Skip to content

Commit

Permalink
[TEP-0135] Update Affinity Assistant documentation
Browse files Browse the repository at this point in the history
Part of [#6740][#6740]. This commit updates the Affinity Assistant related documentation to reflect recent changes in the Affinity Assistant modes.
This commit also adds a chart summarizing how to configure the Affinity Assistant modes using `coschedule` and `disabled-affinity-assistant` feature
flags.

/kind documentation

[#6740]: #6740
  • Loading branch information
QuanZhang-William authored and tekton-robot committed Jul 26, 2023
1 parent 5703d8f commit c38c220
Show file tree
Hide file tree
Showing 4 changed files with 125 additions and 26 deletions.
7 changes: 4 additions & 3 deletions config/config-feature-flags.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -26,8 +26,8 @@ data:
#
# The default behaviour is for Tekton to create Affinity Assistants
#
# See more in the workspace documentation about Affinity Assistant
# https://github.com/tektoncd/pipeline/blob/main/docs/workspaces.md#affinity-assistant-and-specifying-workspace-order-in-a-pipeline
# See more in the Affinity Assistant documentation
# https://github.com/tektoncd/pipeline/blob/main/docs/affinityassistants.md
# or https://github.com/tektoncd/pipeline/pull/2630 for more info.
disable-affinity-assistant: "false"
# Setting this flag will determine how PipelineRun Pods are scheduled with Affinity Assistant.
Expand All @@ -39,7 +39,8 @@ data:
# and only allows one pipelinerun to run on a node at a time.
# Setting it to "disabled" will not apply any coschedule policy.
#
# TODO: add links to documentation and migration strategy
# See more in the Affinity Assistant documentation
# https://github.com/tektoncd/pipeline/blob/main/docs/affinityassistants.md
# NOTE: this feature is still under development and not yet functional.
coschedule: "workspaces"
# Setting this flag to "true" will prevent Tekton scanning attached
Expand Down
10 changes: 9 additions & 1 deletion docs/additional-configs.md
Original file line number Diff line number Diff line change
Expand Up @@ -210,7 +210,7 @@ that are running while the change occurs.

The flags in this ConfigMap are as follows:

- `disable-affinity-assistant` - set this flag to `true` to disable the [Affinity Assistant](./workspaces.md#specifying-workspace-order-in-a-pipeline-and-affinity-assistants)
- `disable-affinity-assistant` - set this flag to `true` to disable the [Affinity Assistant](./affinityassistants)
that is used to provide Node Affinity for `TaskRun` pods that share workspace volume.
The Affinity Assistant is incompatible with other affinity rules
configured for `TaskRun` pods.
Expand All @@ -223,6 +223,13 @@ The flags in this ConfigMap are as follows:
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.

- `coschedule`: set this flag determines how PipelineRun Pods are scheduled with [Affinity Assistant](./affinityassistants).
Acceptable values are "workspaces" (default), "pipelineruns", "isolate-pipelinerun", or "disabled".
Setting it to "workspaces" will schedule all the taskruns sharing the same PVC-based workspace in a pipelinerun to the same node.
Setting it to "pipelineruns" will schedule all the taskruns in a pipelinerun to the same node.
Setting it to "isolate-pipelinerun" will schedule all the taskruns in a pipelinerun to the same node,
and only allows one pipelinerun to run on a node at a time. Setting it to "disabled" will not apply any coschedule policy.

- `await-sidecar-readiness`: set this flag to `"false"` to allow the Tekton controller to start a
TasksRun's first step immediately without waiting for sidecar containers to be running first. Using
this option should decrease the time it takes for a TaskRun to start running, and will allow TaskRun
Expand Down Expand Up @@ -307,6 +314,7 @@ Features currently in "alpha" are:
| [Trusted Resources](./trusted-resources.md) | [TEP-0091](https://github.com/tektoncd/community/blob/main/teps/0091-trusted-resources.md) | N/A | `trusted-resources-verification-no-match-policy` |
| [Larger Results via Sidecar Logs](#enabling-larger-results-using-sidecar-logs) | [TEP-0127](https://github.com/tektoncd/community/blob/main/teps/0127-larger-results-via-sidecar-logs.md) | [v0.43.0](https://github.com/tektoncd/pipeline/releases/tag/v0.43.0) | `results-from` |
| [Configure Default Resolver](./resolution.md#configuring-built-in-resolvers) | [TEP-0133](https://github.com/tektoncd/community/blob/main/teps/0133-configure-default-resolver.md) | N/A | |
| [Coschedule](./affinityassistants.md) | [TEP-0135](https://github.com/tektoncd/community/blob/main/teps/0135-coscheduling-pipelinerun-pods.md) | N/A |`coschedule` |

### Beta Features

Expand Down
111 changes: 111 additions & 0 deletions docs/affinityassistants.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,111 @@
<!--
---
linkTitle: "Affinity Assistants"
weight: 405
---
-->

# Affinity Assistants
Affinity Assistant is a feature to coschedule `PipelineRun` `pods` to the same node
based on [kubernetes pod affinity](https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/#inter-pod-affinity-and-anti-affinity) so that it possible for the taskruns to execute parallel while sharing volume.
Available Affinity Assistant Modes are **coschedule workspaces**, **coschedule pipelineruns**,
**isolate pipelinerun** and **disabled**.

> :seedling: **coschedule pipelineruns** and **isolate pipelinerun** modes are [**alpha features**](./additional-configs.md#alpha-features).
> **coschedule workspaces** is a **stable feature**
* **coschedule workspaces** - When a `PersistentVolumeClaim` is used as volume source for a `Workspace` in a `PipelineRun`,
all `TaskRun` pods within the `PipelineRun` that share the `Workspace` will be scheduled to the same Node. (**Note:** Only one pvc-backed workspace can be mounted to each TaskRun in this mode.)

* **coschedule pipelineruns** - All `TaskRun` pods within the `PipelineRun` will be scheduled to the same Node.

* **isolate pipelinerun** - All `TaskRun` pods within the `PipelineRun` will be scheduled to the same Node,
and only one PipelineRun is allowed to run on a node at a time.

* **disabled** - The Affinity Assistant is disabled. No pod coscheduling behavior.

This means that Affinity Assistant is incompatible with other affinity rules
configured for the `TaskRun` pods (i.e. other affinity rules specified in custom [PodTemplate](pipelineruns.md#specifying-a-pod-template) will be overwritten by Affinity Assistant).
If the `PipelineRun` has a custom [PodTemplate](pipelineruns.md#specifying-a-pod-template) configured, the `NodeSelector` and `Tolerations` fields will also be set on the Affinity Assistant pod. The Affinity Assistant
is deleted when the `PipelineRun` is completed.

Currently, the Affinity Assistant Modes can be configured by the `disable-affinity-assistant` and `coschedule` feature flags. In 9 months, the `disable-affinity-assistant` feature flag will be deprecated, and the Affinity Assistant Modes will be only determined by the `coschedule` feature flag.

The following chart summarizes the Affinity Assistant Modes with different combinations of the `disable-affinity-assistant` and `coschedule` feature flags during migration (when both feature flags are present) and after the migration (when only the `coschedule` flag is present).

<table>
<thead>
<tr>
<th>disable-affinity-assistant</th>
<th>coschedule</th>
<th>behavior during migration</th>
<th>behavior after migration</th>
</tr>
</thead>
<tbody>
<tr>
<td>false (default)</td>
<td>disabled</td>
<td>N/A: invalid</td>
<td>disabled</td>
</tr>
<tr>
<td>false (default)</td>
<td>workspaces (default)</td>
<td>coschedule workspaces</td>
<td>coschedule workspaces</td>
</tr>
<tr>
<td>false (default)</td>
<td>pipelineruns</td>
<td>N/A: invalid</td>
<td>coschedule pipelineruns</td>
</tr>
<tr>
<td>false (default)</td>
<td>isolate-pipelinerun</td>
<td>N/A: invalid</td>
<td>isolate pipelinerun</td>
</tr>
<tr>
<td>true</td>
<td>disabled</td>
<td>disabled</td>
<td>disabled</td>
</tr>
<tr>
<td>true</td>
<td>workspaces (default)</td>
<td>disabled</td>
<td>coschedule workspaces</td>
</tr>
<tr>
<td>true</td>
<td>pipelineruns</td>
<td>coschedule pipelineruns</td>
<td>coschedule pipelineruns</td>
</tr>
<tr>
<td>true</td>
<td>isolate-pipelinerun</td>
<td>isolate pipelinerun</td>
<td>isolate pipelinerun</td>
</tr>
</tbody>
</table>

**Note:** For users who previously accepted the default behavior (`disable-affinity-assistant`: `false`) but now want one of the new features, you need to set `disable-affinity-assistant` to "true" and then turn on the new behavior by setting the `coschedule` flag. For users who previously disabled the affinity assistant but want one of the new features, just set the `coschedule` flag accordingly.

**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 the affinity assistant 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.

**Note:** Any time during the execution of a `pipelineRun`, if the node with a placeholder Affinity Assistant pod and
the `taskRun` pods sharing a `workspace` is `cordoned` or disabled for scheduling anything new (`tainted`), the
`pipelineRun` controller deletes the placeholder pod. The `taskRun` pods on a `cordoned` node continues running
until completion. The deletion of a placeholder pod triggers creating a new placeholder pod on any available node
such that the rest of the `pipelineRun` can continue without any disruption until it finishes.
23 changes: 1 addition & 22 deletions docs/workspaces.md
Original file line number Diff line number Diff line change
Expand Up @@ -364,28 +364,7 @@ write to or read from that `Workspace`. Use the `runAfter` field in your `Pipeli
to define when a `Task` should be executed. For more information, see the [`runAfter` documentation](pipelines.md#using-the-runafter-parameter).

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. other affinity rules configured for the `TaskRun` pods. If the `PipelineRun` has a custom
[PodTemplate](pipelineruns.md#specifying-a-pod-template) configured, the `NodeSelector` and `Tolerations` fields
will also be set on the Affinity Assistant pod. 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 to `true`.

**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.

**Note:** Any time during the execution of a `pipelineRun`, if the node with a placeholder Affinity Assistant pod and
the `taskRun` pods sharing a `workspace` is `cordoned` or disabled for scheduling anything new (`tainted`), the
`pipelineRun` controller deletes the placeholder pod. The `taskRun` pods on a `cordoned` node continues running
until completion. The deletion of a placeholder pod triggers creating a new placeholder pod on any available node
such that the rest of the `pipelineRun` can continue without any disruption until it finishes.
an Affinity Assistant will be created. For more information, see the [`Affinity Assistants` documentation](affinityassistants.md).

#### Specifying `Workspaces` in `PipelineRuns`

Expand Down

0 comments on commit c38c220

Please sign in to comment.