lichenblankie/tinyweb
1
0
Fork 0
Code Issues Pull requests Projects 1 Releases 1 Packages Wiki Activity Actions

fixed reindex, preserved summaries

Browse source 
Previously reindex skipped pages that already had chunks, leaving stale
embeddings in place. It also overwrote good meta description summaries
with auto-generated ones. Now it clears all chunks first so everything
is re-embedded, and only generates summaries for pages missing one.
This commit is contained in:
lichenblankie 2026-03-27 14:08:04 -07:00
parent cf536a860c
commit 3f8ebdab1d
1 changed files with 8 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

24
embeddings.py
View file

@ -506,21 +506,24 @@ def hybrid_search(query_text, bm25_ranked_ids, limit=10, db=None, use_reranker=F
# --------------------------------------------------------------------------- # ---------------------------------------------------------------------------
def reindex_all(db=None, progress_callback=None): def reindex_all(db=None, progress_callback=None):
"""Embed all pages that don't yet have chunks. Also generates missing summaries. Rebuilds HNSW index.""" """Re-embed all pages and regenerate all summaries. Rebuilds HNSW index."""
from db import get_db, return_db, _generate_summary from db import get_db, return_db, _generate_summary
own_db = db is None own_db = db is None
if own_db: if own_db:
db = get_db() db = get_db()
try: try:
# Clear existing chunks so everything is regenerated
db.execute("DELETE FROM chunks")
db.commit()
rows = db.execute( rows = db.execute(
"SELECT p.id, p.title, p.body, p.summary FROM pages p " "SELECT p.id, p.title, p.body, p.summary FROM pages p"
"WHERE p.id NOT IN (SELECT DISTINCT page_id FROM chunks WHERE page_id IS NOT NULL)"
).fetchall() ).fetchall()
total = len(rows) total = len(rows)
for i, row in enumerate(rows): for i, row in enumerate(rows):
store_embeddings(row["id"], row["title"], row["body"], db) store_embeddings(row["id"], row["title"], row["body"], db)
# Generate summary if missing # Only regenerate summary if missing
if not row["summary"]: if not row["summary"]:
summary = _generate_summary(row["title"], row["body"]) summary = _generate_summary(row["title"], row["body"])
db.execute("UPDATE pages SET summary = ? WHERE id = ?", (summary, row["id"])) db.execute("UPDATE pages SET summary = ? WHERE id = ?", (summary, row["id"]))
@ -528,20 +531,9 @@ def reindex_all(db=None, progress_callback=None):
if progress_callback: if progress_callback:
progress_callback(i + 1, total) progress_callback(i + 1, total)
# Generate summaries for pages that already have chunks but no summary
no_summary = db.execute(
"SELECT id, title, body FROM pages WHERE summary = '' OR summary IS NULL"
).fetchall()
for row in no_summary:
summary = _generate_summary(row["title"], row["body"])
db.execute("UPDATE pages SET summary = ? WHERE id = ?", (summary, row["id"]))
if no_summary:
db.commit()
# Also handle remote pages # Also handle remote pages
remote_rows = db.execute( remote_rows = db.execute(
"SELECT rp.id, rp.title, rp.note FROM remote_pages rp " "SELECT rp.id, rp.title, rp.note FROM remote_pages rp"
"WHERE rp.id NOT IN (SELECT DISTINCT remote_page_id FROM chunks WHERE remote_page_id IS NOT NULL)"
).fetchall() ).fetchall()
for rp in remote_rows: for rp in remote_rows: