[WIP] Feature Autoscaling: Enable organization runner autoscaler

README.md

@@ -205,7 +205,9 @@ example-runnerdeploy2475h595fr   mumoshu/actions-runner-controller-ci   Running
 example-runnerdeploy2475ht2qbr   mumoshu/actions-runner-controller-ci   Running
 ```
 
-#### Autoscaling
+### Autoscaling
+
+#### Repository runners Autoscaling
 
 `RunnerDeployment` can scale the number of runners between `minReplicas` and `maxReplicas` fields, depending on pending workflow runs.
 
@@ -241,6 +243,8 @@ The scale out performance is controlled via the manager containers startup `--sy
 Additionally, the autoscaling feature has an anti-flapping option that prevents periodic loop of scaling up and down.
 By default, it doesn't scale down until the grace period of 10 minutes passes after a scale up. The grace period can be configured by setting `scaleDownDelaySecondsAfterScaleUp`:
 
+Please note that if your `RunnerDeployment` has the `Repository` key set, then do not use the `<owner/repository_name>` notation, only provide the `<repository_name>`.
+
 ```yaml
 apiVersion: actions.summerwind.dev/v1alpha1
 kind: RunnerDeployment
@@ -267,6 +271,36 @@ spec:
     - summerwind/actions-runner-controller
 ```
 
+#### Organization runners Autoscaling
+To autoscale on an organizational level, you need to remove the `repositoryNames` mapping and leave it empty. The Github Actions API doesn’t offer an endpoint to check the currently queued workflows on an organizational level. The way how the controller tries to get around this is by - after each `sync-period` - select the repositories with the latest `pushed` time and check the Actions workflow queue of those repositories. At the moment, the controller checks the last 10 repositories.
+
+Please note, in case you want to autoscale your organization runners, that you should modify your Github organization permissions accordingly; for instance, if you are using an organization PAT (Personal Access Token), update the permissions of your PAT to allow the controller to list all the repositories under your organization.
+
+An example of scaling your organization runners is shown below:
+
+```yaml
+apiVersion: actions.summerwind.dev/v1alpha1
+kind: RunnerDeployment
+metadata:
+  name: example-runnerdeploy
+spec:
+  template:
+    spec:
+      organization: "your-organization-name"
+---
+apiVersion: actions.summerwind.dev/v1alpha1
+kind: HorizontalRunnerAutoscaler
+metadata:
+  name: example-runnerdeploy-autoscaler
+spec:
+  scaleTargetRef:
+    name: example-runnerdeploy
+  minReplicas: 1
+  maxReplicas: 3
+  metrics:
+  - type: TotalNumberOfQueuedAndInProgressWorkflowRuns
+```
+
 ## Runner with DinD
 
 When using default runner, runner pod starts up 2 containers: runner and DinD (Docker-in-Docker). This might create issues if there's `LimitRange` set to namespace.
@@ -321,7 +355,7 @@ spec:
         requests:
           cpu: "2.0"
           memory: "4Gi"
-      # If set to false, there are no privileged container and you cannot use docker. 
+      # If set to false, there are no privileged container and you cannot use docker.
       dockerEnabled: false
       # If set to true, runner pod container only 1 container that's expected to be able to run docker, too.
       # image summerwind/actions-runner-dind or custom one should be used with true -value

controllers/autoscaling.go

@@ -2,9 +2,10 @@ package controllers
 
 import (
 	"context"
-	"errors"
 	"fmt"
+	"log"
 	"strings"
+	"time"
 
 	"github.com/summerwind/actions-runner-controller/api/v1alpha1"
 )
@@ -18,30 +19,61 @@ func (r *HorizontalRunnerAutoscalerReconciler) determineDesiredReplicas(rd v1alp
 
 	var repos [][]string
 
-	repoID := rd.Spec.Template.Spec.Repository
-	if repoID == "" {
-		orgName := rd.Spec.Template.Spec.Organization
-		if orgName == "" {
-			return nil, fmt.Errorf("asserting runner deployment spec to detect bug: spec.template.organization should not be empty on this code path")
-		}
+	orgName := rd.Spec.Template.Spec.Organization
+	if orgName == "" {
+		return nil, fmt.Errorf("asserting runner deployment spec to detect bug: spec.template.organization should not be empty on this code path")
+	}
 
-		metrics := hra.Spec.Metrics
+	metrics := hra.Spec.Metrics
+	if len(metrics) == 0 {
+		return nil, fmt.Errorf("validating autoscaling metrics: one or more metrics is required")
+	} else if tpe := metrics[0].Type; tpe != v1alpha1.AutoscalingMetricTypeTotalNumberOfQueuedAndInProgressWorkflowRuns {
+		return nil, fmt.Errorf("validting autoscaling metrics: unsupported metric type %q: only supported value is %s", tpe, v1alpha1.AutoscalingMetricTypeTotalNumberOfQueuedAndInProgressWorkflowRuns)
+	}
+
+	// Github Enterprise API doesn't offer an endpoint for checking the Actions queue on org level, neither does it offer an endpoint for checking what repos Actions is enabled. This means that, in case of GH Enterprise, you must pass a slice of repositories.
