sg: cloud handle deploying a version (#62447)

* wait for build

* make not found error more descriptive

* bazel

* fixup

* fix list instances

* fix deployment key

* fixup

* handle deploying a version

* fix docker image filtering from registry

- make filtering more efficient by filtering while we're iterating
- match docker image format

* merge fixup

* fix formatting

dev/sg/internal/bk/bk.go

@@ -121,7 +121,7 @@ func (c *Client) GetBuildByNumber(ctx context.Context, pipeline string, number s
 	b, _, err := c.bk.Builds.Get(BuildkiteOrg, pipeline, number, nil)
 	if err != nil {
 		if strings.Contains(err.Error(), "404 Not Found") {
-			return nil, errors.New("no build found")
+			return nil, errors.Newf("build %s not found on pipeline %q", number, pipeline)
 		}
 		return nil, err
 	}

dev/sg/internal/cloud/BUILD.bazel

@@ -4,6 +4,7 @@ load("@io_bazel_rules_go//go:def.bzl", "go_library")
 go_library(
     name = "cloud",
     srcs = [
+        "artifact_registry.go",
         "client.go",
         "common.go",
         "delete_command.go",

dev/sg/internal/cloud/artifact_registry.go

@@ -0,0 +1,159 @@
+package cloud
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/grafana/regexp"
+
+	artifactregistry "cloud.google.com/go/artifactregistry/apiv1"
+	artifactregistrypb "cloud.google.com/go/artifactregistry/apiv1/artifactregistrypb"
+	"google.golang.org/api/iterator"
+)
+
+const DefaultArtifactRegistryPageSize = 10000
+
+// DockerImage is a type alias around the Google artifact registry Docker Image type
+type DockerImage artifactregistrypb.DockerImage
+
+type DockerImageFilterOpt func(image *DockerImage) bool
+
+// FilterTagByRegex filters the Docker Image tags by the given regular expression
+func FilterTagByRegex(regex *regexp.Regexp) DockerImageFilterOpt {
+	return func(image *DockerImage) bool {
+		for _, tag := range image.Tags {
+			if regex.MatchString(tag) {
+				return true
+			}
+		}
+		return false
+	}
+}
+
+// FilterNameByRegex filters the Docker Image names by the given regular expression
+func FilterNameByRegex(regex *regexp.Regexp) DockerImageFilterOpt {
+	return func(image *DockerImage) bool {
+		return regex.MatchString(image.Name)
+	}
+}
+
+// FilterByName filters the Docker Images by name
+func FilterByName(name string) DockerImageFilterOpt {
+	// the name format is projects/*/locations/*/repositories/*/dockerImages/<name>@sha256:digest
+	regexp := regexp.MustCompile(fmt.Sprintf("/%s@", name))
+	return FilterNameByRegex(regexp)
+}
+
+// FilterByTag filters the Docker Images by tag
+func FilterByTag(tag string) DockerImageFilterOpt {
+	return func(image *DockerImage) bool {
+		for _, imageTag := range image.Tags {
+			if imageTag == tag {
+				return true
+			}
+		}
+		return false
+	}
+}
+
+// ArtifactRegistry is wrapper around the Google Artifact Registry client
+type ArtifactRegistry struct {
+	Project        string
+	Location       string
+	RepositoryName string
+	PageSize       int32
+
+	client *artifactregistry.Client
+}
+
+// NewDefaultCloudEphemeralRegistry creates an Artifact Registry with all the details set to point to the cloud
+// ephemeral registry
+func NewDefaultCloudEphemeralRegistry(ctx context.Context) (*ArtifactRegistry, error) {
+	return NewArtifactRegistry(ctx, "sourcegraph-ci", "us-central1", "cloud-ephemeral")
+}
+
+func NewArtifactRegistry(ctx context.Context, project, location, repositoryName string) (*ArtifactRegistry, error) {
+	client, err := artifactregistry.NewClient(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	return &ArtifactRegistry{
+		Project:        project,
+		Location:       location,
+		RepositoryName: repositoryName,
+		PageSize:       DefaultArtifactRegistryPageSize,
+
+		client: client,
+	}, nil
+}
+
+// Parent returns the parent string in the format of projects/<project>/locations/<location>/repositories/<repositoryName>
+func (ar *ArtifactRegistry) Parent() string {
+	return fmt.Sprintf("projects/%s/locations/%s/repositories/%s", ar.Project, ar.Location, ar.RepositoryName)
+}
+
+func (ar *ArtifactRegistry) String() string {
+	return ar.Parent()
+}
+
+// ListVersions lists all Docker Images present in the Artifact Registry and optionally filter the images as the images are iterated upon.
+func (ar *ArtifactRegistry) ListDockerImages(ctx context.Context, filterOpts ...DockerImageFilterOpt) ([]*DockerImage, error) {
+	req := &artifactregistrypb.ListDockerImagesRequest{
+		Parent:   ar.Parent(),
+		PageSize: ar.PageSize,
+		OrderBy:  "upload_time",
+	}
+
+	images := []*DockerImage{}
+	resp := ar.client.ListDockerImages(ctx, req)
+	for {
+		image, err := resp.Next()
+		if err != nil {
+			if err == iterator.Done {
+				break
+			}
+			return nil, err
+		}
+
+		dockerImage := (*DockerImage)(image)
+		var shouldAdd bool
+		for _, filterOpt := range filterOpts {
+			shouldAdd = filterOpt(dockerImage)
+			if !shouldAdd {
+				break
+			}
+		}
+		if shouldAdd {
+			images = append(images, dockerImage)
+		}
+	}
+
+	return images, nil
+}
+
+// FindDockerImageExact finds Docker Images that match the name and tag exactly in the Artifact Registry.
+func (ar *ArtifactRegistry) FindDockerImageExact(ctx context.Context, name string, tag string) ([]*DockerImage, error) {
+	return ar.ListDockerImages(ctx, FilterByName(name), FilterByTag(tag))
+}
+
+// FindDockerImageByTagPattern finds all Docker Images that have a tag that matches the given tag pattern.
+func (ar *ArtifactRegistry) FindDockerImageByTagPattern(ctx context.Context, tagPattern string) ([]*DockerImage, error) {
+	tagRegex := regexp.MustCompile(tagPattern)
+	return ar.ListDockerImages(ctx, FilterTagByRegex(tagRegex))
+}
+
+// GetDockerImage gets a Docker Image by name and digest from the Artifact Registry.
+func (ar *ArtifactRegistry) GetDockerImage(ctx context.Context, name, digest string) (*DockerImage, error) {
+	name = fmt.Sprintf("%s/dockerImages/%s@sha256:%s", ar.Parent(), name, digest)
+	req := &artifactregistrypb.GetDockerImageRequest{
+		Name: name,
+	}
+
+	image, err := ar.client.GetDockerImage(ctx, req)
+	if err != nil {
+		return nil, err
+	}
+
+	return (*DockerImage)(image), nil
+}

