Backup automatico script del 2026-09-27 07:00

services/loogle-mcp/app/knowledge/indexer.py

@@ -2,10 +2,15 @@
 """Paperless → Qdrant indexing pipeline."""
 
 import logging
+import zlib
+from concurrent.futures import ThreadPoolExecutor
 from typing import Optional
 
 LOGGER = logging.getLogger("loogle_mcp.indexer")
 
+# Indicizzazioni avviate dai tool MCP: fuori dalla richiesta, una alla volta (thermal gate)
+_background = ThreadPoolExecutor(max_workers=1, thread_name_prefix="bg-index")
+
 from ..db import get_conn
 from . import embeddings, gitea_indexer, paperless, qdrant_store
 from .text_chunk import chunk_text as _chunk_text
@@ -104,7 +109,8 @@ def index_context_snippet(
     doc_key = f"ctx-{username}-{project_id}"
     if snippet_id:
         doc_key = f"{doc_key}-{snippet_id}"
-    doc_id = hash(doc_key) % (2**31)
+    # crc32 e non hash(): hash() cambia a ogni processo e il delete non trovava i vecchi chunk
+    doc_id = zlib.crc32(doc_key.encode()) % (2**31)
     qdrant_store.delete_by_doc(collection, doc_id)
     ids = []
     payloads = []
@@ -127,12 +133,82 @@ def index_context_snippet(
     qdrant_store.upsert_chunks(collection, ids, vectors, payloads)
 
 
+def index_project_context(username: str, project_id: str) -> dict:
+    """Indicizza il context.md completo del progetto sostituendo la versione precedente.
+
+    Personali → ctx_{utente}; condivisi → ctx_shared_family (visibili a tutta la famiglia).
+    Così un contesto che cambia (es. inventario) non lascia in ricerca informazioni superate.
+    """
+    from ..context import store as context_store
+
+    data = context_store.get_project_context(username, project_id, session_limit=0, include_gitea=False)
+    meta = data["meta"]
+    shared = bool(meta.get("shared"))
+    collection = qdrant_store.CTX_SHARED_COLLECTION if shared else qdrant_store.ctx_collection(username)
+    namespace = context_store.SHARED_NS if shared else username
+    drop_project_context(collection, project_id)
+    chunks = _chunk_text(data["context_md"])
+    if not chunks:
+        return {"project_id": project_id, "collection": collection, "chunks": 0}
+    vectors = embeddings.embed_texts(chunks)
+    doc_key = f"ctx-{namespace}-{project_id}"
+    doc_id = zlib.crc32(doc_key.encode()) % (2**31)
+    ids, payloads = [], []
+    for i, chunk in enumerate(chunks):
+        ids.append(f"{doc_key}-chunk-{i}")
+        payloads.append(
+            {
+                "doc_id": doc_id,
+                "project_id": project_id,
+                "chunk_index": i,
+                "title": meta.get("title") or project_id,
+                "text": chunk,
+                "owner": meta.get("owner") or username,
+                "visibility": "family" if shared else "personal",
+                "shared": shared,
+                "updated_by": meta.get("updated_by"),
+                "source": "context",
+                "snippet_id": "full",
+            }
+        )
+    qdrant_store.upsert_chunks(collection, ids, vectors, payloads)
+    return {"project_id": project_id, "collection": collection, "chunks": len(chunks)}
+
+
+_DIGEST_SNIPPET_PREFIXES = ("daily-", "plan-")
+
+
+def drop_project_context(collection: str, project_id: str) -> int:
+    """Rimuove tutti i chunk di un progetto (anche quelli indicizzati a pezzi in passato),
+    tranne gli snippet del digest giornaliero (snippet_id daily-*/plan-*)."""
+    return qdrant_store.delete_matching(
+        collection,
+        "project_id",
+        project_id,
+        lambda payload: not str(payload.get("snippet_id") or "").startswith(_DIGEST_SNIPPET_PREFIXES),
+    )
+
+
+def submit_background(label: str, fn, *args, **kwargs) -> None:
+    """Esegue fn in background; gli errori finiscono nel log invece che al client."""
+
+    def _run():
+        try:
+            fn(*args, **kwargs)
+            LOGGER.info("Indicizzazione background completata: %s", label)
+        except Exception:
+            LOGGER.exception("Indicizzazione background fallita: %s", label)
+
+    _background.submit(_run)
+
+
 def search_knowledge(username: str, query: str, limit: int = 8, is_admin: bool = False) -> list:
-    vectors = embeddings.embed_texts([query])
+    vectors = [embeddings.embed_query(query)]
     collections = [
         qdrant_store.SHARED_COLLECTION,
         qdrant_store.kb_collection(username),
         qdrant_store.ctx_collection(username),
+        qdrant_store.CTX_SHARED_COLLECTION,
     ]
     collections.extend(gitea_indexer.gitea_collections_for_user(username, is_admin=is_admin))
     collections.append(qdrant_store.APPS_SHARED_COLLECTION)
@@ -156,8 +232,9 @@ def search_gitea_knowledge(username: str, query: str, limit: int = 8, is_admin:
 
 
 def search_context(username: str, query: str, limit: int = 8) -> list:
-    vectors = embeddings.embed_texts([query])
-    return qdrant_store.search([qdrant_store.ctx_collection(username)], vectors[0], limit=limit)
+    vectors = [embeddings.embed_query(query)]
+    collections = [qdrant_store.ctx_collection(username), qdrant_store.CTX_SHARED_COLLECTION]
+    return qdrant_store.search(collections, vectors[0], limit=limit)
 
 
 def list_recent_documents(limit: int = 20) -> list: