389 lines
14 KiB
Python
389 lines
14 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Obsidian Vault Reindex Endpoint
|
|
Lightweight HTTP server that triggers incremental or full Obsidian vault reindex.
|
|
|
|
Listens on 0.0.0.0:18810 (configurable via PORT env var).
|
|
Called by n8n webhooks or systemd timers.
|
|
|
|
Endpoints:
|
|
POST /reindex -> trigger incremental reindex, returns JSON stats
|
|
POST /reindex?full=true -> trigger full semantic Chroma rebuild
|
|
GET /reindex/status -> check last index state
|
|
GET /semantic-health -> verify state plus semantic search smoke check
|
|
POST /semantic-search -> query the Obsidian Chroma semantic index
|
|
GET /healthz -> returns ok
|
|
"""
|
|
|
|
import http.server
|
|
import json
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
import threading
|
|
import time
|
|
from pathlib import Path
|
|
from urllib.parse import parse_qs, urlparse
|
|
from urllib import request, error
|
|
|
|
PORT = int(os.environ.get("PORT", 18810))
|
|
REINDEX_TIMEOUT = int(os.environ.get("REINDEX_TIMEOUT", "1800"))
|
|
RAG_COLLECTION = os.environ.get("RAG_COLLECTION", "obsidian").strip() or "obsidian"
|
|
RAG_EMBED_MODEL = os.environ.get("RAG_EMBED_MODEL", "nomic-embed-text").strip() or "nomic-embed-text"
|
|
OLLAMA_BASE_URL = (os.environ.get("OLLAMA_BASE_URL") or "http://127.0.0.1:18807").rstrip("/")
|
|
RAG_RERANK_ENABLED = (os.environ.get("RAG_RERANK_ENABLED") or "false").strip().lower() in {
|
|
"1",
|
|
"true",
|
|
"yes",
|
|
"on",
|
|
}
|
|
RAG_RERANK_URL = (os.environ.get("RAG_RERANK_URL") or "http://127.0.0.1:18818/rerank").strip()
|
|
RAG_RERANK_INITIAL_K = max(1, int(os.environ.get("RAG_RERANK_INITIAL_K") or "20"))
|
|
RAG_RERANK_TOP_K = max(1, int(os.environ.get("RAG_RERANK_TOP_K") or "5"))
|
|
RAG_RERANK_TIMEOUT_MS = max(1, int(os.environ.get("RAG_RERANK_TIMEOUT_MS") or "3000"))
|
|
RAG_RERANK_REQUIRE_NPU_PROOF = (os.environ.get("RAG_RERANK_REQUIRE_NPU_PROOF") or "true").strip().lower() in {
|
|
"1",
|
|
"true",
|
|
"yes",
|
|
"on",
|
|
}
|
|
|
|
REINDEX_SCRIPT = str(
|
|
Path.home()
|
|
/ ".hermes/skills/note-taking/rag-search/scripts/reindex_obsidian.sh"
|
|
)
|
|
STATE_FILE = Path(
|
|
os.environ.get("RAG_STATE_FILE")
|
|
or Path.home() / ".hermes/data/rag-search" / (
|
|
"obsidian_index_state.json" if RAG_COLLECTION == "obsidian" else f"{RAG_COLLECTION}_index_state.json"
|
|
)
|
|
).expanduser()
|
|
SEARCH_SCRIPT = str(Path.home() / ".hermes/skills/note-taking/rag-search/scripts/search.py")
|
|
VENV_PYTHON = str(Path.home() / ".hermes/skills/note-taking/rag-search/venv/bin/python")
|
|
|
|
# Lock to prevent concurrent reindexing
|
|
_reindex_lock = threading.Lock()
|
|
|
|
|
|
def run_reindex(full: bool = False) -> dict:
|
|
"""Run the reindex script. Returns stats dict."""
|
|
if not _reindex_lock.acquire(blocking=False):
|
|
return {"error": "reindex already in progress", "status": "locked"}
|
|
|
|
try:
|
|
cmd = [REINDEX_SCRIPT]
|
|
if full:
|
|
cmd.append("--full")
|
|
env = os.environ.copy()
|
|
env.setdefault("RAG_COLLECTION", RAG_COLLECTION)
|
|
env.setdefault("RAG_EMBED_MODEL", RAG_EMBED_MODEL)
|
|
env.setdefault("OLLAMA_BASE_URL", OLLAMA_BASE_URL)
|
|
result = subprocess.run(
|
|
cmd,
|
|
capture_output=True,
|
|
text=True,
|
|
timeout=REINDEX_TIMEOUT,
|
|
env=env,
|
|
)
|
|
if result.returncode != 0:
|
|
return {
|
|
"error": "reindex failed",
|
|
"exit_code": result.returncode,
|
|
"stderr": result.stderr.strip()[-2000:],
|
|
}
|
|
try:
|
|
payload = json.loads(result.stdout)
|
|
if result.stderr.strip():
|
|
payload["progress_log_tail"] = result.stderr.strip()[-2000:]
|
|
return payload
|
|
except json.JSONDecodeError:
|
|
return {
|
|
"error": "invalid json output",
|
|
"stdout": result.stdout.strip()[:500],
|
|
"stderr": result.stderr.strip()[-2000:],
|
|
}
|
|
except subprocess.TimeoutExpired:
|
|
return {"error": f"reindex timed out ({REINDEX_TIMEOUT}s)"}
|
|
except Exception as e:
|
|
return {"error": str(e)}
|
|
finally:
|
|
_reindex_lock.release()
|
|
|
|
|
|
def get_status() -> dict:
|
|
"""Read the last index state file."""
|
|
if not STATE_FILE.exists():
|
|
return {"indexed": False, "message": "no state file"}
|
|
try:
|
|
return json.loads(STATE_FILE.read_text())
|
|
except (json.JSONDecodeError, IOError) as e:
|
|
return {"error": str(e)}
|
|
|
|
|
|
def _result_text(result: dict) -> str:
|
|
"""Return the text field sent to the reranker without changing response shape."""
|
|
return str(result.get("text") or result.get("content") or "")
|
|
|
|
|
|
def _apply_rerank(query: str, results: list[dict], final_k: int) -> tuple[list[dict], dict]:
|
|
"""Optionally rerank semantic results, falling back to vector order on any error."""
|
|
metadata = {
|
|
"enabled": RAG_RERANK_ENABLED,
|
|
"attempted": False,
|
|
"ok": False,
|
|
"url": RAG_RERANK_URL,
|
|
"initial_k": len(results),
|
|
"top_k": final_k,
|
|
}
|
|
if not RAG_RERANK_ENABLED:
|
|
metadata["ok"] = True
|
|
metadata["reason"] = "disabled"
|
|
return results[:final_k], metadata
|
|
if not results:
|
|
metadata["ok"] = True
|
|
metadata["reason"] = "no_results"
|
|
return [], metadata
|
|
|
|
metadata["attempted"] = True
|
|
documents = []
|
|
for idx, item in enumerate(results):
|
|
text = _result_text(item)
|
|
if not text:
|
|
continue
|
|
documents.append(
|
|
{
|
|
"id": str(item.get("id") or idx),
|
|
"text": text,
|
|
"metadata": {
|
|
"index": idx,
|
|
"path": item.get("path"),
|
|
"source": item.get("source"),
|
|
"chunk": item.get("chunk"),
|
|
},
|
|
}
|
|
)
|
|
if not documents:
|
|
metadata["ok"] = True
|
|
metadata["reason"] = "no_text_documents"
|
|
return results[:final_k], metadata
|
|
|
|
started = time.monotonic()
|
|
try:
|
|
body = json.dumps(
|
|
{
|
|
"query": query,
|
|
"documents": documents,
|
|
"top_k": final_k,
|
|
"return_documents": False,
|
|
}
|
|
).encode("utf-8")
|
|
req = request.Request(
|
|
RAG_RERANK_URL,
|
|
data=body,
|
|
headers={"Content-Type": "application/json"},
|
|
method="POST",
|
|
)
|
|
with request.urlopen(req, timeout=RAG_RERANK_TIMEOUT_MS / 1000.0) as resp:
|
|
payload = json.loads(resp.read().decode("utf-8"))
|
|
except (OSError, TimeoutError, json.JSONDecodeError, error.URLError, error.HTTPError) as exc:
|
|
metadata["duration_ms"] = round((time.monotonic() - started) * 1000, 2)
|
|
metadata["error"] = f"{type(exc).__name__}: {exc}"
|
|
