Skip to content

buildstrategies.md

Latest commit

 

History

History
535 lines (424 loc) · 21.7 KB

buildstrategies.md

File metadata and controls

535 lines (424 loc) · 21.7 KB
Raw

BuildStrategies

Overview

There are two types of strategies, the ClusterBuildStrategy (clusterbuildstrategies.shipwright.io/v1alpha1) and the BuildStrategy (buildstrategies.shipwright.io/v1alpha1). Both strategies define a shared group of steps, needed to fullfil the application build.

A ClusterBuildStrategy is available cluster-wide, while a BuildStrategy is available within a namespace.

Available ClusterBuildStrategies

Well-known strategies can be bootstrapped from here. The currently supported Cluster BuildStrategy are:

Name Supported platforms
buildah linux/amd64 only
BuildKit all
buildpacks-v3-heroku linux/amd64 only
buildpacks-v3 linux/amd64 only
kaniko all
ko all
source-to-image linux/amd64 only

Available BuildStrategies

The current supported namespaces BuildStrategy are:

Name Supported platforms
buildpacks-v3-heroku linux/amd64 only
buildpacks-v3 linux/amd64 only

Buildah

The buildah ClusterBuildStrategy consists of using buildah to build and push a container image, out of a Dockerfile. The Dockerfile should be specified on the Build resource.

Installing Buildah Strategy

To install use:

kubectl apply -f samples/buildstrategy/buildah/buildstrategy_buildah_cr.yaml

Buildpacks v3

The buildpacks-v3 BuildStrategy/ClusterBuildStrategy uses a Cloud Native Builder (CNB) container image, and is able to implement lifecycle commands. The following CNB images are the most common options:

Installing Buildpacks v3 Strategy

You can install the BuildStrategy in your namespace or install the ClusterBuildStrategy at cluster scope so that it can be shared across namespaces.

To install the cluster scope strategy, use (below is a heroku example, you can also use paketo sample):

kubectl apply -f samples/buildstrategy/buildpacks-v3/buildstrategy_buildpacks-v3-heroku_cr.yaml

To install the namespaced scope strategy, use:

kubectl apply -f samples/buildstrategy/buildpacks-v3/buildstrategy_buildpacks-v3-heroku_namespaced_cr.yaml

Kaniko

The kaniko ClusterBuildStrategy is composed by Kaniko's executor kaniko, with the objective of building a container-image, out of a Dockerfile and context directory.

Installing Kaniko Strategy

To install the cluster scope strategy, use:

kubectl apply -f samples/buildstrategy/kaniko/buildstrategy_kaniko_cr.yaml

BuildKit

BuildKit is composed of the buildctl client and the buildkitd daemon. For the buildkit ClusterBuildStrategy, it runs on a daemonless mode, where both client and ephemeral daemon run in a single container. In addition, it runs without privileges ( rootless ).

The buildkit-insecure ClusterBuildStrategy exists to support users pushing to an insecure HTTP registry. We use this strategy at the moment only for testing purposes against a local in-cluster registry. In the future, this strategy will be removed in favor of a single one where users can parameterize the secure/insecure behaviour.

Cache Exporters

By default, the buildkit ClusterBuildStrategy will use caching to optimize the build times. When pushing an image to a registry, it will use the inline export cache, which pushes the image and cache together. Please refer to export-cache docs for more information.

Known Limitations

The buildkit ClusterBuildStrategy currently locks the following parameters:

  • A Dockerfile name needs to be Dockerfile, this is currently not configurable.
  • Exporter caches are enable by default, this is currently not configurable.
  • To allow running rootless, it requires both AppArmor as well as SecComp to be disabled using the unconfined profile.

Installing BuildKit Strategy

To install the cluster scope strategy, use:

kubectl apply -f samples/buildstrategy/buildkit/buildstrategy_buildkit_cr.yaml

ko

The ko ClusterBuilderStrategy is using ko's publish command to build an image from a Golang main package.

Installing ko Strategy

To install the cluster scope strategy, use:

kubectl apply -f samples/buildstrategy/ko/buildstrategy_ko_cr.yaml

Note: The build strategy currently uses the spec.contextDir of the Build in a different way than this property is designed for: the Git repository must be a Go module with the go.mod file at the root. The contextDir specifies the path to the main package. You can check the example which is set up to build the Shipwright Build controller. This behavior will eventually be corrected once Exhaustive list of generalized Build API/CRD attributes #184 / Custom attributes from the Build CR could be used as parameters while defining a BuildStrategy #537 are done.

Note: The build strategy is setup to build for the platform that your Kubernetes cluster is running. Exposing the platform configuration to the Build requires the same features mentioned on the previous note.

Source to Image

This BuildStrategy is composed by source-to-image and kaniko in order to generate a Dockerfile and prepare the application to be built later on with a builder.

s2i requires a specially crafted image, which can be informed as builderImage parameter on the Build resource.

Installing Source to Image Strategy

To install the cluster scope strategy use:

kubectl apply -f samples/buildstrategy/source-to-image/buildstrategy_source-to-image_cr.yaml

Build Steps

  1. s2i in order to generate a Dockerfile and prepare source-code for image build;
  2. kaniko to create and push the container image to what is defined as output.image;

System parameters

You can use parameters when defining the steps of a build strategy to access system information as well as information provided by the user in his Build or BuildRun. The following parameters are available:

Parameter Description
$(params.shp-source-root) The absolute path to the directory that contains the user's sources.
$(params.shp-source-context) The absolute path to the context directory of the user's sources. If the user specified no value for spec.source.contextDir in his Build, then this value will equal the value for $(params.shp-source-root). Note that this directory is not guaranteed to exist at the time the container for your step is started, you can therefore not use this parameter as a step's working directory.
$(params.shp-output-image) The URL of the image that the user wants to push as specified in the Build's spec.output.image, or the override from the BuildRun's spec.output.image.

Steps Resource Definition

All strategies steps can include a definition of resources(limits and requests) for CPU, memory and disk. For strategies with more than one step, each step(container) could require more resources than others. Strategy admins are free to define the values that they consider the best fit for each step. Also, identical strategies with the same steps that are only different in their name and step resources can be installed on the cluster to allow users to create a build with smaller and larger resource requirements.

Strategies with different resources

If the strategy admins would require to have multiple flavours of the same strategy, where one strategy has more resources that the other. Then, multiple strategies for the same type should be defined on the cluster. In the following example, we use Kaniko as the type:

---
apiVersion: shipwright.io/v1alpha1
kind: ClusterBuildStrategy
metadata:
  name: kaniko-small
spec:
  buildSteps:
    - name: build-and-push
      image: gcr.io/kaniko-project/executor:v1.6.0
      workingDir: $(params.shp-source-root)
      securityContext:
        runAsUser: 0
        capabilities:
          add:
            - CHOWN
            - DAC_OVERRIDE
            - FOWNER
            - SETGID
            - SETUID
            - SETFCAP
            - KILL
      env:
        - name: DOCKER_CONFIG
          value: /tekton/home/.docker
        - name: AWS_ACCESS_KEY_ID
          value: NOT_SET
        - name: AWS_SECRET_KEY
          value: NOT_SET
      command:
        - /kaniko/executor
      args:
        - --skip-tls-verify=true
        - --dockerfile=$(build.dockerfile)
        - --context=$(params.shp-source-context)
        - --destination=$(params.shp-output-image)
        - --oci-layout-path=/workspace/output/image
        - --snapshotMode=redo
        - --push-retry=3
      resources:
        limits:
          cpu: 250m
          memory: 65Mi
        requests:
          cpu: 250m
          memory: 65Mi
---
apiVersion: shipwright.io/v1alpha1
kind: ClusterBuildStrategy
metadata:
  name: kaniko-medium
spec:
  buildSteps:
    - name: build-and-push
      image: gcr.io/kaniko-project/executor:v1.6.0
      workingDir: $(params.shp-source-root)
      securityContext:
        runAsUser: 0
        capabilities:
          add:
            - CHOWN
            - DAC_OVERRIDE
            - FOWNER
            - SETGID
            - SETUID
            - SETFCAP
            - KILL
      env:
        - name: DOCKER_CONFIG
          value: /tekton/home/.docker
        - name: AWS_ACCESS_KEY_ID
          value: NOT_SET
        - name: AWS_SECRET_KEY
          value: NOT_SET
      command:
        - /kaniko/executor
      args:
        - --skip-tls-verify=true
        - --dockerfile=$(build.dockerfile)
        - --context=$(params.shp-source-context)
        - --destination=$(params.shp-output-image)
        - --oci-layout-path=/workspace/output/image
        - --snapshotMode=redo
        - --push-retry=3
      resources:
        limits:
          cpu: 500m
          memory: 1Gi
        requests:
          cpu: 500m
          memory: 1Gi

