corpus integrity: ingest_failures tracking in watcher, reconciliation script, corpus status/retry/reconcile endpoints

This commit is contained in:
2026-04-30 21:54:39 +00:00
parent 655dea6ae5
commit 74e2c34f43
3 changed files with 439 additions and 0 deletions
+173
View File
@@ -957,6 +957,179 @@ async def clear_all_conversations(auth: str = Depends(require_auth)):
return JSONResponse({"cleared": True})
# ─── Corpus Integrity Endpoints ─────────────────────────────────────────────
CORPUS_INTEGRITY_SCRIPT = str(Path.home() / "aaronai" / "scripts" / "corpus_integrity.py")
CORPUS_REPORT_PATH = Path.home() / "aaronai" / "corpus_integrity_report.json"
SUPPORTED_EXTS = {".pdf", ".docx", ".pptx", ".txt", ".md"}
MIGRATION_STATE_PATH = Path.home() / "aaronai" / "experiments" / "tier1_migration_state.json"
def get_corpus_status_data():
fs_count = 0
try:
root = Path(NEXTCLOUD_PATH)
for path in root.rglob("*"):
if path.is_file() and path.suffix.lower() in SUPPORTED_EXTS:
if path.name.startswith((".", "~$")): continue
if "Admin/Backups" in str(path) or "Backups" in path.parts: continue
if "Journal/Media" in str(path): continue
fs_count += 1
except Exception:
pass
pv_count = 0
try:
pg = get_pg()
cur = pg.cursor()
cur.execute("SELECT COUNT(DISTINCT source) FROM embeddings WHERE source IS NOT NULL")
pv_count = cur.fetchone()[0]
pg.close()
except Exception:
pass
gr_sources = set()
try:
if MIGRATION_STATE_PATH.exists():
state = json.loads(MIGRATION_STATE_PATH.read_text())
for fp in state.get("ingested", []):
gr_sources.add(Path(fp).name)
except Exception:
pass
try:
pg = get_pg()
cur = pg.cursor()
cur.execute("SELECT DISTINCT source FROM stage_3_queue WHERE completed_at IS NOT NULL")
for row in cur.fetchall(): gr_sources.add(row[0])
pg.close()
except Exception:
pass
failures = []
try:
pg = get_pg()
cur = pg.cursor()
cur.execute("""
SELECT source, filepath, error, retry_count, first_failed_at, last_failed_at
FROM ingest_failures WHERE resolved = FALSE
ORDER BY last_failed_at DESC LIMIT 50
""")
for row in cur.fetchall():
failures.append({
"source": row[0], "filepath": row[1], "error": row[2],
"retry_count": row[3], "first_failed_at": str(row[4]),
"last_failed_at": str(row[5]),
})
pg.close()
except Exception:
pass
last_report = None
try:
if CORPUS_REPORT_PATH.exists():
report = json.loads(CORPUS_REPORT_PATH.read_text())
last_report = {
"timestamp": report.get("timestamp"),
"gaps": report.get("summary", {}).get("neither", 0),
"auto_queued": len(report.get("auto_queued", [])),
}
except Exception:
pass
return {
"filesystem": fs_count,
"pgvector": pv_count,
"graphiti": len(gr_sources),
"failures": failures,
"failure_count": len(failures),
"last_reconciliation": last_report,
}
@app.get("/api/corpus/status")
async def corpus_status(auth: str = Depends(require_auth)):
try:
return JSONResponse(get_corpus_status_data())
except Exception as e:
return JSONResponse({"error": str(e)}, status_code=500)
@app.post("/api/corpus/retry")
async def corpus_retry(request: Request, auth: str = Depends(require_auth)):
try:
body = await request.json()
source = body.get("source", "")
if not source:
return JSONResponse({"error": "source required"}, status_code=400)
pg = get_pg()
cur = pg.cursor()
cur.execute("SELECT filepath FROM ingest_failures WHERE source = %s", (source,))
row = cur.fetchone()
pg.close()
if not row:
return JSONResponse({"error": "source not found in failures"}, status_code=404)
filepath = Path(row[0])
if not filepath.exists():
return JSONResponse({"error": f"file not found: {filepath}"}, status_code=404)
suffix = filepath.suffix.lower()
text = ""
try:
if suffix in {".txt", ".md"}:
text = filepath.read_text(encoding="utf-8", errors="ignore")
elif suffix == ".pdf":
from pypdf import PdfReader
text = "".join(p.extract_text() + "\n" for p in PdfReader(filepath).pages if p.extract_text())
elif suffix == ".docx":
from docx import Document as DocxDocument
text = "\n".join(p.text for p in DocxDocument(filepath).paragraphs if p.text.strip())
elif suffix == ".pptx":
from pptx import Presentation
prs = Presentation(filepath)
text = "\n".join(shape.text for slide in prs.slides for shape in slide.shapes
if hasattr(shape, "text") and shape.text.strip())
except Exception as e:
return JSONResponse({"error": f"extraction failed: {e}"}, status_code=500)
if not text.strip():
return JSONResponse({"error": "file produces empty text — may be corrupt"}, status_code=422)
pg = get_pg()
cur = pg.cursor()
cur.execute("""
INSERT INTO stage_2_queue (source, full_text, char_length)
VALUES (%s, %s, %s)
ON CONFLICT (source) DO UPDATE SET
full_text = EXCLUDED.full_text, char_length = EXCLUDED.char_length,
enqueued_at = NOW(), completed_at = NULL, failed_at = NULL, attempts = 0
""", (source, text[:50000], len(text)))
cur.execute("""
UPDATE ingest_failures SET retry_count = retry_count + 1, last_failed_at = NOW()
WHERE source = %s
""", (source,))
pg.commit()
pg.close()
return JSONResponse({"queued": True, "source": source})
except Exception as e:
return JSONResponse({"error": str(e)}, status_code=500)
@app.post("/api/corpus/reconcile")
async def corpus_reconcile(request: Request, background_tasks: BackgroundTasks, auth: str = Depends(require_auth)):
try:
body = await request.json()
fix = body.get("fix", True)
except Exception:
fix = True
def run_reconcile():
try:
cmd = [PYTHON, CORPUS_INTEGRITY_SCRIPT]
if fix:
cmd.append("--fix")
subprocess.run(cmd, cwd=str(Path.home() / "aaronai"), timeout=300)
except Exception as e:
print(f"Reconciliation failed: {e}")
background_tasks.add_task(run_reconcile)
return JSONResponse({"started": True, "fix": fix})
# ─── Scheduler ──────────────────────────────────────────────────────────────
scheduler = BackgroundScheduler()