Max parked nodes by using percentages

.github/workflows/ci-chart.yaml

@@ -18,7 +18,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v6
         with:
-          python-version: '3.14'
+          python-version: '3.12'
           check-latest: true
       - name: Set up chart-testing
         uses: helm/[email protected]

README.md

@@ -58,7 +58,7 @@ The following options can be used to customize the k8s-shredder controller:
 | ParkedNodeTaint                    | "shredder.ethos.adobe.net/upgrade-status=parked:NoSchedule" | Taint to apply to parked nodes in format key=value:effect                                            |
 | EnableNodeLabelDetection           | false                                                       | Controls whether to scan for nodes with specific labels and automatically park them                  |
 | NodeLabelsToDetect                 | []                                                          | List of node labels to detect. Supports both key-only and key=value formats                          |
-| MaxParkedNodes                     | 0                                                           | Maximum number of nodes that can be parked simultaneously. Set to 0 (default) for no limit.         |
+| MaxParkedNodes                     | "0"                                                         | Maximum number of nodes that can be parked simultaneously. Can be an integer (e.g., "5") or percentage (e.g., "20%"). Set to "0" (default) for no limit. |
 | ExtraParkingLabels                 | {}                                                          | (Optional) Map of extra labels to apply to nodes and pods during parking. Example: `{ "example.com/owner": "infrastructure" }` |
 | EvictionSafetyCheck                | true                                                        | Controls whether to perform safety checks before force eviction. If true, nodes will be unparked if pods don't have required parking labels. |
 | ParkingReasonLabel                 | "shredder.ethos.adobe.net/parked-reason"                   | Label used to track why a node or pod was parked (values: node-label, karpenter-drifted, karpenter-disrupted) |
@@ -131,25 +131,49 @@ This integration allows k8s-shredder to automatically handle node lifecycle mana
 
 k8s-shredder supports limiting the maximum number of nodes that can be parked simultaneously using the `MaxParkedNodes` configuration option. This feature helps prevent overwhelming the cluster with too many parked nodes at once, which could impact application availability.
 
-When `MaxParkedNodes` is set to a positive integer:
+`MaxParkedNodes` can be specified as either:
+- **Integer value** (e.g., `"5"`): Absolute maximum number of nodes that can be parked
+- **Percentage value** (e.g., `"20%"`): Maximum percentage of total cluster nodes that can be parked (calculated dynamically each cycle)
 
-1. **Before parking nodes**: k8s-shredder counts the number of currently parked nodes
-2. **Calculate available slots**: `availableSlots = MaxParkedNodes - currentlyParked`
-3. **Limit parking**: If the number of eligible nodes exceeds available slots, only the first `availableSlots` nodes are parked
-4. **Skip if full**: If no slots are available (currentlyParked >= MaxParkedNodes), parking is skipped for that eviction interval
+When `MaxParkedNodes` is set to a non-zero value:
 
-**Examples:**
-- `MaxParkedNodes: 0` (default): No limit, all eligible nodes are parked
-- `MaxParkedNodes: 5`: Maximum 5 nodes can be parked at any time
-- `MaxParkedNodes: -1`: Invalid value, treated as 0 (no limit) with a warning logged
+1. **Parse the limit**: The configuration is parsed to determine the actual limit
+   - For percentages: `limit = (percentage / 100) * totalNodes` (rounded down)
+   - For integers: `limit = configured value`
+2. **Count parked nodes**: k8s-shredder counts the number of currently parked nodes
+3. **Calculate available slots**: `availableSlots = limit - currentlyParked`
+4. **Sort by age**: Eligible nodes are sorted by creation timestamp (oldest first) to ensure predictable parking order
+5. **Limit parking**: If the number of eligible nodes exceeds available slots, only the oldest `availableSlots` nodes are parked
+6. **Skip if full**: If no slots are available (currentlyParked >= limit), parking is skipped for that eviction interval
 
