feat: playbook agents

* playbook run listener support playbook_run as the parent not just
  notification_send

Author: adityathebe

api/v1/playbook_actions.go

@@ -300,7 +300,7 @@ func (t *AIActionClient) Populate(ctx context.Context) error {
 		if err != nil {
 			return err
 		} else if conn == nil {
-			return fmt.Errorf("connection(%s) was not found: %w", *t.Connection, err)
+			return fmt.Errorf("connection(%s) was not found", *t.Connection)
 		}
 
 		if err := t.APIKey.Scan(conn.Password); err != nil {
@@ -385,6 +385,8 @@ type AIAction struct {
 	// When enabled, the prompt is simply saved without passing it on to the LLM.
 	DryRun bool `json:"dryRun,omitempty"`
 
+	PlaybookAgents []types.ResourceSelector `json:"playbookAgents,omitempty"`
+
 	// Specify selectors for playbooks. The LLM will recommend the best suited playbooks
 	// in response to the prompt.
 	RecommendPlaybooks []types.ResourceSelector `json:"recommendPlaybooks,omitempty"`

api/v1/zz_generated.deepcopy.go

@@ -140,6 +140,66 @@ spec:
                             Model name based on the backend chosen.
                             Example: gpt-4o for openai, claude-3-5-sonnet-latest for Anthropic, llama3.1:8b for Ollama
                           type: string
+                        playbookAgents:
+                          items:
+                            properties:
+                              agent:
+                                description: |-
+                                  Agent can be the agent id or the name of the agent.
+                                   Additionally, the special "self" value can be used to select resources without an agent.
+                                type: string
+                              cache:
+                                description: |-
+                                  Cache directives
+                                   'no-cache' (should not fetch from cache but can be cached)
+                                   'no-store' (should not cache)
+                                   'max-age=X' (cache for X duration)
+                                type: string
+                              fieldSelector:
+                                type: string
+                              health:
+                                description: |-
+                                  Health filters resources by the health.
+                                  Multiple healths can be provided separated by comma.
+                                type: string
+                              id:
+                                type: string
+                              includeDeleted:
+                                type: boolean
+                              labelSelector:
+                                type: string
+                              limit:
+                                type: integer
+                              name:
+                                type: string
+                              namespace:
+                                type: string
+                              scope:
+                                description: |-
+                                  Scope is the reference for parent of the resource to select.
+                                  For config items, the scope is the scraper id
+                                  For checks, it's canaries and
+                                  For components, it's topology.
+                                  It can either be a uuid or namespace/name
+                                type: string
+                              search:
+                                description: Search query that applies to the resource
+                                  name, tag & labels.
+                                type: string
+                              statuses:
+                                description: Statuses filter resources by the status
+                                items:
+                                  type: string
+                                type: array
+                              tagSelector:
+                                type: string
+                              types:
+                                description: Types filter resources by the type
+                                items:
+                                  type: string
+                                type: array
+                            type: object
+                          type: array
                         prompt:
                           description: Prompt is the human prompt
                           type: string

config/crds/mission-control.flanksource.com_playbooks.yaml

@@ -38,6 +38,12 @@
         "dryRun": {
           "type": "boolean"
         },
+        "playbookAgents": {
+          "items": {
+            "$ref": "#/$defs/ResourceSelector"
+          },
+          "type": "array"
+        },
         "recommendPlaybooks": {
           "items": {
             "$ref": "#/$defs/ResourceSelector"

config/schemas/playbook-spec.schema.json

@@ -38,6 +38,12 @@
         "dryRun": {
           "type": "boolean"
         },
+        "playbookAgents": {
+          "items": {
+            "$ref": "#/$defs/ResourceSelector"
+          },
+          "type": "array"
+        },
         "recommendPlaybooks": {
           "items": {
             "$ref": "#/$defs/ResourceSelector"

config/schemas/playbook.schema.json

@@ -12,6 +12,10 @@ import (
 	"gorm.io/gorm/clause"
 )
 
+// events received on this channel are saved to DB.
+// NOTE: Not sure about this one. will probably remove it.
+var EventChan = make(chan models.Event)
+
 type asyncHandlerData struct {
 	fn           func(ctx context.Context, e models.Events) models.Events
 	batchSize    int
@@ -67,6 +71,15 @@ func StartConsumers(ctx context.Context) {
 	for _, fn := range registers {
 		fn(ctx)
 	}
+
+	go func() {
+		for e := range EventChan {
+			if err := ctx.DB().Create(&e).Error; err != nil {
+				log.Errorf("failed to create event: %w", err)
+			}
+		}
+	}()
+
 	// We listen to all PG Notifications on one channel and distribute it to other consumers
 	// based on the events.
 	notifyRouter := pg.NewNotifyRouter()

events/event_queue.go

@@ -0,0 +1,47 @@
+---
+# yaml-language-server: $schema=../../../config/schemas/playbook.schema.json
+apiVersion: mission-control.flanksource.com/v1
+kind: Playbook
+metadata:
+  name: use-agents
+spec:
+  description: Use AI to diagnose unhealthy config items and send a notification slack
+  category: AI
+  configs:
+    - name: '*'
+  parameters:
+    - name: prompt
+      label: Prompt
+      default: Find out why {{.config.name}} is unhealthy and report in one short sentence.
+      properties:
+        multiline: 'true'
+  actions:
+    - name: query LLM
+      ai:
+        connection: connection://mc/anthropic
+        playbookAgents:
+          - namespace: mc
+        systemPrompt: |
+          **Role:** Seasoned Kubernetes engineer and Diagnostic Expert
+
+          **Objective:** Assist users in diagnosing issues with unhealthy Kubernetes resources by analyzing provided manifests and related resources.
+
+          **Instructions:**
+
+          1. **Resource Analysis:** Examine the manifest of the unhealthy resource thoroughly.
+          2. **Contextual Investigation:** Consider additional related resources provided (e.g., pods, replica sets, namespaces) to gain a comprehensive understanding of the issue.
+          4. **One-Time Diagnosis:** Aim to diagnose the issue in a single response without requiring follow-up questions.
+        prompt: '{{.params.prompt}}'
+        changes:
+          since: 2d
+        analysis:
+          since: 2d
+        relationships:
+          - depth: 3
+            direction: outgoing
+            changes:
+              since: 24h
+          - depth: 5
+            direction: incoming
+            changes:
+              since: 24h

fixtures/playbooks/ai/ai-agents.yaml

@@ -1,11 +1,12 @@
 ---
-# yaml-language-server: $schema=../../config/schemas/playbook.schema.json
+# yaml-language-server: $schema=../../../config/schemas/playbook.schema.json
 apiVersion: mission-control.flanksource.com/v1
 kind: Playbook
 metadata:
   name: diagnose-aws-resource
 spec:
   description: Use AI to diagnose AWS resources
+  category: AI
   configs:
     - types:
         - AWS::CloudFormation::Stack

…es/playbooks/ai-diagnose-aws-ollama.yaml …playbooks/ai/ai-diagnose-aws-ollama.yamlfixtures/playbooks/ai-diagnose-aws-ollama.yaml renamed to fixtures/playbooks/ai/ai-diagnose-aws-ollama.yaml fixtures/playbooks/ai-diagnose-aws-ollama.yaml renamed to fixtures/playbooks/ai/ai-diagnose-aws-ollama.yaml

@@ -1,11 +1,12 @@
 ---
-# yaml-language-server: $schema=../../config/schemas/playbook.schema.json
+# yaml-language-server: $schema=../../../config/schemas/playbook.schema.json
 apiVersion: mission-control.flanksource.com/v1
 kind: Playbook
 metadata:
   name: diagnose-kubernetes-resource
 spec:
   description: Use AI to diagnose unhealthy kubernetes resources
+  category: AI
   configs:
     - types:
         - Kubernetes::DaemonSet

…oks/ai-diagnose-kubernetes-resource.yaml …/ai/ai-diagnose-kubernetes-resource.yamlfixtures/playbooks/ai-diagnose-kubernetes-resource.yaml renamed to fixtures/playbooks/ai/ai-diagnose-kubernetes-resource.yaml fixtures/playbooks/ai-diagnose-kubernetes-resource.yaml renamed to fixtures/playbooks/ai/ai-diagnose-kubernetes-resource.yaml

@@ -1,13 +1,14 @@
 ---
-# yaml-language-server: $schema=../../config/schemas/playbook.schema.json
+# yaml-language-server: $schema=../../../config/schemas/playbook.schema.json
 apiVersion: mission-control.flanksource.com/v1
 kind: Playbook
 metadata:
   name: diagnose-and-send-to-slack
 spec:
   description: Use AI to diagnose unhealthy config items and send a notification slack
+  category: AI
   configs:
-    - health: unhealthy,warning
+    - health: unhealthy,warning,unknown,healthy
   parameters:
     - name: prompt
       label: Prompt