feat: rollback windows. Fixes argoproj#574 (argoproj#2394)

* feature: introduce rollback windows

fixes: argoproj#574
Signed-off-by: Alex Eftimie <[email protected]>

* feature: introduce rollback windows - generated files

fixes: argoproj#574
Signed-off-by: Alex Eftimie <[email protected]>

* ran codegen again

Signed-off-by: Alex Eftimie <[email protected]>

* More unit tests. New e2e

Signed-off-by: Alex Eftimie <[email protected]>

* More tests to make codecov happy

Signed-off-by: Alex Eftimie <[email protected]>

* increas lint timeout

Signed-off-by: Alex Eftimie <[email protected]>

* Exclude Experiment RS when computing rollback window

Signed-off-by: Alex Eftimie <[email protected]>

* Add documentation around new feature rollbackWindow

Signed-off-by: Alex Eftimie <[email protected]>

* Fix rollback window; cancel pauses and abort and skip to the end of analysis when the window is detected

Signed-off-by: Alex Eftimie <[email protected]>

Signed-off-by: Alex Eftimie <[email protected]>

Author: alexef

.github/workflows/go.yml

@@ -45,7 +45,7 @@ jobs:
         uses: golangci/golangci-lint-action@v3
         with:
           version: v1.49.0
-          args: --timeout 5m
+          args: --timeout 6m
   build:
     name: Build
     runs-on: ubuntu-latest

docs/features/specification.md

@@ -78,6 +78,12 @@ spec:
   # than or equal to this value.
   restartAt: "2020-03-30T21:19:35Z"
 
+  # The rollback window provides a way to fast track deployments to
+  # previously deployed versions.
+  # Optional, and by default is not set.
+  rollbackWindow:
+    revisions: 3
+
   strategy:
 
     # Blue-green update strategy

docs/rollback.md

@@ -0,0 +1,23 @@
+# Rollback Windows
+
+!!! important
+
+    Available for blue-green and canary rollouts since v1.4
+
+By default, when an older Rollout manifest is re-applied, the controller treats it the same as a spec change, and will execute the full list of steps, and perform analysis too. There are two exceptions to this rule: 
+1. the controller detects if it is moving back to a blue-green ReplicaSet which exists and is still scaled up (within its `scaleDownDelay`) 
+2. the controller detects it is moving back to the canary's "stable" ReplicaSet, and the upgrade had not yet completed.
+
+It is often undesirable to re-run analysis and steps for a rollout, when the desired behavior is to rollback as soon as possible. To help with this, a rollback window feature allows users to indicate that the promotion to the ReplicaSet within the window will skip all steps.
+
+Example:
+
+```yaml
+spec:
+  rollbackWindow:
+    revisions: 3
+
+  revisionHistoryLimit: 5
+```
+
+Assume a linear revision history: `1`, `2`, `3`, `4`, `5 (current)`. A rollback from revision 5 back to 4 or 3 will fall within the window, so it will be fast tracked.

manifests/crds/rollout-crd.yaml

@@ -79,6 +79,12 @@ spec:
               revisionHistoryLimit:
                 format: int32
                 type: integer
+              rollbackWindow:
+                properties:
+                  revisions:
+                    format: int32
+                    type: integer
+                type: object
               selector:
                 properties:
                   matchExpressions:

manifests/install.yaml

@@ -11089,6 +11089,12 @@ spec:
               revisionHistoryLimit:
                 format: int32
                 type: integer
+              rollbackWindow:
+                properties:
+                  revisions:
+                    format: int32
+                    type: integer
+                type: object
               selector:
                 properties:
                   matchExpressions:

manifests/namespace-install.yaml

@@ -11089,6 +11089,12 @@ spec:
               revisionHistoryLimit:
                 format: int32
                 type: integer
+              rollbackWindow:
+                properties:
+                  revisions:
+                    format: int32
+                    type: integer
+                type: object
               selector:
                 properties:
                   matchExpressions:

mkdocs.yml

@@ -37,6 +37,7 @@ nav:
   - Ephemeral Metadata: features/ephemeral-metadata.md
   - Restarting Rollouts: features/restart.md
   - Scaledown Aborted Rollouts: features/scaledown-aborted-rs.md
+  - Rollback Window: features/rollback.md
   - Anti Affinity: features/anti-affinity/anti-affinity.md
   - Helm: features/helm.md
   - Kustomize: features/kustomize.md

pkg/apiclient/rollout/rollout.swagger.json

@@ -1129,6 +1129,15 @@
       "type": "object",
       "title": "RequiredDuringSchedulingIgnoredDuringExecution defines inter-pod scheduling rule to be RequiredDuringSchedulingIgnoredDuringExecution"
     },
+    "github.com.argoproj.argo_rollouts.pkg.apis.rollouts.v1alpha1.RollbackWindowSpec": {
+      "type": "object",
+      "properties": {
+        "revisions": {
+          "type": "integer",
+          "format": "int32"
+        }
+      }
+    },
     "github.com.argoproj.argo_rollouts.pkg.apis.rollouts.v1alpha1.Rollout": {
       "type": "object",
       "properties": {
@@ -1368,6 +1377,10 @@
           "format": "int32",
           "title": "Minimum number of seconds for which a newly created pod should be ready\nwithout any of its container crashing, for it to be considered available.\nDefaults to 0 (pod will be considered available as soon as it is ready)\n+optional"
         },
+        "rollbackWindow": {
+          "$ref": "#/definitions/github.com.argoproj.argo_rollouts.pkg.apis.rollouts.v1alpha1.RollbackWindowSpec",
+          "title": "The window in which a rollback will be fast tracked (fully promoted)\n+optional"
+        },
         "strategy": {
           "$ref": "#/definitions/github.com.argoproj.argo_rollouts.pkg.apis.rollouts.v1alpha1.RolloutStrategy",
           "title": "The deployment strategy to use to replace existing pods with new ones.\n+optional"