K8s: Distributor uses Greedy algo as slot selector strategy in autoscaling (#2875)

VietND96 · web-flow · commit 3c1453ecf482 · 2025-07-03T06:00:19.000+07:00
K8s: Distributor uses Greedy as slot selector strategy in autoscaling

Signed-off-by: Viet Nguyen Duc &lt;nguyenducviet4496@gmail.com&gt;
diff --git a/.github/workflows/deploy.yml b/.github/workflows/deploy.yml
@@ -128,7 +128,7 @@ jobs:
         if: github.event.inputs.skip-build-push-image != 'true'
         uses: nick-invision/retry@master
         with:
-          timeout_minutes: 90
+          timeout_minutes: 160
           max_attempts: 3
           retry_wait_seconds: 60
           command: PLATFORMS="${PLATFORMS}" VERSION="${GRID_VERSION}" BUILD_DATE=${BUILD_DATE} make build
diff --git a/.github/workflows/docker-test.yml b/.github/workflows/docker-test.yml
@@ -222,7 +222,7 @@ jobs:
         uses: nick-invision/retry@master
         if: matrix.build-all == true
         with:
-          timeout_minutes: 20
+          timeout_minutes: 90
           max_attempts: 3
           retry_wait_seconds: 60
           command: |
diff --git a/.github/workflows/helm-chart-test.yml b/.github/workflows/helm-chart-test.yml
@@ -197,7 +197,7 @@ jobs:
       - name: Build Docker images
         uses: nick-invision/retry@master
         with:
-          timeout_minutes: 12
+          timeout_minutes: 90
           max_attempts: 3
           retry_wait_seconds: 60
           command: NAME=${IMAGE_REGISTRY} VERSION=${BRANCH} BUILD_DATE=${BUILD_DATE} make build
diff --git a/.github/workflows/k8s-scaling-test.yml b/.github/workflows/k8s-scaling-test.yml
@@ -178,7 +178,7 @@ jobs:
       - name: Build Docker images
         uses: nick-invision/retry@master
         with:
-          timeout_minutes: 12
+          timeout_minutes: 30
           max_attempts: 3
           retry_wait_seconds: 60
           command: NAME=${IMAGE_REGISTRY} VERSION=${BRANCH} BUILD_DATE=${BUILD_DATE} make build
diff --git a/.github/workflows/nightly.yml b/.github/workflows/nightly.yml
@@ -78,7 +78,7 @@ jobs:
       - name: Build images
         uses: nick-invision/retry@master
         with:
-          timeout_minutes: 90
+          timeout_minutes: 160
           max_attempts: 3
           retry_wait_seconds: 60
           command: PLATFORMS="${PLATFORMS}" VERSION="${GRID_VERSION}" BUILD_DATE=${BUILD_DATE} make build
diff --git a/Distributor/start-selenium-grid-distributor.sh b/Distributor/start-selenium-grid-distributor.sh
@@ -145,6 +145,10 @@ if [ ! -z "$SE_REJECT_UNSUPPORTED_CAPS" ]; then
   append_se_opts "--reject-unsupported-caps" "${SE_REJECT_UNSUPPORTED_CAPS}"
 fi
 