return results[:final_k], metadata
|
|
|
|
metadata["duration_ms"] = round((time.monotonic() - started) * 1000, 2)
|
|
metadata["ok"] = bool(payload.get("ok", True))
|
|
metadata["model"] = payload.get("model")
|
|
metadata["device"] = payload.get("device")
|
|
metadata["npu_busy_delta_us"] = payload.get("npu_busy_delta_us")
|
|
metadata["require_npu_proof"] = RAG_RERANK_REQUIRE_NPU_PROOF
|
|
metadata["input_count"] = payload.get("input_count")
|
|
ranked = payload.get("results") or []
|
|
if RAG_RERANK_REQUIRE_NPU_PROOF and int(payload.get("npu_busy_delta_us") or 0) <= 0:
|
|
metadata["ok"] = False
|
|
metadata["error"] = "reranker response lacked positive npu_busy_delta_us"
|
|
return results[:final_k], metadata
|
|
if not metadata["ok"] or not ranked:
|
|
metadata["error"] = payload.get("error") or "reranker returned no ranked results"
|
|
return results[:final_k], metadata
|
|
|
|
by_id = {str(item.get("id") or idx): item for idx, item in enumerate(results)}
|
|
reranked = []
|
|
for rank, ranked_item in enumerate(ranked):
|
|
source_item = None
|
|
if "id" in ranked_item:
|
|
source_item = by_id.get(str(ranked_item.get("id")))
|
|
if source_item is None and isinstance(ranked_item.get("index"), int):
|
|
idx = ranked_item["index"]
|
|
if 0 <= idx < len(results):
|
|
source_item = results[idx]
|
|
if source_item is None:
|
|
continue
|
|
merged = dict(source_item)
|
|
merged["rerank_score"] = ranked_item.get("score")
|
|
merged["rerank_rank"] = rank + 1
|
|
reranked.append(merged)
|
|
if len(reranked) >= final_k:
|
|
break
|
|
if not reranked:
|
|
metadata["ok"] = False
|
|
metadata["error"] = "reranker result IDs did not match search results"
|
|
return results[:final_k], metadata
|
|
return reranked, metadata
|
|
|
|
|
|
def run_semantic_search(query: str, top_k: int = 5) -> dict:
|
|
"""Query the local Obsidian Chroma index via the rag-search script."""
|
|
query = (query or "").strip()
|
|
if not query:
|
|
return {"ok": False, "error": "query is required", "results": []}
|
|
top_k = max(1, min(int(top_k or 5), 20))
|
|
search_k = max(top_k, min(RAG_RERANK_INITIAL_K, 100)) if RAG_RERANK_ENABLED else top_k
|
|
final_k = min(top_k, RAG_RERANK_TOP_K) if RAG_RERANK_ENABLED else top_k
|
|
env = os.environ.copy()
|
|
env.setdefault("RAG_COLLECTION", RAG_COLLECTION)
|
|
env.setdefault("RAG_EMBED_MODEL", RAG_EMBED_MODEL)
|
|
env.setdefault("OLLAMA_BASE_URL", OLLAMA_BASE_URL)
|
|
result = subprocess.run(
|
|
[
|
|
VENV_PYTHON if Path(VENV_PYTHON).exists() else sys.executable,
|
|
SEARCH_SCRIPT,
|
|
"--index",
|
|
RAG_COLLECTION,
|
|
"--top-k",
|
|
str(search_k),
|
|
"--raw",
|
|
query,
|
|
],
|
|
capture_output=True,
|
|
text=True,
|
|
timeout=90,
|
|
env=env,
|
|
)
|
|
if result.returncode != 0:
|
|
return {
|
|
"ok": False,
|
|
"query": query,
|
|
"top_k": top_k,
|
|
"search_k": search_k,
|
|
"error": result.stderr.strip()[-2000:] or result.stdout.strip()[-2000:],
|
|
"results": [],
|
|
"rerank": {
|
|
"enabled": RAG_RERANK_ENABLED,
|
|
"attempted": False,
|
|
"ok": False,
|
|
"error": "vector search failed before rerank",
|
|
},
|
|
}
|
|
payload = json.loads(result.stdout)
|
|
results = payload.get("results") or []
|
|
results, rerank_meta = _apply_rerank(query, results, final_k)
|
|
return {
|
|
"ok": True,
|
|
"query": query,
|
|
"index": payload.get("index", RAG_COLLECTION),
|
|
"top_k": top_k,
|
|
"search_k": search_k,
|
|
"result_count": len(results),
|
|
"rerank": rerank_meta,
|
|
"results": results,
|
|
}
|
|
|
|
|
|
def semantic_health() -> dict:
|
|
"""Return state plus a tiny semantic-search smoke check."""
|
|
status = get_status()
|
|
health = {
|
|
"status": "ok" if status.get("status") == "ok" and status.get("vector_count", 0) > 0 else "degraded",
|
|
"state": {
|
|
k: status.get(k)
|
|
for k in (
|
|
"status",
|
|
"note_count",
|
|
"vector_count",
|
|
"collection",
|
|
"embedding_backend",
|
|
"embedding_model",
|
|
"last_full_index",
|
|
"last_incremental_index",
|
|
)
|
|
},
|
|
}
|
|
try:
|
|
payload = run_semantic_search("Obsidian reindex", top_k=1)
|
|
health["search_ok"] = bool(payload.get("results"))
|
|
health["result_count"] = len(payload.get("results", []))
|
|
if not payload.get("ok"):
|
|
health["search_error"] = payload.get("error")
|
|
except Exception as e:
|
|
health["status"] = "degraded"
|
|
health["search_ok"] = False
|
|
health["search_error"] = str(e)
|
|
if not health.get("search_ok"):
|
|
health["status"] = "degraded"
|
|
return health
|
|
|
|
|
|
class ReindexHandler(http.server.BaseHTTPRequestHandler):
|
|
def do_GET(self):
|
|
path = urlparse(self.path).path.rstrip("/")
|
|
if path == "/healthz":
|
|
self._json_response({"status": "ok"})
|
|
elif path == "/reindex/status":
|
|
self._json_response(get_status())
|
|
elif path in ("/semantic-health", "/reindex/semantic-health"):
|
|
data = semantic_health()
|
|
self._json_response(data, status=200 if data.get("status") == "ok" else 503)
|
|
else:
|
|
self._json_response({"error": "not found"}, status=404)
|
|
|
|
def do_POST(self):
|
|
parsed = urlparse(self.path)
|
|
path = parsed.path.rstrip("/")
|
|
if path == "/reindex":
|
|
params = parse_qs(parsed.query)
|
|
full = (params.get("full") or [""])[0].lower() in {"1", "true", "yes"}
|
|
result = run_reindex(full=full)
|
|
status = 200 if "error" not in result else 500
|
|
self._json_response(result, status=status)
|
|
elif path == "/semantic-search":
|
|
try:
|
|
length = int(self.headers.get("Content-Length") or 0)
|
|
body = self.rfile.read(length).decode("utf-8") if length else "{}"
|
|
payload = json.loads(body or "{}")
|
|
query = payload.get("query") or payload.get("q") or ""
|
|
top_k = payload.get("top_k") or payload.get("topK") or 5
|
|
result = run_semantic_search(str(query), int(top_k))
|
|
self._json_response(result, status=200 if result.get("ok") else 400)
|
|
except json.JSONDecodeError:
|
|
self._json_response({"ok": False, "error": "invalid json", "results": []}, status=400)
|
|
except Exception as exc:
|
|
self._json_response({"ok": False, "error": str(exc), "results": []}, status=500)
|
|
else:
|
|
self._json_response({"error": "not found"}, status=404)
|
|
|
|
def _json_response(self, data, status=200):
|
|
body = json.dumps(data, indent=2).encode()
|
|
self.send_response(status)
|
|
self.send_header("Content-Type", "application/json")
|
|
self.send_header("Content-Length", str(len(body)))
|
|
self.end_headers()
|
|
self.wfile.write(body)
|
|
|
|
def log_message(self, format, *args):
|
|
# Minimal logging
|
|
pass
|
|
|
|
|
|
def main():
|
|
server = http.server.HTTPServer(("0.0.0.0", PORT), ReindexHandler)
|
|
print(f"obsidian-reindex-server listening on 0.0.0.0:{PORT}", flush=True)
|
|
try:
|
|
server.serve_forever()
|
|
except KeyboardInterrupt:
|
|
pass
|
|
server.server_close()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|