The above provides more control and flexibility for the strategy admins. For end-users, all they need to do, is to reference the proper strategy. For example:

---
apiVersion: shipwright.io/v1alpha1
kind: Build
metadata:
  name: kaniko-medium
spec:
  source:
    url: https://github.com/shipwright-io/sample-go
    contextDir: docker-build
  strategy:
    name: kaniko
    kind: ClusterBuildStrategy
  dockerfile: Dockerfile

How does Tekton Pipelines handle resources

The Build controller relies on the Tekton pipeline controller to schedule the pods that execute the above strategy steps. In a nutshell, the Build controller creates on run-time a Tekton TaskRun, and the TaskRun generates a new pod in the particular namespace. In order to build an image, the pod executes all the strategy steps one-by-one.

Tekton manage each step resources request in a very particular way, see the docs. From this document, it mentions the following:

The CPU, memory, and ephemeral storage resource requests will be set to zero, or, if specified, the minimums set through LimitRanges in that Namespace, if the container image does not have the largest resource request out of all container images in the Task. This ensures that the Pod that executes the Task only requests enough resources to run a single container image in the Task rather than hoard resources for all container images in the Task at once.

Examples of Tekton resources management

For a more concrete example, let´s take a look on the following scenarios:

Scenario 1. Namespace without LimitRange, both steps with the same resource values.

If we will apply the following resources:

We will see some differences between the TaskRun definition and the pod definition.

For the TaskRun, as expected we can see the resources on each step, as we previously define on our strategy.

$ kubectl -n test-build get tr buildah-golang-buildrun-9gmcx-pod-lhzbc -o json | jq '.spec.taskSpec.steps[] | select(.name == "step-buildah-bud" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "250m",
    "memory": "65Mi"
  }
}

$ kubectl -n test-build get tr buildah-golang-buildrun-9gmcx-pod-lhzbc -o json | jq '.spec.taskSpec.steps[] | select(.name == "step-buildah-push" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "250m",
    "memory": "65Mi"
  }
}

The pod definition is different, while Tekton will only use the highest values of one container, and set the rest(lowest) to zero:

$ kubectl -n test-build get pods buildah-golang-buildrun-9gmcx-pod-lhzbc -o json | jq '.spec.containers[] | select(.name == "step-step-buildah-bud" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "250m",
    "ephemeral-storage": "0",
    "memory": "65Mi"
  }
}

$ kubectl -n test-build get pods buildah-golang-buildrun-9gmcx-pod-lhzbc -o json | jq '.spec.containers[] | select(.name == "step-step-buildah-push" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "0",               <------------------- See how the request is set to ZERO.
    "ephemeral-storage": "0", <------------------- See how the request is set to ZERO.
    "memory": "0"             <------------------- See how the request is set to ZERO.
  }
}

In this scenario, only one container can have the spec.resources.requests definition. Even when both steps have the same values, only one container will get them, the others will be set to zero.