+	if len(metrics[0].RepositoryNames) < 1 && r.GitHubClient.GithubEnterprise {
+		return nil, fmt.Errorf("[ERROR] user didn't pass any repository! Please pass a list of repositories the controller has to monitor")
+	}
 
-		if len(metrics) == 0 {
-			return nil, fmt.Errorf("validating autoscaling metrics: one or more metrics is required")
-		} else if tpe := metrics[0].Type; tpe != v1alpha1.AutoscalingMetricTypeTotalNumberOfQueuedAndInProgressWorkflowRuns {
-			return nil, fmt.Errorf("validting autoscaling metrics: unsupported metric type %q: only supported value is %s", tpe, v1alpha1.AutoscalingMetricTypeTotalNumberOfQueuedAndInProgressWorkflowRuns)
-		} else if len(metrics[0].RepositoryNames) == 0 {
-			return nil, errors.New("validating autoscaling metrics: spec.autoscaling.metrics[].repositoryNames is required and must have one more more entries for organizational runner deployment")
+	// If the above conditionally didn't return an error, we  automatically assume, in case on an empty repository slice, that an organization is used.
+	if len(metrics[0].RepositoryNames) == 0 {
+		enabledRepos, _, err := r.GitHubClient.Actions.ListEnabledReposInOrg(context.Background(), orgName, nil)
+		if err != nil {
+			return nil, fmt.Errorf("[ERROR] 'ListEnabledReposInOrg' failed with error message: %s", err)
 		}
 
-		for _, repoName := range metrics[0].RepositoryNames {
-			repos = append(repos, []string{orgName, repoName})
+		if len(enabledRepos.Repositories) < 1 {
+			return nil, fmt.Errorf("[ERROR] 'ListEnabledReposInOrg' returned an empty slice of repositories, check your permissions. Error message: %s", err)
 		}
-	} else {
-		repo := strings.Split(repoID, "/")
 
-		repos = append(repos, repo)
+		for _, v := range enabledRepos.Repositories {
+			repoName := fmt.Sprint(*v.Name)
+
+			if *v.Archived || *v.Disabled {
+				continue
+			}
+
+			lastChange := (int(time.Now().UTC().Sub(v.PushedAt.Time).Minutes()))
+			// We need a conditional here, since the `ListEnabledReposInOrg(ListOptions)` doesn't allow us to filter on `pushedAt`: https://docs.github.com/en/free-pro-team@latest/rest/reference/actions#list-selected-repositories-enabled-for-github-actions-in-an-organization
+			if lastChange > int(r.SyncPeriod.Minutes()) {
+				continue
+			} else if len(repos) < 20 {
+				repos = append(repos, []string{orgName, repoName})
+			}
+		}
+		log.Printf("[INFO] watching the following organizational repositories: %s", repos)
+
+	} else {
+		repoID := rd.Spec.Template.Spec.Repository
+		if repoID == "" {
+			for _, repoName := range metrics[0].RepositoryNames {
+				repos = append(repos, []string{orgName, repoName})
+			}
+		} else {
+			repo := strings.Split(repoID, "/")
+			repos = append(repos, repo)
+		}
 	}
 
 	var total, inProgress, queued, completed, unknown int

controllers/horizontalrunnerautoscaler_controller.go

@@ -46,6 +46,7 @@ type HorizontalRunnerAutoscalerReconciler struct {
 	Log          logr.Logger
 	Recorder     record.EventRecorder
 	Scheme       *runtime.Scheme
+	SyncPeriod   *time.Duration
 }
 
 // +kubebuilder:rbac:groups=actions.summerwind.dev,resources=runnerdeployments,verbs=get;list;watch;update;patch

github/github.go

@@ -29,16 +29,19 @@ type Client struct {
 	regTokens map[string]*github.RegistrationToken
 	mu        sync.Mutex
 	// GithubBaseURL to Github without API suffix.
