release: support retrying rollbacks

This commit adds support for retrying rollbacks using
`.spec.rollback.retry` until the configured `.spec.rolback.maxRetries`
count is reached (defaults to 5, setting it to 0 means infinite). The
reset of the counter happens after a successful upgrade or a change
(e.g. chart change, `.spec` change).

This makes it possible to for example recover from transient errors
that were the cause of the rollback, and not a persistent chart or
values failure.

chart/helm-operator/templates/crd.yaml

@@ -84,6 +84,15 @@ spec:
                 enable:
                   description: If set, will perform rollbacks for this release on upgrade failures
                   type: boolean
+                retry:
+                  description: If set, the upgrade of a rolled back release will be retried until
+                    the maximum amount of retries is reached
+                  type: boolean
+                maxRetries:
+                  description: The maximum amount of retries that should be attempted for a rolled
+                    back release if retries are enabled, defaults to 5, 0 equals infinite
+                  type: integer
+                  format: int64
                 force:
                   description: If set, will force resource update through delete/recreate if needed
                   type: boolean

deploy/flux-helm-release-crd.yaml

@@ -76,6 +76,15 @@ spec:
                 enable:
                   description: If set, will perform rollbacks for this release on upgrade failures
                   type: boolean
+                retry:
+                  description: If set, the upgrade of a rolled back release will be retried until
+                    the maximum amount of retries is reached
+                  type: boolean
+                maxRetries:
+                  description: The maximum amount of retries that should be attempted for a rolled
+                    back release if retries are enabled, defaults to 5, 0 equals infinite
+                  type: integer
+                  format: int64
                 force:
                   description: If set, will force resource update through delete/recreate if needed
                   type: boolean

pkg/apis/helm.fluxcd.io/v1/types.go

@@ -164,6 +164,8 @@ func (s RepoChartSource) CleanRepoURL() string {
 
 type Rollback struct {
 	Enable       bool   `json:"enable,omitempty"`
+	Retry        bool   `json:"retry,omitempty"`
+	MaxRetries   *int64 `json:"maxRetries,omitempty"`
 	Force        bool   `json:"force,omitempty"`
 	Recreate     bool   `json:"recreate,omitempty"`
 	DisableHooks bool   `json:"disableHooks,omitempty"`
@@ -178,6 +180,13 @@ func (r Rollback) GetTimeout() time.Duration {
 	return time.Duration(*r.Timeout) * time.Second
 }
 
+func (r Rollback) GetMaxRetries() int64 {
+	if r.MaxRetries == nil {
+		return 5
+	}
+	return *r.MaxRetries
+}
+
 // HelmReleaseSpec is the spec for a HelmRelease resource
 type HelmReleaseSpec struct {
 	ChartSource      `json:"chart"`
@@ -268,6 +277,11 @@ type HelmReleaseStatus struct {
 	// +optional
 	Revision string `json:"revision,omitempty"`
 
+	// RollbackCount defines the amount of rollback attempts made,
+	// it is incremented after a rollback failure and reset after a
+	// successful upgrade or revision change.
+	RollbackCount int64 `json:"retryCount,omitempty"`
+
 	// Conditions contains observations of the resource's state, e.g.,
 	// has the chart which it refers to been fetched.
 	// +optional

pkg/install/templates/flux-helm-release-crd.yaml.tmpl

@@ -76,6 +76,15 @@ spec:
                 enable:
                   description: If set, will perform rollbacks for this release on upgrade failures
                   type: boolean
+                retry:
+                  description: If set, the upgrade of a rolled back release will be retried until
+                    the maximum amount of retries is reached
+                  type: boolean
+                maxRetries:
+                  description: The maximum amount of retries that should be attempted for a rolled
+                    back release if retries are enabled, defaults to 5, 0 equals infinite
+                  type: integer
+                  format: int64
                 force:
                   description: If set, will force resource update through delete/recreate if needed
                   type: boolean