dev/sg/internal/cloud/client.go

@@ -17,6 +17,8 @@ import (
 	"github.com/sourcegraph/sourcegraph/lib/pointers"
 )
 
+var ErrInstanceNotFound error = errors.New("instance not found")
+
 // HeaderUserToken is the header name for the user token when communicating with the Cloud API.
 const HeaderUserToken = "X-GCP-User-Token"
 

dev/sg/internal/cloud/deploy_command.go

@@ -1,8 +1,8 @@
 package cloud
 
 import (
-	"cmp"
 	"context"
+	"strings"
 	"time"
 
 	"github.com/buildkite/go-buildkite/v3/buildkite"
@@ -18,6 +18,8 @@ import (
 	"github.com/sourcegraph/sourcegraph/lib/pointers"
 )
 
+var ErrDeploymentExists error = errors.New("deployment already exists")
+
 var DeployEphemeralCommand = cli.Command{
 	Name:        "deploy",
 	Usage:       "sg could deploy --branch <branch> --tag <tag>",
@@ -32,14 +34,6 @@ var DeployEphemeralCommand = cli.Command{
 			Name:        "version",
 			DefaultText: "deploys an ephemeral cloud Sourcegraph environment with the specified version. The version MUST exist and implies that no build will be created",
 		},
-		&cli.StringFlag{
-			Name:        "tag",
-			DefaultText: "the git tag that should be deployed",
-		},
-		&cli.BoolFlag{
-			Name:        "skip-wip-notice",
-			DefaultText: "skips the EXPERIMENTAL notice prompt",
-		},
 	},
 }
 
@@ -65,6 +59,42 @@ func determineVersion(build *buildkite.Build, tag string) (string, error) {
 	), nil
 }
 
+func createDeploymentForVersion(ctx context.Context, email, name, version string) error {
+	cloudClient, err := NewClient(ctx, email, APIEndpoint)
+	if err != nil {
+		return err
+	}
+
+	cloudEmoji := "☁️"
+	pending := std.Out.Pending(output.Linef(cloudEmoji, output.StylePending, "Starting deployment %q for version %q", name, version))
+	spec := NewDeploymentSpec(
+		sanitizeInstanceName(name),
+		version,
+	)
+
+	// Check if the deployment already exists
+	_, err = cloudClient.GetInstance(ctx, spec.Name)
+	if err != nil {
+		if !errors.Is(err, ErrInstanceNotFound) {
+			return errors.Wrapf(err, "failed to determine if instance %q already exists", spec.Name)
+		}
+	} else {
+		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "Cannot deploy %q", err))
+		// Deployment exists
+		return ErrDeploymentExists
+	}
+
+	inst, err := cloudClient.CreateInstance(ctx, spec)
+	if err != nil {
+		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "Deployment failed: %v", err))
+		return errors.Wrapf(err, "failed to deploy version %v", version)
+	}
+
+	pending.Writef("Deploy instance details: \n%s", inst.String())
+	pending.Complete(output.Linef(output.EmojiSuccess, output.StyleSuccess, "Deployment %q created for version %q - access at: %s", name, version, inst.URL))
+	return nil
+}
+
 func triggerEphemeralBuild(ctx context.Context, currRepo *repo.GitRepo) (*buildkite.Build, error) {
 	pending := std.Out.Pending(output.Linef("🔨", output.StylePending, "Checking if branch %q is up to date with remote", currRepo.Branch))
 	if isOutOfSync, err := currRepo.IsOutOfSync(ctx); err != nil {
@@ -76,9 +106,9 @@ func triggerEphemeralBuild(ctx context.Context, currRepo *repo.GitRepo) (*buildk
 
 	client, err := bk.NewClient(ctx, std.Out)
 	if err != nil {
-		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "failed to create client to trigger build"))
 		return nil, err
 	}
+
 	pending.Updatef("Starting cloud ephemeral build for %q on commit %q", currRepo.Branch, currRepo.Ref)
 	build, err := client.TriggerBuild(ctx, "sourcegraph", currRepo.Branch, currRepo.Ref, bk.WithEnvVar("CLOUD_EPHEMERAL", "true"))
 	if err != nil {
@@ -90,34 +120,21 @@ func triggerEphemeralBuild(ctx context.Context, currRepo *repo.GitRepo) (*buildk
 	return build, nil
 }
 
-func createDeploymentForVersion(ctx context.Context, name, version string) error {
-	email, err := GetGCloudAccount(ctx)
+func checkVersionExistsInRegistry(ctx context.Context, version string) error {
+	ar, err := NewDefaultCloudEphemeralRegistry(ctx)
 	if err != nil {
+		std.Out.WriteFailuref("failed to create Cloud Ephemeral registry")
 		return err
 	}
-	if err := validateEmail(email); err != nil {
+	pending := std.Out.Pending(output.Linef(CloudEmoji, output.StylePending, "Checking if version %q exists in Cloud ephemeral registry", version))
+	if images, err := ar.FindDockerImageExact(ctx, "gitserver", version); err != nil {
+		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "failed to check if version %q exists in Cloud ephemeral registry", version))
 		return err
+	} else if len(images) == 0 {
+		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "no version %q found in Cloud ephemeral registry!", version))
+		return errors.Newf("no image with tag %q found", version)
 	}
-	cloudClient, err := NewClient(ctx, email, APIEndpoint)
-	if err != nil {
-		return err
-	}
-
-	cloudEmoji := "☁️"
-	pending := std.Out.Pending(output.Linef(cloudEmoji, output.StylePending, "Creating deployment %q for version %q", name, version))
-
-	spec := NewDeploymentSpec(
-		sanitizeInstanceName(name),
-		version,
-	)
-	inst, err := cloudClient.CreateInstance(ctx, spec)
-	if err != nil {
-		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "deployment failed: %v", err))
-		return errors.Wrapf(err, "failed to deploy version %v", version)
-	}
-
-	pending.Writef("Deploy instance details: \n%s", inst.String())
-	pending.Complete(output.Linef(output.EmojiSuccess, output.StyleSuccess, "Deployment %q created for version %q - access at: %s", name, version, inst.URL))
+	pending.Complete(output.Linef(output.EmojiSuccess, output.StyleSuccess, "Version %q found in Cloud ephemeral registry", version))
 	return nil
 }
 
@@ -145,15 +162,35 @@ func deployCloudEphemeral(ctx *cli.Context) error {
 				std.Out.WriteWarningf(`Your branch %q is out of sync with remote.
 