-	GithubBaseURL string
+	GithubBaseURL    string
+	GithubEnterprise bool
 }
 
 // NewClient creates a Github Client
 func (c *Config) NewClient() (*Client, error) {
 	var (
-		httpClient *http.Client
-		client     *github.Client
+		httpClient       *http.Client
+		client           *github.Client
+		githubEnterprise bool
 	)
 	githubBaseURL := "https://github.com/"
+
 	if len(c.Token) > 0 {
 		httpClient = oauth2.NewClient(context.Background(), oauth2.StaticTokenSource(
 			&oauth2.Token{AccessToken: c.Token},
@@ -60,20 +63,24 @@ func (c *Config) NewClient() (*Client, error) {
 
 	if len(c.EnterpriseURL) > 0 {
 		var err error
+		githubEnterprise = true
+
 		client, err = github.NewEnterpriseClient(c.EnterpriseURL, c.EnterpriseURL, httpClient)
 		if err != nil {
 			return nil, fmt.Errorf("enterprise client creation failed: %v", err)
 		}
 		githubBaseURL = fmt.Sprintf("%s://%s%s", client.BaseURL.Scheme, client.BaseURL.Host, strings.TrimSuffix(client.BaseURL.Path, "api/v3/"))
 	} else {
 		client = github.NewClient(httpClient)
+		githubEnterprise = false
 	}
 
 	return &Client{
-		Client:        client,
-		regTokens:     map[string]*github.RegistrationToken{},
-		mu:            sync.Mutex{},
-		GithubBaseURL: githubBaseURL,
+		Client:           client,
+		regTokens:        map[string]*github.RegistrationToken{},
+		mu:               sync.Mutex{},
+		GithubBaseURL:    githubBaseURL,
+		GithubEnterprise: githubEnterprise,
 	}, nil
 }
 

go.mod

@@ -6,8 +6,6 @@ require (
 	github.com/bradleyfalzon/ghinstallation v1.1.1
 	github.com/davecgh/go-spew v1.1.1
 	github.com/go-logr/logr v0.1.0
-	github.com/google/go-github v17.0.0+incompatible // indirect
-	github.com/google/go-github/v32 v32.1.1-0.20200822031813-d57a3a84ba04
 	github.com/google/go-github/v33 v33.0.0
 	github.com/google/go-querystring v1.0.0
 	github.com/gorilla/mux v1.8.0

go.sum

@@ -116,12 +116,8 @@ github.com/google/go-cmp v0.3.0 h1:crn/baboCvb5fXaQ0IJ1SGTsTVrWpDsCWC8EGETZijY=
 github.com/google/go-cmp v0.3.0/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU=
 github.com/google/go-cmp v0.3.1 h1:Xye71clBPdm5HgqGwUkwhbynsUJZhDbS20FvLhQ2izg=
 github.com/google/go-cmp v0.3.1/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU=
-github.com/google/go-github v17.0.0+incompatible h1:N0LgJ1j65A7kfXrZnUDaYCs/Sf4rEjNlfyDHW9dolSY=
-github.com/google/go-github v17.0.0+incompatible/go.mod h1:zLgOLi98H3fifZn+44m+umXrS52loVEgC2AApnigrVQ=
 github.com/google/go-github/v29 v29.0.2 h1:opYN6Wc7DOz7Ku3Oh4l7prmkOMwEcQxpFtxdU8N8Pts=
 github.com/google/go-github/v29 v29.0.2/go.mod h1:CHKiKKPHJ0REzfwc14QMklvtHwCveD0PxlMjLlzAM5E=
-github.com/google/go-github/v32 v32.1.1-0.20200822031813-d57a3a84ba04 h1:wEYk2h/GwOhImcVjiTIceP88WxVbXw2F+ARYUQMEsfg=
-github.com/google/go-github/v32 v32.1.1-0.20200822031813-d57a3a84ba04/go.mod h1:rIEpZD9CTDQwDK9GDrtMTycQNA4JU3qBsCizh3q2WCI=
 github.com/google/go-github/v33 v33.0.0 h1:qAf9yP0qc54ufQxzwv+u9H0tiVOnPJxo0lI/JXqw3ZM=
 github.com/google/go-github/v33 v33.0.0/go.mod h1:GMdDnVZY/2TsWgp/lkYnpSAh6TrzhANBBwm6k6TTEXg=
 github.com/google/go-querystring v1.0.0 h1:Xkwi/a1rcvNg1PPYe5vI8GbeBY/jrVuDX5ASuANWTrk=

main.go

@@ -148,6 +148,7 @@ func main() {
 		Log:          ctrl.Log.WithName("controllers").WithName("HorizontalRunnerAutoscaler"),
 		Scheme:       mgr.GetScheme(),
 		GitHubClient: ghClient,
+		SyncPeriod:   &syncPeriod,
 	}
 
 	if err = horizontalRunnerAutoscaler.SetupWithManager(mgr); err != nil {