fix: upsert in bounded chunks, recreate Qdrant indexes on rebuild, larger maintenance tmpfs

backend/src/workspaces/preprocessing-service.ts

@@ -11,6 +11,7 @@ import {
 } from "./preprocessing-state.js";
 import type { DeterministicRuntimeConfigLease } from "./runtime-config-lease.js";
 import { readAnnotationsSync } from "./annotations-sync.js";
+import { reconcileCollection } from "./qdrant-collection.js";
 
 export interface WorkspaceOperationResult {
   schemaVersion: 1;
@@ -150,12 +151,15 @@ export class WorkspacePreprocessingService {
     const q = `${runtime.configLease.semanticQdrantUrl}/collections/${encodeURIComponent(collection)}`;
     const del = await fetch(q, { method: "DELETE" });
     if (!del.ok && del.status !== 404) return baseResult(runtime, "vector rebuild", "failed", "semantic_index_incompatible", { warnings: ["collection delete failed"] });
-    const put = await fetch(q, {
-      method: "PUT",
-      headers: { "content-type": "application/json" },
-      body: JSON.stringify({ vectors: { size: runtime.workspace.semantic_index.vector_store.dimensions, distance: runtime.workspace.semantic_index.vector_store.distance } }),
+    // Recreate the complete contract (dimensions + distance + the 8 required keyword indexes).
+    const recreated = await reconcileCollection({
+      baseUrl: runtime.configLease.semanticQdrantUrl,
+      collection,
+      dimensions: runtime.workspace.semantic_index.vector_store.dimensions,
+      distance: runtime.workspace.semantic_index.vector_store.distance,
+      mode: "self_heal",
     });
-    if (!put.ok) return baseResult(runtime, "vector rebuild", "failed", "semantic_index_incompatible", { warnings: ["collection recreate failed"] });
+    if (!recreated.ok) return baseResult(runtime, "vector rebuild", "failed", "semantic_index_incompatible", { warnings: ["collection recreate failed"] });
     return baseResult(runtime, "vector rebuild", "succeeded", "ok", { warnings: [`recreated collection=${collection}`] });
   }
   async inspect(options: { workspaceId: string }): Promise<WorkspaceOperationResult> {

backend/test/workspace-preprocessing-service.test.ts

@@ -123,6 +123,7 @@ function runtime(workspace = baseWorkspace, workspaceId = workspace.workspace.id
       catalogBlob: "c".repeat(40),
       configDigest: "sha256:config",
       bindingDigest: "sha256:bindings",
+      semanticQdrantUrl: "http://qdrant:6333",
       effectiveConfig: {
         schemaVersion: 1,
         dwh: {
@@ -491,3 +492,36 @@ test("full runs continue after schema accept only when the accepted blob matches
   const stillBlocked = await second.service.run({ workspaceId: "psd-clinical", resumeRunId: secondRunId });
   expect(stillBlocked).toMatchObject({ status: "blocked", code: "manual_review_required" });
 });
+
+test("vector rebuild recreates the full collection contract including keyword indexes", async () => {
+  const f = fixture();
+  // mock fetch: DELETE ok, then reconcileCollection self-heals create + indexes (real fetch in deps)
+  const calls: string[] = [];
+  const fakeFetch = async (url: string, init?: any) => {
+    calls.push(`${init?.method ?? "GET"} ${url}`);
+    if ((init?.method ?? "GET") === "DELETE") return new Response("", { status: 200 });
+    if (url.endsWith("/collections/psd-clinical") && init?.method === "PUT") return new Response("", { status: 200 });
+    if (url.endsWith("/collections/psd-clinical") && init?.method === "GET") {
+      return new Response(JSON.stringify({ result: { config: { params: { vectors: { size: 1024, distance: "Cosine" } } }, payload_schema: { content_hash: { data_type: "keyword" }, document_id: { data_type: "keyword" }, kind: { data_type: "keyword" }, record_key: { data_type: "keyword" }, record_kind: { data_type: "keyword" }, vector_generation: { data_type: "keyword" }, workspace_id: { data_type: "keyword" }, workspace_revision: { data_type: "keyword" } } } }), { status: 200 });
+    }
+    if (url.endsWith("/collections/psd-clinical/index") && init?.method === "PUT") return new Response("", { status: 200 });
+    return new Response(JSON.stringify({ result: {} }), { status: 200 });
+  };
+  const service = new WorkspacePreprocessingService({
+    dataRoot: f.dataRoot,
+    acquireActiveRuntime: async () => runtime(baseWorkspace),
+    runChild: vi.fn(),
+    listSessions: async () => [],
+    semanticPreflight: async () => ({ ok: true }),
+  });
+  // replace global fetch used by vectorRebuild/reconcileCollection
+  const original = globalThis.fetch;
+  globalThis.fetch = fakeFetch as any;
+  try {
+    const result = await service.vectorRebuild({ workspaceId: "psd-clinical", collection: "psd-clinical", confirm: "psd-clinical", destroy: true });
+    expect(result).toMatchObject({ status: "succeeded", code: "ok" });
+  } finally {
+    globalThis.fetch = original;
+  }
+  expect(calls.some((c) => c.startsWith("DELETE "))).toBe(true);
+});