Multiple applications per pod

Dockerfile

@@ -1,5 +1,5 @@
 # Build the manager binary
-FROM golang:1.20 as builder
+FROM golang:1.20 AS builder
 
 WORKDIR /workspace
 

api/v1alpha2/minicluster_types.go

@@ -50,6 +50,11 @@ type MiniClusterSpec struct {
 	// +optional
 	Interactive bool `json:"interactive"`
 
+	// Allow >1 Flux running (oversubscribing resources)
+	// +kubebuilder:default=false
+	// +optional
+	Oversubscribe bool `json:"oversubscribe"`
+
 	// Flux options for the broker, shared across cluster
 	// +optional
 	Flux FluxSpec `json:"flux"`
@@ -825,13 +830,6 @@ func (f *MiniCluster) Validate() bool {
 		// Count the FluxRunners
 		if container.RunFlux {
 			fluxRunners += 1
-
-			// Non flux-runners are required to have a name
-		} else {
-			if container.Name == "" {
-				fmt.Printf("😥️ %s is missing a name\n", name)
-				return false
-			}
 		}
 
 		// If a custom script is provided AND a command, no go
@@ -840,7 +838,16 @@ func (f *MiniCluster) Validate() bool {
 			return false
 		}
 	}
-	if fluxRunners != 1 {
+
+	// If we have more than one flux runner, must explicitly oversubscribe
+	if fluxRunners > 1 && !f.Spec.Oversubscribe {
+		fmt.Printf("😥️ More than one flux runner requires oversubscribe: true\n")
+		valid = false
+	}
+
+	// More than one container can run Flux (and the brokers see the same resources)
+	// But we need at least one!
+	if fluxRunners < 1 {
 		valid = false
 	}
 

api/v1alpha2/swagger.json

@@ -626,6 +626,11 @@
           "default": {},
           "$ref": "#/definitions/Network"
         },
+        "oversubscribe": {
+          "description": "Allow \u003e1 Flux running (oversubscribing resources)",
+          "type": "boolean",
+          "default": false
+        },
         "pod": {
           "description": "Pod spec details",
           "default": {},

chart/templates/minicluster-crd.yaml

@@ -523,6 +523,10 @@ spec:
                     description: Name for cluster headless service
                     type: string
                 type: object
+              oversubscribe:
+                default: false
+                description: Allow >1 Flux running (oversubscribing resources)
+                type: boolean
               pod:
                 description: Pod spec details
                 properties:

config/crd/bases/flux-framework.org_miniclusters.yaml

@@ -526,6 +526,10 @@ spec:
                     description: Name for cluster headless service
                     type: string
                 type: object
+              oversubscribe:
+                default: false
+                description: Allow >1 Flux running (oversubscribing resources)
+                type: boolean
               pod:
                 description: Pod spec details
                 properties:

controllers/flux/containers.go

@@ -53,7 +53,7 @@ func getFluxContainer(
 
 func getContainers(
 	specs []api.MiniClusterContainer,
-	defaultName string,
+	customName string,
 	mounts []corev1.VolumeMount,
 	serviceContainer bool,
 ) ([]corev1.Container, error) {
@@ -70,8 +70,15 @@ func getContainers(
 			pullPolicy = corev1.PullAlways
 		}
 
-		// Fluxrunner will use the namespace name
-		containerName := container.Name
+		// Give all flux containers a name, if not provided
+		if container.Name == "" {
+			// Maintain previous behavior to have name == main flux runner
+			if i == 0 {
+				container.Name = customName
+			} else {
+				container.Name = fmt.Sprintf("%s-%d", container.Name, i)
+			}
+		}
 		command := []string{}
 
 		// A Flux runner will have a wait.sh script that waits for the flux view
@@ -82,7 +89,6 @@ func getContainers(
 			// wait.sh path corresponds to container identifier
 			waitScript := fmt.Sprintf("/flux_operator/wait-%d.sh", i)
 			command = []string{"/bin/bash", waitScript}
-			containerName = defaultName
 		}
 
 		// A container not running flux can only have pre/post sections
@@ -140,7 +146,7 @@ func getContainers(
 		newContainer := corev1.Container{
 
 			// Call this the driver container, number 0
-			Name:            containerName,
+			Name:            container.Name,
 			Image:           container.Image,
 			ImagePullPolicy: pullPolicy,
 			WorkingDir:      container.WorkingDir,

controllers/flux/job.go

@@ -98,6 +98,8 @@ func NewMiniClusterJob(cluster *api.MiniCluster) (*batchv1.Job, error) {
 	}
 
 	// Prepare listing of containers for the MiniCluster
+	// We don't provide a default name because defaults are provided in Validate()
+	// Only service containers have a custom name here
 	containers, err := getContainers(
 		cluster.Spec.Containers,
 		cluster.Name,

docs/getting_started/custom-resource-definition.md

@@ -92,6 +92,16 @@ This would be equivalent to giving a start command of `sleep infinity` however o
 (e.g., if there is a flux shutdown from within the Flux instance) the sleep command would
 not exit with a failed code.
 
+### oversubscribe
+
+By default, we treat your single application container _or_ the single container in a MiniCluster pod designated to "runFlux" as the only Flux broker. When oversubscribe is set to true, you are allowed to define more than one "runFlux" container, meaning that multiple brokers will be sharing the same resources.
+
+```yaml
+  oversubscribe: true
+```
+
+We created this use case with the intention of having a service container running fluxion alongside the MiniCluster to orchestrate the N containers. This is consiedered an advanced use case and you should use it with caution!
+
 ### launcher
 
 If you are using an executor that launches Flux Jobs (e.g., workflow managers such as Snakemake and Nextflow do!)

docs/tutorials/index.md

@@ -56,6 +56,7 @@ The following tutorials are provided from their respective directories (and are
 
 These examples show how to interact with your flux queue from a sidecar container (that has access to the flux broker of the pod):
 
+ - [multiple-applications-per-pod](https://github.com/flux-framework/flux-operator/tree/main/examples/experimental/multiple-applications-per-pod): Allow multiple applications to run in a single pod aided by a Fluxion service orchestrator.
  - [flux-sidecar](https://github.com/flux-framework/flux-operator/blob/main/examples/tests/flux-sidecar) to see a sleep job in the main application queue
 
 ### Services

examples/dist/flux-operator-arm.yaml

@@ -532,6 +532,10 @@ spec:
                     description: Name for cluster headless service
                     type: string
                 type: object
+              oversubscribe:
+                default: false
+                description: Allow >1 Flux running (oversubscribing resources)
+                type: boolean
               pod:
                 description: Pod spec details
                 properties:

examples/dist/flux-operator.yaml

@@ -532,6 +532,10 @@ spec:
                     description: Name for cluster headless service
                     type: string
                 type: object
+              oversubscribe:
+                default: false
+                description: Allow >1 Flux running (oversubscribing resources)
+                type: boolean
               pod:
                 description: Pod spec details
                 properties: