feat: add support for Rollout kind in Application (#249)

cmd/argo-watcher/argocd/argo_status_updater.go

@@ -33,9 +33,10 @@ type ArgoStatusUpdater struct {
 	registryProxyUrl string
 	retryOptions     []retry.Option
 	mutex            MutexMap
+	acceptSuspended  bool
 }
 
-func (updater *ArgoStatusUpdater) Init(argo Argo, retryAttempts uint, retryDelay time.Duration, registryProxyUrl string) {
+func (updater *ArgoStatusUpdater) Init(argo Argo, retryAttempts uint, retryDelay time.Duration, registryProxyUrl string, acceptSuspended bool) {
 	updater.argo = argo
 	updater.registryProxyUrl = registryProxyUrl
 	updater.retryOptions = []retry.Option{
@@ -44,6 +45,7 @@ func (updater *ArgoStatusUpdater) Init(argo Argo, retryAttempts uint, retryDelay
 		retry.Delay(retryDelay),
 		retry.LastErrorOnly(true),
 	}
+	updater.acceptSuspended = acceptSuspended
 }
 
 func (updater *ArgoStatusUpdater) collectInitialAppStatus(task *models.Task) error {
@@ -52,7 +54,7 @@ func (updater *ArgoStatusUpdater) collectInitialAppStatus(task *models.Task) err
 		return err
 	}
 
-	status := application.GetRolloutStatus(task.ListImages(), updater.registryProxyUrl)
+	status := application.GetRolloutStatus(task.ListImages(), updater.registryProxyUrl, updater.acceptSuspended)
 
 	// sort images to avoid hash mismatch
 	slices.Sort(application.Status.Summary.Images)
@@ -79,7 +81,7 @@ func (updater *ArgoStatusUpdater) WaitForRollout(task models.Task) {
 	}
 
 	// get application status
-	status := application.GetRolloutStatus(task.ListImages(), updater.registryProxyUrl)
+	status := application.GetRolloutStatus(task.ListImages(), updater.registryProxyUrl, updater.acceptSuspended)
 	if status == models.ArgoRolloutAppSuccess {
 		log.Info().Str("id", task.Id).Msg("App is running on the expected version.")
 		// deployment success
@@ -174,7 +176,7 @@ func (updater *ArgoStatusUpdater) waitForApplicationDeployment(task models.Task)
 			return err
 		}
 
-		status := application.GetRolloutStatus(task.ListImages(), updater.registryProxyUrl)
+		status := application.GetRolloutStatus(task.ListImages(), updater.registryProxyUrl, updater.acceptSuspended)
 
 		switch status {
 		case models.ArgoRolloutAppDegraded:

cmd/argo-watcher/argocd/argo_status_updater_test.go

@@ -29,7 +29,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -70,7 +70,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 3, 0*time.Second, "test-registry")
+		updater.Init(*argo, 3, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -118,7 +118,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -159,7 +159,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "")
+		updater.Init(*argo, 1, 0*time.Second, "", false)
 
 		// prepare test data
 		task := models.Task{
@@ -201,7 +201,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -230,7 +230,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -259,7 +259,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -288,7 +288,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -328,7 +328,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{
@@ -372,7 +372,7 @@ func TestArgoStatusUpdaterCheck(t *testing.T) {
 
 		// argo updater
 		updater := &ArgoStatusUpdater{}
-		updater.Init(*argo, 1, 0*time.Second, "test-registry")
+		updater.Init(*argo, 1, 0*time.Second, "test-registry", false)
 
