fix(kb): recreate a missing Qdrant container instead of spinning on Start Qdrant

The Knowledge Base's "Start Qdrant" button called start on a container
that no longer existed. The server answered 200 with success: false, the
modal ignored that flag, and the button polled the health check forever.

Starting a dependency service whose container is gone now recreates it
from its stored config through the normal install path. Its data lives on
a host bind mount, so nothing is lost. The modal shows a refused start as
an error, and stops waiting after five minutes if Qdrant never comes
online.

Fixes #1426

admin/app/services/docker_service.ts

@@ -5,6 +5,7 @@ import { inject } from '@adonisjs/core'
 import transmit from '@adonisjs/transmit/services/main'
 import { doResumableDownloadWithRetry } from '../utils/downloads.js'
 import { pickAmdHsaOverride } from '../utils/amd_hsa_override.js'
+import { recreatesOnStart } from '../utils/dependency_start.js'
 import { join } from 'path'
 import os from 'node:os'
 import env from '#start/env'
@@ -92,7 +93,7 @@ export class DockerService {
   ): Promise<{ success: boolean; message: string }> {
     try {
       const service = await Service.query().where('service_name', serviceName).first()
-      if (!service || !service.installed) {
+      if (!service) {
         return {
           success: false,
           message: `Service ${serviceName} not found or not installed`,
@@ -101,6 +102,24 @@ export class DockerService {
 
       const containers = await this.docker.listContainers({ all: true })
       const container = containers.find((c) => c.Names.includes(`/${serviceName}`))
+
+      if (recreatesOnStart(action, service.is_dependency_service, !!container)) {
+        logger.info(`[DockerService] ${serviceName} has no container. Recreating it.`)
+        // Preflight refuses a row still marked installed; the next sync would reset it anyway.
+        if (service.installed) {
+          service.installed = false
+          await service.save()
+        }
+        return await this.createContainerPreflight(serviceName)
+      }
+
+      if (!service.installed) {
+        return {
+          success: false,
+          message: `Service ${serviceName} not found or not installed`,
+        }
+      }
+
       if (!container) {
         return {
           success: false,

admin/app/utils/dependency_start.ts

@@ -0,0 +1,18 @@
+/**
+ * Whether a "start" request should recreate the service's container instead of
+ * starting it (#1426).
+ *
+ * A hidden dependency service (Qdrant) has no card on the Apps page, so the
+ * Knowledge Base's "Start Qdrant" button is the only control a user has over it.
+ * When its container is gone there is nothing to start, and the next container sync
+ * marks the row not installed, so the only way back is to recreate it from the
+ * stored config. Its data lives on a host bind mount, so recreating it loses nothing.
+ * Every other service keeps the old answer: the user reinstalls it from the Apps page.
+ */
+export function recreatesOnStart(
+  action: 'start' | 'stop' | 'restart',
+  isDependencyService: boolean,
+  containerExists: boolean
+): boolean {
+  return action === 'start' && isDependencyService && !containerExists
+}

admin/inertia/components/chat/KnowledgeBaseModal.tsx

@@ -55,6 +55,10 @@ const VIEWABLE_EXTENSIONS = new Set([
   'html',
 ])
 
+// How long "Start Qdrant" waits for the health check to come back online. Long
+// enough for a recreate that has to pull the image again.
+const QDRANT_START_TIMEOUT_MS = 5 * 60_000
+
 function isViewableExtension(filename: string): boolean {
   const ext = filename.split('.').at(-1)?.toLowerCase() ?? ''
   return VIEWABLE_EXTENSIONS.has(ext)
@@ -228,6 +232,20 @@ export default function KnowledgeBaseModal({
     if (!qdrantOffline) setIsStartingQdrant(false)
   }, [qdrantOffline])
 
+  // Stop waiting on a Qdrant that started and then crashed, or whose recreate failed.
+  useEffect(() => {
+    if (!isStartingQdrant) return
+    const timer = setTimeout(() => {
+      setIsStartingQdrant(false)
+      addNotification({
+        type: 'error',
+        message:
+          'Qdrant did not come online. Check its logs under Settings → Service Logs & Metrics.',
+      })
+    }, QDRANT_START_TIMEOUT_MS)
+    return () => clearTimeout(timer)
+  }, [isStartingQdrant])
+
   const { data: storedFiles = [], isLoading: isLoadingFiles } = useQuery({
     queryKey: ['storedFiles'],
     queryFn: () => api.getStoredRAGFiles(),
@@ -399,7 +417,12 @@ export default function KnowledgeBaseModal({
 
   const startQdrantMutation = useMutation({
     mutationFn: () => api.affectService(SERVICE_NAMES.QDRANT, 'start'),
-    onSuccess: () => {
+    onSuccess: (result) => {
+      // A refused start comes back as success: false, not as a thrown error.
+      if (!result?.success) {
+        addNotification({ type: 'error', message: result?.message || 'Failed to start Qdrant.' })
+        return
+      }
       setIsStartingQdrant(true)
       queryClient.invalidateQueries({ queryKey: ['qdrantHealth'] })
     },

admin/tests/unit/dependency_start.spec.ts

@@ -0,0 +1,21 @@
+import * as assert from 'node:assert/strict'
+import { test } from 'node:test'
+
+import { recreatesOnStart } from '../../app/utils/dependency_start.js'
+
+test('starting a dependency service whose container is gone recreates it', () => {
+  assert.equal(recreatesOnStart('start', true, false), true)
+})
+
+test('starting a dependency service whose container exists just starts it', () => {
+  assert.equal(recreatesOnStart('start', true, true), false)
+})
+
+test('stop and restart never recreate a missing dependency container', () => {
+  assert.equal(recreatesOnStart('stop', true, false), false)
+  assert.equal(recreatesOnStart('restart', true, false), false)
+})
+
+test('starting a user-facing app with no container does not recreate it', () => {
+  assert.equal(recreatesOnStart('start', false, false), false)
+})