-This limit applies to both Karpenter drift detection and node label detection features. When multiple nodes are eligible for parking but the limit would be exceeded, k8s-shredder will park the nodes in the order they are discovered and skip the remaining nodes until the next eviction interval.
+**Examples:**
+- `MaxParkedNodes: "0"` (default): No limit, all eligible nodes are parked
+- `MaxParkedNodes: "5"`: Maximum 5 nodes can be parked at any time
+- `MaxParkedNodes: "20%"`: Maximum 20% of total cluster nodes can be parked (e.g., 2 nodes in a 10-node cluster)
+- Invalid values (e.g., `"-1"`, `"invalid"`): Treated as 0 (no limit) with a warning logged
+
+**Percentage Benefits:**
+- **Dynamic scaling**: Limit automatically adjusts as cluster size changes
+- **Proportional safety**: Maintains a consistent percentage of available capacity regardless of cluster size
+- **Auto-scaling friendly**: Works well with cluster auto-scaling by recalculating limits each cycle
+
+**Predictable Parking Order:**
+Eligible nodes are **always sorted by creation timestamp (oldest first)**, regardless of whether `MaxParkedNodes` is set. This ensures:
+- **Consistent behavior**: The same nodes will be parked first across multiple eviction cycles
+- **Fair rotation**: Oldest nodes are prioritized for replacement during rolling upgrades
+- **Predictable capacity**: You can anticipate which nodes will be parked next when slots become available
+- **Deterministic ordering**: Even when parking all eligible nodes (no limit), they are processed in a predictable order
+
+This sorting behavior applies to both Karpenter drift detection and node label detection features. When multiple nodes are eligible for parking:
+- **With no limit** (`MaxParkedNodes: "0"`): All nodes are parked in order from oldest to newest
+- **With a limit**: Only the oldest nodes up to the limit are parked; newer nodes wait for the next eviction interval
 
 **Use cases:**
 - **Gradual node replacement**: Control the pace of node cycling during cluster upgrades
 - **Resource management**: Prevent excessive resource pressure from too many parked nodes
 - **Application stability**: Ensure applications have sufficient capacity during node transitions
 - **Cost optimization**: Balance between node replacement speed and cluster stability
+- **Auto-scaling clusters**: Use percentage-based limits to maintain consistent safety margins as cluster size changes
 
 #### ExtraParkingLabels
 

charts/k8s-shredder/Chart.yaml

@@ -12,5 +12,5 @@ maintainers:
   - name: sfotony
     email: [email protected]
     url: https://adobe.com
-version: 0.2.6
-appVersion: v0.3.6
+version: 0.2.7
+appVersion: v0.3.7

charts/k8s-shredder/README.md

@@ -1,6 +1,6 @@
 # k8s-shredder
 
-![Version: 0.2.6](https://img.shields.io/badge/Version-0.2.6-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: v0.3.6](https://img.shields.io/badge/AppVersion-v0.3.6-informational?style=flat-square)
+![Version: 0.2.7](https://img.shields.io/badge/Version-0.2.7-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: v0.3.7](https://img.shields.io/badge/AppVersion-v0.3.7-informational?style=flat-square)
 