+if [ ! -z "$SE_DISTRIBUTOR_SLOT_SELECTOR" ]; then
+  append_se_opts "--slot-selector" "${SE_DISTRIBUTOR_SLOT_SELECTOR}"
+fi
+
 if [ ! -z "$SE_NEW_SESSION_THREAD_POOL_SIZE" ]; then
   append_se_opts "--newsession-threadpool-size" "${SE_NEW_SESSION_THREAD_POOL_SIZE}"
 fi
diff --git a/ENV_VARIABLES.md b/ENV_VARIABLES.md
@@ -154,3 +154,4 @@
 | SE_VIDEO_MAXRATE |  |  |  |
 | SE_NODE_DELETE_SESSION_ON_UI | true | Enable capability to support deleting session on Grid UI | --delete-session-on-ui |
 | SE_UPDATE_CHROME_COMPONENTS |  | Applicable for node-chrome, standalone-chrome (arch linux/amd64). Update the latest version of Chrome and ChromeDriver at the beginning of the container startup. Read more: [#2872](https://github.com/SeleniumHQ/docker-selenium/pull/2872) |  |
+| SE_DISTRIBUTOR_SLOT_SELECTOR |  | Full class name of non-default slot selector. This is used to select a slot in a Node once the Node has been matched. | --slot-selector |
diff --git a/Hub/start-selenium-grid-hub.sh b/Hub/start-selenium-grid-hub.sh
@@ -101,6 +101,10 @@ if [ ! -z "$SE_REJECT_UNSUPPORTED_CAPS" ]; then
   append_se_opts "--reject-unsupported-caps" "${SE_REJECT_UNSUPPORTED_CAPS}"
 fi
 
+if [ ! -z "$SE_DISTRIBUTOR_SLOT_SELECTOR" ]; then
+  append_se_opts "--slot-selector" "${SE_DISTRIBUTOR_SLOT_SELECTOR}"
+fi
+
 if [ ! -z "$SE_NEW_SESSION_THREAD_POOL_SIZE" ]; then
   append_se_opts "--newsession-threadpool-size" "${SE_NEW_SESSION_THREAD_POOL_SIZE}"
 fi
diff --git a/Standalone/start-selenium-standalone.sh b/Standalone/start-selenium-standalone.sh
@@ -128,6 +128,10 @@ if [ ! -z "$SE_REJECT_UNSUPPORTED_CAPS" ]; then
   append_se_opts "--reject-unsupported-caps" "${SE_REJECT_UNSUPPORTED_CAPS}"
 fi
 