 		// prepare test data
 		task := models.Task{

cmd/argo-watcher/config/config.go

@@ -28,23 +28,24 @@ type DatabaseConfig struct {
 }
 
 type ServerConfig struct {
-	ArgoUrl           url.URL        `env:"ARGO_URL,required" json:"argo_cd_url"`
-	ArgoUrlAlias      string         `env:"ARGO_URL_ALIAS" json:"argo_cd_url_alias,omitempty"` // Used to generate App URL. Can be omitted if ArgoUrl is reachable from outside.
-	ArgoToken         string         `env:"ARGO_TOKEN,required" json:"-"`
-	ArgoApiTimeout    int64          `env:"ARGO_API_TIMEOUT" envDefault:"60" json:"argo_api_timeout"`
-	DeploymentTimeout uint           `env:"DEPLOYMENT_TIMEOUT" envDefault:"900" json:"deployment_timeout"`
-	ArgoRefreshApp    bool           `env:"ARGO_REFRESH_APP" envDefault:"true" json:"argo_refresh_app"`
-	RegistryProxyUrl  string         `env:"DOCKER_IMAGES_PROXY" json:"registry_proxy_url,omitempty"`
-	StateType         string         `env:"STATE_TYPE,required" validate:"oneof=postgres in-memory" json:"state_type"`
-	StaticFilePath    string         `env:"STATIC_FILES_PATH" envDefault:"static" json:"-"`
-	SkipTlsVerify     bool           `env:"SKIP_TLS_VERIFY" envDefault:"false" json:"skip_tls_verify"`
-	LogLevel          string         `env:"LOG_LEVEL" envDefault:"info" json:"log_level"`
-	LogFormat         string         `env:"LOG_FORMAT" envDefault:"json" json:"-"`
-	Host              string         `env:"HOST" envDefault:"0.0.0.0" json:"-"`
-	Port              string         `env:"PORT" envDefault:"8080" json:"-"`
-	DeployToken       string         `env:"ARGO_WATCHER_DEPLOY_TOKEN" json:"-"`
-	Db                DatabaseConfig `json:"db,omitempty"`
-	Keycloak          KeycloakConfig `json:"keycloak,omitempty"`
+	ArgoUrl            url.URL        `env:"ARGO_URL,required" json:"argo_cd_url"`
+	ArgoUrlAlias       string         `env:"ARGO_URL_ALIAS" json:"argo_cd_url_alias,omitempty"` // Used to generate App URL. Can be omitted if ArgoUrl is reachable from outside.
+	ArgoToken          string         `env:"ARGO_TOKEN,required" json:"-"`
+	ArgoApiTimeout     int64          `env:"ARGO_API_TIMEOUT" envDefault:"60" json:"argo_api_timeout"`
+	AcceptSuspendedApp bool           `env:"ACCEPT_SUSPENDED_APP" envDefault:"false" json:"accept_suspended_app"` // If true, we will accept "Suspended" health status as valid
+	DeploymentTimeout  uint           `env:"DEPLOYMENT_TIMEOUT" envDefault:"900" json:"deployment_timeout"`
+	ArgoRefreshApp     bool           `env:"ARGO_REFRESH_APP" envDefault:"true" json:"argo_refresh_app"`
+	RegistryProxyUrl   string         `env:"DOCKER_IMAGES_PROXY" json:"registry_proxy_url,omitempty"`
+	StateType          string         `env:"STATE_TYPE,required" validate:"oneof=postgres in-memory" json:"state_type"`
+	StaticFilePath     string         `env:"STATIC_FILES_PATH" envDefault:"static" json:"-"`
+	SkipTlsVerify      bool           `env:"SKIP_TLS_VERIFY" envDefault:"false" json:"skip_tls_verify"`
+	LogLevel           string         `env:"LOG_LEVEL" envDefault:"info" json:"log_level"`
+	LogFormat          string         `env:"LOG_FORMAT" envDefault:"json" json:"-"`
+	Host               string         `env:"HOST" envDefault:"0.0.0.0" json:"-"`
+	Port               string         `env:"PORT" envDefault:"8080" json:"-"`
+	DeployToken        string         `env:"ARGO_WATCHER_DEPLOY_TOKEN" json:"-"`
+	Db                 DatabaseConfig `json:"db,omitempty"`
+	Keycloak           KeycloakConfig `json:"keycloak,omitempty"`
 }
 
 // NewServerConfig parses the server configuration from environment variables using the envconfig package.

cmd/argo-watcher/server/server.go

@@ -69,7 +69,12 @@ func RunServer() {
 
 	// initialize argo updater
 	updater := &argocd.ArgoStatusUpdater{}
-	updater.Init(*argo, serverConfig.GetRetryAttempts(), argocd.ArgoSyncRetryDelay, serverConfig.RegistryProxyUrl)
+	updater.Init(*argo,
+		serverConfig.GetRetryAttempts(),
+		argocd.ArgoSyncRetryDelay,
+		serverConfig.RegistryProxyUrl,
+		serverConfig.AcceptSuspendedApp,
+	)
 
 	// create environment
 	env := &Env{config: serverConfig, argo: argo, metrics: metrics, updater: updater}

internal/models/argo.go

@@ -78,7 +78,7 @@ type Application struct {
 }
 
 // GetRolloutStatus calculates application rollout status depending on the expected images and proxy configuration.
-func (app *Application) GetRolloutStatus(rolloutImages []string, registryProxyUrl string) string {
+func (app *Application) GetRolloutStatus(rolloutImages []string, registryProxyUrl string, acceptSuspended bool) string {
 	// check if all the images rolled out
 	for _, image := range rolloutImages {
 		if !helpers.ImagesContains(app.Status.Summary.Images, image, registryProxyUrl) {
@@ -96,6 +96,12 @@ func (app *Application) GetRolloutStatus(rolloutImages []string, registryProxyUr
 		return ArgoRolloutAppNotSynced
 	}
 