 a novel way of dealing with kubernetes nodes blocked from draining
 
@@ -64,7 +64,7 @@ a novel way of dealing with kubernetes nodes blocked from draining
 | serviceAccount.annotations | object | `{}` | Additional annotations for the service account (useful for IAM roles, etc.) |
 | serviceAccount.create | bool | `true` | Create a service account for k8s-shredder |
 | serviceAccount.name | string | `"k8s-shredder"` | Name of the service account |
-| shredder | object | `{"AllowEvictionLabel":"shredder.ethos.adobe.net/allow-eviction","ArgoRolloutsAPIVersion":"v1alpha1","EnableKarpenterDisruptionDetection":false,"EnableKarpenterDriftDetection":false,"EnableNodeLabelDetection":false,"EvictionLoopInterval":"1h","EvictionSafetyCheck":true,"ExpiresOnLabel":"shredder.ethos.adobe.net/parked-node-expires-on","ExtraParkingLabels":{},"MaxParkedNodes":0,"NamespacePrefixSkipInitialEviction":"ns-ethos-","NodeLabelsToDetect":[],"ParkedByLabel":"shredder.ethos.adobe.net/parked-by","ParkedByValue":"k8s-shredder","ParkedNodeTTL":"168h","ParkedNodeTaint":"shredder.ethos.adobe.net/upgrade-status=parked:NoSchedule","ParkingReasonLabel":"shredder.ethos.adobe.net/parked-reason","RestartedAtAnnotation":"shredder.ethos.adobe.net/restartedAt","RollingRestartThreshold":0.1,"ToBeDeletedTaint":"ToBeDeletedByClusterAutoscaler","UpgradeStatusLabel":"shredder.ethos.adobe.net/upgrade-status"}` | Core k8s-shredder configuration |
+| shredder | object | `{"AllowEvictionLabel":"shredder.ethos.adobe.net/allow-eviction","ArgoRolloutsAPIVersion":"v1alpha1","EnableKarpenterDisruptionDetection":false,"EnableKarpenterDriftDetection":false,"EnableNodeLabelDetection":false,"EvictionLoopInterval":"1h","EvictionSafetyCheck":true,"ExpiresOnLabel":"shredder.ethos.adobe.net/parked-node-expires-on","ExtraParkingLabels":{},"MaxParkedNodes":"0","NamespacePrefixSkipInitialEviction":"ns-ethos-","NodeLabelsToDetect":[],"ParkedByLabel":"shredder.ethos.adobe.net/parked-by","ParkedByValue":"k8s-shredder","ParkedNodeTTL":"168h","ParkedNodeTaint":"shredder.ethos.adobe.net/upgrade-status=parked:NoSchedule","ParkingReasonLabel":"shredder.ethos.adobe.net/parked-reason","RestartedAtAnnotation":"shredder.ethos.adobe.net/restartedAt","RollingRestartThreshold":0.1,"ToBeDeletedTaint":"ToBeDeletedByClusterAutoscaler","UpgradeStatusLabel":"shredder.ethos.adobe.net/upgrade-status"}` | Core k8s-shredder configuration |
 | shredder.AllowEvictionLabel | string | `"shredder.ethos.adobe.net/allow-eviction"` | Label to explicitly allow eviction on specific resources |
 | shredder.ArgoRolloutsAPIVersion | string | `"v1alpha1"` | API version for Argo Rollouts integration |
 | shredder.EnableKarpenterDisruptionDetection | bool | `false` | Enable Karpenter disruption detection for node lifecycle management |
@@ -74,7 +74,7 @@ a novel way of dealing with kubernetes nodes blocked from draining
 | shredder.EvictionSafetyCheck | bool | `true` | Controls whether to perform safety checks before force eviction |
 | shredder.ExpiresOnLabel | string | `"shredder.ethos.adobe.net/parked-node-expires-on"` | Label used to track when a parked node expires |
 | shredder.ExtraParkingLabels | object | `{}` | Additional labels to apply to nodes and pods during parking |
-| shredder.MaxParkedNodes | int | `0` | Maximum number of nodes that can be parked simultaneously (0 = no limit) |
+| shredder.MaxParkedNodes | string | `"0"` | Maximum number of nodes that can be parked simultaneously. Can be an integer (e.g., "5") or percentage (e.g., "20%"). Set to "0" for no limit |
 | shredder.NamespacePrefixSkipInitialEviction | string | `"ns-ethos-"` | Namespace prefix to skip during initial eviction (useful for system namespaces) |
 | shredder.NodeLabelsToDetect | list | `[]` | List of node labels to monitor for triggering shredder actions |
 | shredder.ParkedByLabel | string | `"shredder.ethos.adobe.net/parked-by"` | Label to track which component parked a node |

charts/k8s-shredder/values.yaml

@@ -62,8 +62,8 @@ shredder:
   EnableNodeLabelDetection: false
   # -- List of node labels to monitor for triggering shredder actions
   NodeLabelsToDetect: []
-  # -- Maximum number of nodes that can be parked simultaneously (0 = no limit)
-  MaxParkedNodes: 0
+  # -- Maximum number of nodes that can be parked simultaneously. Can be an integer (e.g., "5") or percentage (e.g., "20%"). Set to "0" for no limit
+  MaxParkedNodes: '0'
   # -- Controls whether to perform safety checks before force eviction
   EvictionSafetyCheck: true
   # -- Label used to track why a node or pod was parked

cmd/root.go

@@ -12,6 +12,7 @@ governing permissions and limitations under the License.
 package cmd
 
 import (
+	"strconv"
 	"strings"
 	"time"
 