+if [ ! -z "$SE_DISTRIBUTOR_SLOT_SELECTOR" ]; then
+  append_se_opts "--slot-selector" "${SE_DISTRIBUTOR_SLOT_SELECTOR}"
+fi
+
 if [ ! -z "$SE_NEW_SESSION_THREAD_POOL_SIZE" ]; then
   append_se_opts "--newsession-threadpool-size" "${SE_NEW_SESSION_THREAD_POOL_SIZE}"
 fi
diff --git a/charts/selenium-grid/CONFIGURATION.md b/charts/selenium-grid/CONFIGURATION.md
@@ -219,6 +219,7 @@ A Helm chart for creating a Selenium Grid Server in Kubernetes
 | components.distributor.imagePullPolicy | string | `"IfNotPresent"` | Image pull policy (see https://kubernetes.io/docs/concepts/containers/images/#updating-images) |
 | components.distributor.imagePullSecret | string | `""` | Image pull secret (see https://kubernetes.io/docs/tasks/configure-pod-container/pull-image-private-registry/) |
 | components.distributor.newSessionThreadPoolSize | string | `nil` | Configure fixed-sized thread pool for the Distributor to create new sessions as it consumes new session requests from the queue |
+| components.distributor.slotSelectorStrategy | string | `""` | Full class name of non-default slot selector. This is used to select a slot in a Node once the Node has been matched |
 | components.distributor.extraEnvironmentVariables | list | `[]` | Specify extra environment variables for Distributor |
 | components.distributor.extraEnvFrom | list | `[]` | Specify extra environment variables from ConfigMap and Secret for Distributor |
 | components.distributor.affinity | object | `{}` | Specify affinity for distributor pods, this overwrites global.seleniumGrid.affinity parameter |
@@ -340,6 +341,7 @@ A Helm chart for creating a Selenium Grid Server in Kubernetes
 | hub.readinessProbe | object | `{"enabled":true,"failureThreshold":10,"initialDelaySeconds":12,"path":"/readyz","periodSeconds":10,"successThreshold":1,"timeoutSeconds":10}` | Readiness probe settings |
 | hub.livenessProbe | object | `{"enabled":true,"failureThreshold":30,"initialDelaySeconds":60,"path":"/readyz","periodSeconds":60,"successThreshold":1,"timeoutSeconds":60}` | Liveness probe settings |
 | hub.subPath | string | `""` | Custom sub path for the hub deployment |
+| hub.slotSelectorStrategy | string | `""` | Full class name of non-default slot selector. This is used to select a slot in a Node once the Node has been matched |
 | hub.extraEnvironmentVariables | list | `[]` | Custom environment variables for selenium-hub |
 | hub.extraEnvFrom | list | `[]` | Custom environment variables by sourcing entire configMap, Secret, etc. for selenium-hub |
 | hub.extraVolumeMounts | list | `[]` | Extra volume mounts for Hub container |
@@ -391,6 +393,7 @@ A Helm chart for creating a Selenium Grid Server in Kubernetes
 | autoscaling.enableWithExistingKEDA | bool | `false` | Enable autoscaling without automatically installing KEDA |
 | autoscaling.scalingType | string | `"job"` | Which type of KEDA scaling to use: job or deployment |
 | autoscaling.setReplicasInSpec | bool | `true` | Force remove replicas in deployment spec in case ArgoCD with AutoSync enabled will try to resolve back to desired state |
+| autoscaling.slotSelectorStrategy | string | `"org.openqa.selenium.grid.distributor.selector.GreedySlotSelector"` | Strategy for Selenium Hub/Distributor select slot to assign to a new session. |
 | autoscaling.authenticationRef | object | `{"annotations":{"helm.sh/hook":"post-install,post-upgrade,post-rollback","helm.sh/hook-weight":"0"},"name":""}` | Specify an external KEDA TriggerAuthentication resource is used for scaler triggers config. Apply for all browser nodes |
 | autoscaling.useCachedMetrics | bool | `false` | Enables caching of metric values during polling interval (as specified in .spec.pollingInterval, the default: false in KEDA). |
 | autoscaling.triggerName | string | `""` | Set trigger name. |
diff --git a/charts/selenium-grid/templates/_helpers.tpl b/charts/selenium-grid/templates/_helpers.tpl
@@ -202,6 +202,16 @@ based on sum of maxReplicaCount of all enabled Nodes in autoscaling
 {{- $threadPoolSize -}}
 {{- end -}}
 
+{{- define "seleniumGrid.autoscaling.distributor.slotSelector" -}}
+{{- $slotSelector := "" -}}
+{{- if eq (include "seleniumGrid.useKEDA" $) "true" -}}
+{{- $slotSelector = $.Values.autoscaling.slotSelectorStrategy -}}
+{{- else -}}
+{{- $slotSelector = $.Values.isolateComponents | ternary $.Values.components.distributor.slotSelectorStrategy $.Values.hub.slotSelectorStrategy -}}
+{{- end -}}
+{{- $slotSelector -}}
+{{- end -}}
+
 {{/*
 Common autoscaling spec template
 */}}
diff --git a/charts/selenium-grid/templates/distributor-deployment.yaml b/charts/selenium-grid/templates/distributor-deployment.yaml
@@ -73,6 +73,10 @@ spec:
             - name: SE_NEW_SESSION_THREAD_POOL_SIZE
               value: '{{ template "seleniumGrid.autoscaling.distributor.threadPoolSize" $ }}'
             {{- end }}
+            {{- if not (empty (include "seleniumGrid.autoscaling.distributor.slotSelector" $)) }}
+            - name: SE_DISTRIBUTOR_SLOT_SELECTOR
+              value: {{ include "seleniumGrid.autoscaling.distributor.slotSelector" $ | quote }}
+            {{- end }}
           {{- with .Values.components.extraEnvironmentVariables }}
             {{- tpl (toYaml .) $ | nindent 12 }}
           {{- end }}
diff --git a/charts/selenium-grid/templates/hub-deployment.yaml b/charts/selenium-grid/templates/hub-deployment.yaml
@@ -130,6 +130,10 @@ spec:
             - name: SE_NEW_SESSION_THREAD_POOL_SIZE
               value: '{{ template "seleniumGrid.autoscaling.distributor.threadPoolSize" $ }}'
             {{- end }}
+            {{- if not (empty (include "seleniumGrid.autoscaling.distributor.slotSelector" $)) }}
+            - name: SE_DISTRIBUTOR_SLOT_SELECTOR
+              value: {{ include "seleniumGrid.autoscaling.distributor.slotSelector" $ | quote }}
+            {{- end }}
           {{- with .Values.hub.extraEnvironmentVariables }}
             {{- tpl (toYaml .) $ | nindent 12 }}
           {{- end }}
diff --git a/charts/selenium-grid/values.yaml b/charts/selenium-grid/values.yaml
@@ -529,6 +529,8 @@ components:
 