 Please make sure you have either pushed or pulled the latest changes before trying again`, currRepo.Branch)
+			} else {
+				std.Out.WriteFailuref("Cannot start deployment as there was problem with the ephemeral build")
 			}
-			return err
+			return errors.Wrapf(err, "cloud ephemeral deployment failure")
 		}
+
 		version, err = determineVersion(build, ctx.String("tag"))
 		if err != nil {
 			return err
 		}
+	} else if err = checkVersionExistsInRegistry(ctx.Context, version); err != nil {
+		return err
+	}
+	email, err := GetGCloudAccount(ctx.Context)
+	if err != nil {
+		return err
 	}
 
-	name := cmp.Or(ctx.String("name"), currRepo.Branch)
-	return createDeploymentForVersion(ctx.Context, name, version)
+	var deploymentName string
+	if ctx.String("name") != "" {
+		deploymentName = ctx.String("name")
+	} else if ctx.String("version") != "" {
+		// if a version is given we generate a name based on the email user and the given version
+		// to make sure the deployment is unique
+		user := strings.ReplaceAll(email[0:strings.Index(email, "@")], ".", "_")
+		deploymentName = user[:min(12, len(user))] + "_" + version
+	} else {
+		deploymentName = currRepo.Branch
+	}
+
+	return createDeploymentForVersion(ctx.Context, email, deploymentName, version)
 }

dev/sg/internal/cloud/instance.go

@@ -213,7 +213,7 @@ func (f *InstanceFeatures) IsEphemeralInstance() bool {
 }
 
 func (f *InstanceFeatures) SetEphemeralInstance(v bool) {
-	f.features["ephemeral_instance"] = strconv.FormatBool(v)
+	f.features["ephemeral"] = strconv.FormatBool(v)
 }
 
 func (f *InstanceFeatures) SetEphemeralLeaseTime(expiresAt time.Time) {

dev/sg/internal/cloud/list_versions_command.go

@@ -1,34 +1,30 @@
 package cloud
 
 import (
-	"context"
 	"encoding/json"
-	"fmt"
 	"os"
 	"time"
 
-	artifactregistry "cloud.google.com/go/artifactregistry/apiv1"
-	artifactregistrypb "cloud.google.com/go/artifactregistry/apiv1/artifactregistrypb"
 	"github.com/grafana/regexp"
 	"github.com/urfave/cli/v2"
-	"google.golang.org/api/iterator"
 
 	"github.com/sourcegraph/sourcegraph/dev/sg/internal/std"
 	"github.com/sourcegraph/sourcegraph/lib/output"
 )
 
-const DefaultArtifactRegistryPageSize = 10000
-
 var ListVersionsEphemeralCommand = cli.Command{
 	Name:        "list-versions",
 	Usage:       "sg could list-versions",
 	Description: "list ephemeral cloud instances attached to your GCP account",
 	Action:      listTagsCloudEphemeral,
 	Flags: []cli.Flag{
-		&cli.StringFlag{
-			Name:  "format",
-			Usage: "format to print out the list of versions out - can be one json, raw or formatted",
-			Value: "formatted",
+		&cli.BoolFlag{
+			Name:  "json",
+			Usage: "print the instance details in JSON",
+		},
+		&cli.BoolFlag{
+			Name:  "raw",
+			Usage: "print all of the instance details",
 		},
 		&cli.IntFlag{
 			Name:  "limit",
@@ -36,87 +32,13 @@ var ListVersionsEphemeralCommand = cli.Command{
 			Value: 100,
 		},
 		&cli.StringFlag{
-			Name:  "filter",
-			Usage: "filter versions by regex",
+			Name:    "filter",
+			Usage:   "filter versions by regex",
+			Aliases: []string{"f"},
 		},
 	},
 }
 