@@ -123,7 +124,7 @@ func discoverConfig() {
 	viper.SetDefault("ParkedNodeTaint", "shredder.ethos.adobe.net/upgrade-status=parked:NoSchedule")
 	viper.SetDefault("EnableNodeLabelDetection", false)
 	viper.SetDefault("NodeLabelsToDetect", []string{})
-	viper.SetDefault("MaxParkedNodes", 0)
+	viper.SetDefault("MaxParkedNodes", "0")
 	viper.SetDefault("ExtraParkingLabels", map[string]string{})
 	viper.SetDefault("EvictionSafetyCheck", true)
 	viper.SetDefault("ParkingReasonLabel", "shredder.ethos.adobe.net/parked-reason")
@@ -149,9 +150,22 @@ func parseConfig() {
 	}
 
 	// Validate MaxParkedNodes configuration
-	if cfg.MaxParkedNodes < 0 {
-		log.WithField("MaxParkedNodes", cfg.MaxParkedNodes).Warn("MaxParkedNodes is negative, treating as 0 (no limit)")
-		cfg.MaxParkedNodes = 0
+	// Basic validation - detailed parsing happens in LimitNodesToPark
+	if cfg.MaxParkedNodes != "" && cfg.MaxParkedNodes != "0" {
+		// Check if it's a percentage
+		if strings.HasSuffix(cfg.MaxParkedNodes, "%") {
+			percentageStr := strings.TrimSuffix(cfg.MaxParkedNodes, "%")
+			if _, err := strconv.ParseFloat(percentageStr, 64); err != nil {
+				log.WithField("MaxParkedNodes", cfg.MaxParkedNodes).Warn("Invalid MaxParkedNodes percentage format, treating as 0 (no limit)")
+				cfg.MaxParkedNodes = "0"
+			}
+		} else {
+			// Check if it's a valid integer
+			if _, err := strconv.Atoi(cfg.MaxParkedNodes); err != nil {
+				log.WithField("MaxParkedNodes", cfg.MaxParkedNodes).Warn("Invalid MaxParkedNodes format, treating as 0 (no limit)")
+				cfg.MaxParkedNodes = "0"
+			}
+		}
 	}
 
 	log.WithFields(log.Fields{

config.yaml

@@ -32,7 +32,7 @@ NodeLabelsToDetect: []  # List of node labels to detect. Supports both key-only
   # - "node.example.com/park"  # Matches any node with the "node.example.com/park" label
 
 # Parking limits
-MaxParkedNodes: 0  # Maximum number of nodes that can be parked simultaneously. Set to 0 (default) for no limit.
+MaxParkedNodes: '0'  # Maximum number of nodes that can be parked simultaneously. Can be an integer (e.g., "5") or percentage (e.g., "20%"). Set to "0" (default) for no limit.
 
 # Extra labels to apply to parked nodes and pods
 # ExtraParkingLabels:  # (optional) Additional labels to apply to nodes and pods during parking

internal/testing/k8s-shredder-karpenter.yaml

@@ -80,7 +80,7 @@ data:
     ParkedNodeTaint: "shredder.ethos.adobe.net/upgrade-status=parked:NoSchedule"
     EnableNodeLabelDetection: false
     NodeLabelsToDetect: []
-    MaxParkedNodes: 0
+    MaxParkedNodes: "0"
 ---
 apiVersion: v1
 kind: Service

internal/testing/k8s-shredder-node-labels.yaml

@@ -83,7 +83,7 @@ data:
       - "test-label"
       - "maintenance=scheduled"
       - "node.test.io/park"
-    MaxParkedNodes: 0
+    MaxParkedNodes: "0"
 ---
 apiVersion: v1
 kind: Service

