeline
+  workspaces:
+    - name: ws
+      volumeClaimTemplate:
+        spec:
+          accessModes:
+            - ReadWriteOnce
+          resources:
+            requests:
+              storage: 1Gi

pkg/pod/pod.go

@@ -26,6 +26,7 @@ import (
 	"github.com/tektoncd/pipeline/pkg/names"
 	"github.com/tektoncd/pipeline/pkg/system"
 	"github.com/tektoncd/pipeline/pkg/version"
+	"github.com/tektoncd/pipeline/pkg/workspace"
 	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime/schema"
@@ -219,6 +220,17 @@ func MakePod(images pipeline.Images, taskRun *v1beta1.TaskRun, taskSpec v1beta1.
 		return nil, err
 	}
 
+	// Using node affinity on taskRuns sharing PVC workspace, with an Affinity Assistant
+	// is mutual exclusive with other affinity on taskRun pods. If other
+	// affinity is wanted, that should be added on the Affinity Assistant pod unless
+	// assistant is disabled. When Affinity Assistant is disabled, an affinityAssistantName is not set.
+	var affinity *corev1.Affinity
+	if affinityAssistantName := taskRun.Annotations[workspace.AnnotationAffinityAssistantName]; affinityAssistantName != "" {
+		affinity = nodeAffinityUsingAffinityAssistant(affinityAssistantName)
+	} else {
+		affinity = podTemplate.Affinity
+	}
+
 	mergedPodContainers := stepContainers
 
 	// Merge sidecar containers with step containers.
@@ -265,7 +277,7 @@ func MakePod(images pipeline.Images, taskRun *v1beta1.TaskRun, taskSpec v1beta1.
 			Volumes:                      volumes,
 			NodeSelector:                 podTemplate.NodeSelector,
 			Tolerations:                  podTemplate.Tolerations,
-			Affinity:                     podTemplate.Affinity,
+			Affinity:                     affinity,
 			SecurityContext:              podTemplate.SecurityContext,
 			RuntimeClassName:             podTemplate.RuntimeClassName,
 			AutomountServiceAccountToken: podTemplate.AutomountServiceAccountToken,
@@ -296,6 +308,25 @@ func MakeLabels(s *v1beta1.TaskRun) map[string]string {
 	return labels
 }
 
+// nodeAffinityUsingAffinityAssistant achieves Node Affinity for taskRun pods
+// sharing PVC workspace by setting PodAffinity so that taskRuns is
+// scheduled to the Node were the Affinity Assistant pod is scheduled.
+func nodeAffinityUsingAffinityAssistant(affinityAssistantName string) *corev1.Affinity {
+	return &corev1.Affinity{
+		PodAffinity: &corev1.PodAffinity{
+			RequiredDuringSchedulingIgnoredDuringExecution: []corev1.PodAffinityTerm{{
+				LabelSelector: &metav1.LabelSelector{
+					MatchLabels: map[string]string{
+						workspace.LabelInstance:  affinityAssistantName,
+						workspace.LabelComponent: workspace.ComponentNameAffinityAssistant,
+					},
+				},
+				TopologyKey: "kubernetes.io/hostname",
+			}},
+		},
+	}
+}
+
 // getLimitRangeMinimum gets all LimitRanges in a namespace and
 // searches for if a container minimum is specified. Due to
 // https://github.com/kubernetes/kubernetes/issues/79496, the