Scenario 2. Namespace without LimitRange, steps with different resources:

If we will apply the following resources:

  • buildahBuild

  • buildahBuildRun

  • We will use a modified buildah strategy, with the following steps resources:

      - name: buildah-bud
    image: quay.io/containers/buildah:v1.20.1
    workingDir: $(params.shp-source-root)
    securityContext:
      privileged: true
    command:
      - /usr/bin/buildah
    args:
      - bud
      - --tag=$(params.shp-output-image)
      - --file=$(build.dockerfile)
      - $(build.source.contextDir)
    resources:
      limits:
        cpu: 500m
        memory: 1Gi
      requests:
        cpu: 250m
        memory: 65Mi
    volumeMounts:
      - name: buildah-images
        mountPath: /var/lib/containers/storage
  - name: buildah-push
    image: quay.io/containers/buildah:v1.20.1
    securityContext:
      privileged: true
    command:
      - /usr/bin/buildah
    args:
      - push
      - --tls-verify=false
      - docker://$(params.shp-output-image)
    resources:
      limits:
        cpu: 500m
        memory: 1Gi
      requests:
        cpu: 250m
        memory: 100Mi  <------ See how we provide more memory to step-buildah-push, compared to the 65Mi of the other step

For the TaskRun, as expected we can see the resources on each step.

$ kubectl -n test-build get tr buildah-golang-buildrun-skgrp -o json | jq '.spec.taskSpec.steps[] | select(.name == "step-buildah-bud" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "250m",
    "memory": "65Mi"
  }
}

$ kubectl -n test-build get tr buildah-golang-buildrun-skgrp -o json | jq '.spec.taskSpec.steps[] | select(.name == "step-buildah-push" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "250m",
    "memory": "100Mi"
  }
}

The pod definition is different, while Tekton will only use the highest values of one container, and set the rest(lowest) to zero:

$ kubectl -n test-build get pods buildah-golang-buildrun-95xq8-pod-mww8d -o json | jq '.spec.containers[] | select(.name == "step-step-buildah-bud" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "250m",                <------------------- See how the CPU is preserved
    "ephemeral-storage": "0",
    "memory": "0"                 <------------------- See how the memory is set to ZERO
  }
}
$ kubectl -n test-build get pods buildah-golang-buildrun-95xq8-pod-mww8d -o json | jq '.spec.containers[] | select(.name == "step-step-buildah-push" ) | .resources'
{
  "limits": {
    "cpu": "500m",
    "memory": "1Gi"
  },
  "requests": {
    "cpu": "0",                     <------------------- See how the CPU is set to zero.
    "ephemeral-storage": "0",
    "memory": "100Mi"               <------------------- See how the memory is preserved on this container
  }
}

In the above scenario, we can see how the maximum numbers for resource requests are distributed between containers. The container step-buildah-push gets the 100mi for the memory requests, while it was the one defining the highest number. At the same time, the container step-buildah-bud is assigned a 0 for its memory request.

Scenario 3. Namespace with a LimitRange.

When a LimitRange exists on the namespace, Tekton Pipeline controller will do the same approach as stated in the above two scenarios. The difference is that for the containers that have lower values, instead of zero, they will get the minimum values of the LimitRange.

Annotations

Annotations can be defined for a BuildStrategy/ClusterBuildStrategy as for any other Kubernetes object. Annotations are propagated to the TaskRun and from there, Tekton propagates them to the Pod. Use cases for this are for example:

  • The Kubernetes Network Traffic Shaping feature looks for the kubernetes.io/ingress-bandwidth and kubernetes.io/egress-bandwidth annotations to limit the network bandwidth the Pod is allowed to use.
  • The AppArmor profile of a container is defined using the container.apparmor.security.beta.kubernetes.io/<container_name> annotation.

The following annotations are not propagated:

  • kubectl.kubernetes.io/last-applied-configuration
  • clusterbuildstrategy.shipwright.io/*
  • buildstrategy.shipwright.io/*
  • build.shipwright.io/*
  • buildrun.shipwright.io/*

A Kubernetes administrator can further restrict the usage of annotations by using policy engines like Open Policy Agent.