-// DockerImage is a type alias around the Google artifact registry Docker Image type
-type DockerImage artifactregistrypb.DockerImage
-
-// ArtifactRegistry is wrapper around the Google Artifact Registry client
-type ArtifactRegistry struct {
-	Project        string
-	Location       string
-	RepositoryName string
-	PageSize       int32
-
-	client *artifactregistry.Client
-}
-
-func NewArtifactRegistry(ctx context.Context, project, location, repositoryName string) (*ArtifactRegistry, error) {
-	client, err := artifactregistry.NewClient(ctx)
-	if err != nil {
-		return nil, err
-	}
-
-	return &ArtifactRegistry{
-		Project:        project,
-		Location:       location,
-		RepositoryName: repositoryName,
-		PageSize:       DefaultArtifactRegistryPageSize,
-
-		client: client,
-	}, nil
-}
-
-// Parent returns the parent string in the format of projects/<project>/locations/<location>/repositories/<repositoryName>
-func (ar *ArtifactRegistry) Parent() string {
-	return fmt.Sprintf("projects/%s/locations/%s/repositories/%s", ar.Project, ar.Location, ar.RepositoryName)
-}
-
-func (ar *ArtifactRegistry) String() string {
-	return ar.Parent()
-}
-
-// ListVersions lists all Docker Images present in the Artifact Registry.
-func (ar *ArtifactRegistry) ListDockerImages(ctx context.Context) ([]*DockerImage, error) {
-	req := &artifactregistrypb.ListDockerImagesRequest{
-		Parent:   ar.Parent(),
-		PageSize: ar.PageSize,
-	}
-
-	images := []*DockerImage{}
-	resp := ar.client.ListDockerImages(ctx, req)
-	for {
-		image, err := resp.Next()
-		if err != nil {
-			if err == iterator.Done {
-				break
-			}
-			return nil, err
-		}
-		images = append(images, (*DockerImage)(image))
-	}
-
-	return images, nil
-}
-
-// GetDockerImage gets a Docker Image by name from the Artifact Registry.
-func (ar *ArtifactRegistry) GetDockerImage(ctx context.Context, name string) (*DockerImage, error) {
-	req := &artifactregistrypb.GetDockerImageRequest{
-		Name: name,
-	}
-
-	image, err := ar.client.GetDockerImage(ctx, req)
-	if err != nil {
-		return nil, err
-	}
-
-	return (*DockerImage)(image), nil
-}
-
 func listTagsCloudEphemeral(ctx *cli.Context) error {
 	var filterRegex *regexp.Regexp
 	if ctx.String("filter") != "" {
@@ -127,9 +49,9 @@ func listTagsCloudEphemeral(ctx *cli.Context) error {
 		return err
 	}
 	pending := std.Out.Pending(output.Linef(CloudEmoji, output.StylePending, "Retrieving docker images from registry %q", ar.RepositoryName))
-	images, err := ar.ListDockerImages(ctx.Context)
+	images, err := ar.ListDockerImages(ctx.Context, FilterTagByRegex(filterRegex))
 	if err != nil {
-		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "failed to retreive images from registry %q", ar.RepositoryName))
+		pending.Complete(output.Linef(output.EmojiFailure, output.StyleFailure, "Failed to retreive images from registry %q", ar.RepositoryName))
 		return err
 	}
 	pending.Complete(output.Linef(CloudEmoji, output.StyleSuccess, "Retrieved %d docker images from registry %q", len(images), ar.RepositoryName))
@@ -143,20 +65,20 @@ func listTagsCloudEphemeral(ctx *cli.Context) error {
 		}
 	}
 
-	switch ctx.String("format") {
-	case "json":
+	switch {
+	case ctx.Bool("json"):
 		{
 			return json.NewEncoder(os.Stdout).Encode(imagesByTag)
 		}
-	case "raw":
+	case ctx.Bool("raw"):
 		{
 			count := 0
 			limit := ctx.Int("limit")
 			for tag, images := range imagesByTag {
 				image := images[0]
-				std.Out.Writef("Tag: %s\n", tag)
-				std.Out.Writef(" %-50s %-20s %s", "Name", "Upload time", "URI")
-				std.Out.Writef("- %-50s %-20s %s", image.Name, image.UploadTime.AsTime().Format(time.DateTime), image.Uri)
+				std.Out.Writef(`Tag                   : %s
+Upload Time           : %s
+Image count           : %d`, tag, image.UploadTime.AsTime().Format(time.DateTime), len(images))
 				count++
 				if limit >= 1 && count >= limit {
 					break
@@ -171,13 +93,16 @@ func listTagsCloudEphemeral(ctx *cli.Context) error {
 			for tag, images := range imagesByTag {
 				// we use the first image to get the upload time
 				image := images[0]
-				tag := tag[:min(50, len(tag))]
+				if len(tag) > 50 {
+					tag = tag[:47] + "..."
+				}
 				std.Out.Writef("%-50s %-20s %-5d", tag, image.UploadTime.AsTime().Format(time.DateTime), len(images))
 				count++
 				if limit >= 1 && count >= limit {
 					break
 				}
 			}
+			std.Out.WriteSuggestionf("Some tags might have been truncated. To see the full tag ouput use the --raw format or filter the tags by using --filter")
 		}
 	}
 	return nil