+	// an optional check that helps when we are dealing with Rollout object that can be in a suspended state
+	// during the rollout process
+	if app.Status.Health.Status == "Suspended" && app.Status.Sync.Status == "Synced" && acceptSuspended {
+		return ArgoRolloutAppSuccess
+	}
+
 	// verify app health status
 	if app.Status.Health.Status != "Healthy" {
 		return ArgoRolloutAppNotHealthy

internal/models/argo_test.go

@@ -71,7 +71,7 @@ func TestArgoRolloutStatus(t *testing.T) {
 		images := []string{"ghcr.io/shini4i/argo-watcher:version2"}
 		registryProxyUrl := ""
 		// test status
-		assert.Equal(t, ArgoRolloutAppNotAvailable, application.GetRolloutStatus(images, registryProxyUrl))
+		assert.Equal(t, ArgoRolloutAppNotAvailable, application.GetRolloutStatus(images, registryProxyUrl, false))
 	})
 
 	t.Run("Rollout status - ArgoRolloutAppNotSynced", func(t *testing.T) {
@@ -83,7 +83,7 @@ func TestArgoRolloutStatus(t *testing.T) {
 		images := []string{"ghcr.io/shini4i/argo-watcher:version1"}
 		registryProxyUrl := ""
 		// test status
-		assert.Equal(t, ArgoRolloutAppNotSynced, application.GetRolloutStatus(images, registryProxyUrl))
+		assert.Equal(t, ArgoRolloutAppNotSynced, application.GetRolloutStatus(images, registryProxyUrl, false))
 	})
 
 	t.Run("Rollout status - ArgoRolloutAppNotHealthy", func(t *testing.T) {
@@ -96,7 +96,7 @@ func TestArgoRolloutStatus(t *testing.T) {
 		images := []string{"ghcr.io/shini4i/argo-watcher:version1"}
 		registryProxyUrl := ""
 		// test status
-		assert.Equal(t, ArgoRolloutAppNotHealthy, application.GetRolloutStatus(images, registryProxyUrl))
+		assert.Equal(t, ArgoRolloutAppNotHealthy, application.GetRolloutStatus(images, registryProxyUrl, false))
 	})
 
 	t.Run("Rollout status - ArgoRolloutAppSuccess", func(t *testing.T) {
@@ -109,7 +109,7 @@ func TestArgoRolloutStatus(t *testing.T) {
 		images := []string{"ghcr.io/shini4i/argo-watcher:version1"}
 		registryProxyUrl := ""
 		// test status
-		assert.Equal(t, ArgoRolloutAppSuccess, application.GetRolloutStatus(images, registryProxyUrl))
+		assert.Equal(t, ArgoRolloutAppSuccess, application.GetRolloutStatus(images, registryProxyUrl, false))
 	})
 
 	t.Run("Rollout status - ArgoRolloutAppDegraded", func(t *testing.T) {
@@ -122,7 +122,29 @@ func TestArgoRolloutStatus(t *testing.T) {
 		images := []string{"ghcr.io/shini4i/argo-watcher:version1"}
 		registryProxyUrl := ""
 		// test status
-		assert.Equal(t, ArgoRolloutAppDegraded, application.GetRolloutStatus(images, registryProxyUrl))
+		assert.Equal(t, ArgoRolloutAppDegraded, application.GetRolloutStatus(images, registryProxyUrl, false))
+	})
+
+	t.Run("acceptSuspended is true", func(t *testing.T) {
+		application := Application{}
+		application.Status.Health.Status = "Suspended"
+		application.Status.Sync.Status = "Synced"
+
+		status := application.GetRolloutStatus([]string{}, "", true)
+		if status != ArgoRolloutAppSuccess {
+			t.Errorf("Expected status to be %s, but got %s", ArgoRolloutAppSuccess, status)
+		}
+	})
+
+	t.Run("acceptSuspended is false", func(t *testing.T) {
+		application := Application{}
+		application.Status.Health.Status = "Suspended"
+		application.Status.Sync.Status = "Synced"
+
+		status := application.GetRolloutStatus([]string{}, "", false)
+		if status == ArgoRolloutAppSuccess {
+			t.Errorf("Expected status to not be %s", ArgoRolloutAppSuccess)
+		}
 	})
 }