internal/testing/k8s-shredder.yaml

@@ -80,7 +80,7 @@ data:
     ParkedNodeTaint: "shredder.ethos.adobe.net/upgrade-status=parked:NoSchedule"
     EnableNodeLabelDetection: false
     NodeLabelsToDetect: []
-    MaxParkedNodes: 0
+    MaxParkedNodes: "0"
 ---
 apiVersion: v1
 kind: Service

pkg/config/config.go

@@ -49,8 +49,11 @@ type Config struct {
 	EnableNodeLabelDetection bool
 	// NodeLabelsToDetect is a list of node labels to look for. Can be just keys or key=value pairs
 	NodeLabelsToDetect []string
-	// MaxParkedNodes is the maximum number of nodes that can be parked simultaneously. If set to 0 (default), no limit is applied.
-	MaxParkedNodes int
+	// MaxParkedNodes is the maximum number of nodes that can be parked simultaneously.
+	// Can be either an integer (e.g. "5") or a percentage (e.g. "20%").
+	// If set to "0" or empty (default), no limit is applied.
+	// When a percentage is specified, the limit is calculated as (percentage/100) * (total nodes in cluster).
+	MaxParkedNodes string
 	// ExtraParkingLabels is a map of additional labels to apply to nodes and pods during the parking process. If not set, no extra labels are applied.
 	ExtraParkingLabels map[string]string
 	// EvictionSafetyCheck controls whether to perform safety checks before force eviction. If true, nodes will be unparked if pods don't have required parking labels.

pkg/utils/karpenter_detection_test.go

@@ -205,7 +205,7 @@ func TestLabelDriftedNodes(t *testing.T) {
 			name:              "No drifted node claims",
 			driftedNodeClaims: []KarpenterNodeClaimInfo{},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -225,7 +225,7 @@ func TestLabelDriftedNodes(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -245,7 +245,7 @@ func TestLabelDriftedNodes(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -265,7 +265,7 @@ func TestLabelDriftedNodes(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -484,7 +484,7 @@ func TestLabelDisruptedNodes(t *testing.T) {
 			name:                "No disrupted node claims",
 			disruptedNodeClaims: []KarpenterNodeClaimInfo{},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -505,7 +505,7 @@ func TestLabelDisruptedNodes(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -526,7 +526,7 @@ func TestLabelDisruptedNodes(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -547,7 +547,7 @@ func TestLabelDisruptedNodes(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 				ParkingReasonLabel: "parked-reason",
 			},
@@ -614,7 +614,7 @@ func TestProcessDriftedKarpenterNodes(t *testing.T) {
 			appContext: &AppContext{
 				Config: config.Config{
 					UpgradeStatusLabel: "upgrade-status",
-					MaxParkedNodes:     5,
+					MaxParkedNodes:     "5",
 					ParkingReasonLabel: "parked-reason",
 				},
 				K8sClient:        fake.NewSimpleClientset(),

pkg/utils/node_label_detection_test.go

@@ -430,7 +430,7 @@ func TestParkNodesWithLabels(t *testing.T) {
 			name:          "No matching nodes",
 			matchingNodes: []NodeLabelInfo{},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 			},
 			dryRun:      false,
@@ -448,7 +448,7 @@ func TestParkNodesWithLabels(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 			},
 			dryRun:      false,
@@ -472,7 +472,7 @@ func TestParkNodesWithLabels(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 			},
 			dryRun:      false,
@@ -490,7 +490,7 @@ func TestParkNodesWithLabels(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     5,
+				MaxParkedNodes:     "5",
 				UpgradeStatusLabel: "upgrade-status",
 			},
 			dryRun:      true,
@@ -520,7 +520,7 @@ func TestParkNodesWithLabels(t *testing.T) {
 				},
 			},
 			cfg: config.Config{
-				MaxParkedNodes:     2,
+				MaxParkedNodes:     "2",
 				UpgradeStatusLabel: "upgrade-status",
 			},
 			dryRun:      false,