Support 'cull connected kernels' as a config in helm and create a way to dynamically pass other environments variables to helm deployment (#1231)

* support cull connected kernels as config in helm
* support passing multiple env vars dynamically
* add documentation
* add missing env keywords to fix helm lint
* change example to current env variables that eg actually use
* change example to a random string instead of EG_ env vars
* address yaml boolean in doc for cullConnected

Signed-off-by: Luong Vo <[email protected]>
Signed-off-by: Luong Vo <[email protected]>

Author: luong-komorebi

docs/source/operators/deploy-kubernetes.md

@@ -474,13 +474,15 @@ can override them with helm's `--set` or `--values` options. Always use `--set`
 | `deployment.terminationGracePeriodSeconds` | Time to wait for Enterprise Gateway to gracefully shutdown.                                                                                                                                                                                      | `30`                                                                           |
 | `deployment.resources`                     | set Enterprise Gateway container resources.                                                                                                                                                                                                      | valid Yaml resources, see values file for example                              |
 | `deployment.replicas`                      | Update to deploy multiple replicas of EG.                                                                                                                                                                                                        | `1`                                                                            |
+| `deployment.extraEnv`                      | Additional environment variables to set for Enterprise Gateway.                                                                                                                                                                                  | `{}`                                                                           |
 | `logLevel`                                 | Log output level.                                                                                                                                                                                                                                | `DEBUG`                                                                        |
 | `mirrorWorkingDirs`                        | Whether to mirror working directories. NOTE: This requires appropriate volume mounts to make notebook dir accessible.                                                                                                                            | `false`                                                                        |
 | `authToken`                                | Optional authorization token passed in all requests (see --EnterpriseGatewayApp.auth_token)                                                                                                                                                      | `nil`                                                                          |
 | `kernel.clusterRole`                       | Kernel cluster role created by this chart. Used if no KERNEL_NAMESPACE is provided by client.                                                                                                                                                    | `kernel-controller`                                                            |
 | `kernel.shareGatewayNamespace`             | Will start kernels in the same namespace as EG if True.                                                                                                                                                                                          | `false`                                                                        |
 | `kernel.launchTimeout`                     | Timeout for kernel launching in seconds.                                                                                                                                                                                                         | `60`                                                                           |
 | `kernel.cullIdleTimeout`                   | Idle timeout in seconds. Default is 1 hour.                                                                                                                                                                                                      | `3600`                                                                         |
+| `kernel.cullConnected`                     | Whether to cull idle kernels that still have clients connected.                                                                                                                                                                                  | `false`                                                                        |
 | `kernel.allowedKernels`                    | List of kernel names that are available for use.                                                                                                                                                                                                 | `{r_kubernetes,...}` (see `values.yaml`)                                       |
 | `kernel.defaultKernelName`                 | Default kernel name should be something from the allowedKernels                                                                                                                                                                                  | `python-kubernetes`                                                            |
 | `kernelspecs.image`                        | Optional custom data image containing kernelspecs to use. Cannot be used with NFS enabled.                                                                                                                                                       | `nil`                                                                          |

etc/kubernetes/helm/enterprise-gateway/templates/deployment.yaml

@@ -65,6 +65,8 @@ spec:
           value: {{ if .Values.mirrorWorkingDirs }}"True"{{ else }}"False"{{ end }}
         - name: EG_CULL_IDLE_TIMEOUT
           value: !!str {{ .Values.kernel.cullIdleTimeout }}
+        - name: EG_CULL_CONNECTED
+          value: {{ if .Values.kernel.cullConnected }}"True"{{ else }}"False"{{ end }}
         - name: EG_LOG_LEVEL
           value: {{ .Values.logLevel }}
         - name: EG_KERNEL_LAUNCH_TIMEOUT
@@ -78,6 +80,13 @@ spec:
         - name: EG_AUTH_TOKEN
           value: {{ .Values.authToken }}
         {{- end }}
+        {{- if .Values.deployment.extraEnv }}
+        {{- range $key, $val := .Values.deployment.extraEnv }}
+        - name: {{ $key }}
+          value: |-
+{{ $val | indent 12 }}
+        {{- end }}
+        {{- end }}
         ports:
         {{ with index .Values.service.ports 0 }}
         - containerPort: {{ .port }}

etc/kubernetes/helm/enterprise-gateway/values.yaml

@@ -62,6 +62,16 @@ deployment:
   tolerations: []
   affinity: {}
   nodeSelector: {}
+  extraEnv: {}
+    # SOME_ENV_VAR_WITH_SIMPLE_VALUE: "example"
+    # SOME_ENV_VAR_WITH_LONG_VALUE: >
+    #   'this example',
+    #   'will not preserve',
+    #   'line breaks',
+    # SOME_ENV_VAR_WITH_MULTILINE_VALUE: |
+    #   this example
+    #   will preserve
+    #   line breaks
 
 # Log output level.
 logLevel: DEBUG
@@ -79,6 +89,8 @@ kernel:
   launchTimeout: 60
   # Timeout for an idle kernel before its culled in seconds. Default is 1 hour.
   cullIdleTimeout: 3600
+  # Whether to cull idle kernels with connecting clients
+  cullConnected: false
   # List of kernel names that are available for use. To allow additional kernelspecs without
   # requiring redeployment (and assuming kernelspecs are mounted or otherwise accessible
   # outside the pod), comment out (or remove) the entries, leaving only `allowedKernels:`.