     # -- Configure fixed-sized thread pool for the Distributor to create new sessions as it consumes new session requests from the queue
     newSessionThreadPoolSize:
+    # -- Full class name of non-default slot selector. This is used to select a slot in a Node once the Node has been matched
+    slotSelectorStrategy: ""
     # -- Specify extra environment variables for Distributor
     extraEnvironmentVariables: []
     # -- Specify extra environment variables from ConfigMap and Secret for Distributor
@@ -885,6 +887,8 @@ hub:
     successThreshold: 1
   # -- Custom sub path for the hub deployment
   subPath: ""
+  # -- Full class name of non-default slot selector. This is used to select a slot in a Node once the Node has been matched
+  slotSelectorStrategy: ""
   # -- Custom environment variables for selenium-hub
   extraEnvironmentVariables: []
     # - name: SE_JAVA_OPTS
@@ -1015,6 +1019,12 @@ autoscaling:
   scalingType: job
   # -- Force remove replicas in deployment spec in case ArgoCD with AutoSync enabled will try to resolve back to desired state
   setReplicasInSpec: true
+  # By default, it is "org.openqa.selenium.grid.distributor.selector.DefaultSlotSelector". This is ensuring load balancing sessions across all available nodes.
+  # New strategy "org.openqa.selenium.grid.distributor.selector.GreedySlotSelector". This might be helpful in case autoscaling overprovisioned with max sessions greater than 1 per node.
+  # Ensure all slots in a node are filled with sessions before selecting the next available node. Note this is available since Selenium Server 4.34.0.
+  # Change this default chart value if you prefer default strategy or using this chart with older Selenium Server versions.
+  # -- Strategy for Selenium Hub/Distributor select slot to assign to a new session.
+  slotSelectorStrategy: "org.openqa.selenium.grid.distributor.selector.GreedySlotSelector"
   # -- Specify an external KEDA TriggerAuthentication resource is used for scaler triggers config. Apply for all browser nodes
   authenticationRef:
     name: ""
diff --git a/scripts/generate_list_env_vars/description.yaml b/scripts/generate_list_env_vars/description.yaml
@@ -473,3 +473,7 @@
     Update the latest version of Chrome and ChromeDriver at the beginning of the container
     startup. Read more: [#2872](https://github.com/SeleniumHQ/docker-selenium/pull/2872)'
   cli: ''
+- name: SE_DISTRIBUTOR_SLOT_SELECTOR
+  description: Full class name of non-default slot selector. This is used to select
+    a slot in a Node once the Node has been matched.
+  cli: --slot-selector
diff --git a/scripts/generate_list_env_vars/value.yaml b/scripts/generate_list_env_vars/value.yaml
@@ -18,6 +18,8 @@
   default: ''
 - name: SE_DISTRIBUTOR_PORT
   default: '5553'
+- name: SE_DISTRIBUTOR_SLOT_SELECTOR
+  default: ''
 - name: SE_DRAIN_AFTER_SESSION_COUNT
   default: '0'
 - name: SE_ENABLE_BROWSER_LEFTOVERS_CLEANUP
