Files
devplacepy/devplacepy/services/jobs/deepsearch/worker.py
T
2026-07-07 15:28:28 +02:00

269 lines
8.2 KiB
Python

# retoor <retoor@molodetz.nl>
from __future__ import annotations
import asyncio
import json
import sys
from datetime import datetime, timezone
from pathlib import Path
from devplacepy.services.deepsearch.embeddings import embed_texts, local_embed
from devplacepy.services.deepsearch.store import Chunk, VectorStore
from devplacepy.utils import generate_uid
from .chunking import chunk_text
from .crawl import content_hash, crawl, search_queries, url_hash
from .enhance import plan_queries
from .orchestrate import orchestrate
from .phases import (
PHASE_ANALYSIS,
PHASE_CRAWLING,
PHASE_INDEXING,
PHASE_LABELS,
PHASE_PLANNING,
PHASE_SEARCHING,
PHASE_SYNTHESIS,
TOTAL_PHASES,
phase_index,
)
CONTROL_FILE = "control.json"
PAUSE_POLL_SECONDS = 1.0
EMBED_BATCH = 64
FRAME_VERSION = 1
_first_frame_sent = False
def _emit(frame: dict) -> None:
global _first_frame_sent
if not _first_frame_sent:
frame = {"version": FRAME_VERSION, **frame}
_first_frame_sent = True
sys.stdout.write(json.dumps(frame, ensure_ascii=False) + "\n")
sys.stdout.flush()
def _stage(stage: str, message: str, phase: str) -> None:
_emit({"type": "stage", "stage": stage, "message": message})
_emit(
{
"type": "phase",
"phase": phase,
"index": phase_index(phase),
"total": TOTAL_PHASES,
"label": PHASE_LABELS.get(phase, phase),
"message": message,
}
)
def _read_control(output_dir: Path) -> str:
path = output_dir / CONTROL_FILE
if not path.is_file():
return "running"
try:
return str(json.loads(path.read_text(encoding="utf-8")).get("state", "running"))
except (ValueError, OSError):
return "running"
def _make_stop(output_dir: Path):
async def should_stop() -> bool:
while True:
state = _read_control(output_dir)
if state == "cancelled":
return True
if state == "paused":
_emit({"type": "stage", "stage": "paused", "message": "Paused"})
await asyncio.sleep(PAUSE_POLL_SECONDS)
continue
return False
return should_stop
async def _index_chunks(
store: VectorStore, pages: list, api_key: str, emit
) -> tuple[int, str]:
chunks: list[Chunk] = []
for page in pages:
for position, text in enumerate(chunk_text(page.text)):
chunks.append(
Chunk(
uid=generate_uid(),
text=text,
url=page.url,
title=page.title,
depth=page.depth,
source=page.source,
position=position,
)
)
total = len(chunks)
if not chunks:
emit({"type": "embed_done", "backend": "empty", "chunk_count": 0})
return 0, "empty"
total_batches = (total + EMBED_BATCH - 1) // EMBED_BATCH
backend = "gateway"
forced_local = False
done = 0
collected: list[list[float]] = []
for batch_no, start in enumerate(range(0, total, EMBED_BATCH), start=1):
batch = chunks[start : start + EMBED_BATCH]
emit(
{
"type": "embed_batch",
"batch": batch_no,
"total_batches": total_batches,
"backend": "local" if forced_local else backend,
"chunks": len(batch),
"done": done,
"total": total,
"message": f"Embedding batch {batch_no}/{total_batches}",
}
)
if forced_local:
result = local_embed([chunk.text for chunk in batch])
else:
result = await embed_texts([chunk.text for chunk in batch], api_key)
if not result.vectors or result.backend != "gateway":
forced_local = True
result = local_embed([chunk.text for chunk in batch])
backend = result.backend
collected.extend(result.vectors)
done += len(batch)
emit(
{
"type": "embed_batch",
"batch": batch_no,
"total_batches": total_batches,
"backend": backend,
"chunks": len(batch),
"done": done,
"total": total,
"message": f"Embedded batch {batch_no}/{total_batches} ({backend})",
}
)
if forced_local:
collected = local_embed([chunk.text for chunk in chunks]).vectors
backend = "local"
store.add(chunks, collected)
final_backend = "local" if forced_local else backend
emit({"type": "embed_done", "backend": final_backend, "chunk_count": total})
return total, final_backend
async def _run(payload: dict, output_dir: Path) -> dict:
query = payload.get("query", "")
max_pages = int(payload.get("max_pages", 12))
depth = int(payload.get("depth", 2))
api_key = payload.get("api_key", "")
collection = payload.get("collection", "")
cached_hashes = set(payload.get("cached_hashes", []))
should_stop = _make_stop(output_dir)
_stage("planning", "Planning research queries", PHASE_PLANNING)
queries = await plan_queries(query, api_key, _emit)
_emit({"type": "queries", "queries": queries})
_stage("searching", "Searching the web", PHASE_SEARCHING)
candidates = await search_queries(queries, _emit)
_emit({"type": "candidates", "count": len(candidates)})
_stage("crawling", "Crawling sources", PHASE_CRAWLING)
outcome = await crawl(
candidates,
max_pages,
_emit,
lambda url: url_hash(url) in cached_hashes,
should_stop,
query=query,
depth=depth,
)
new_cache = [
{
"url_hash": url_hash(page.url),
"url": page.url,
"title": page.title,
"content_hash": content_hash(page.text),
"status": page.status,
"byte_size": len(page.text),
}
for page in outcome.pages
]
store = VectorStore(collection)
_stage("indexing", "Indexing content", PHASE_INDEXING)
chunk_count, embed_backend = await _index_chunks(
store, outcome.pages, api_key, _emit
)
_stage("analysis", "Running research agents", PHASE_ANALYSIS)
result = await orchestrate(
query, outcome.pages, api_key, _emit, store=store, queries=queries
)
_stage("synthesis", "Compiling cited report", PHASE_SYNTHESIS)
sources = [
{"url": page.url, "title": page.title, "source": page.source}
for page in outcome.pages
]
report = {
"query": query,
"depth": depth,
"max_pages": max_pages,
"generated_at": datetime.now(timezone.utc).isoformat(),
"summary": result.summary,
"findings": result.findings,
"sources": sources,
"score": result.score,
"confidence": result.confidence,
"source_diversity": result.source_diversity,
"synthesis": result.synthesis,
"page_count": len(outcome.pages),
"chunk_count": chunk_count,
"embed_backend": embed_backend,
"collection": collection,
}
(output_dir / "report.json").write_text(
json.dumps(report, ensure_ascii=False), encoding="utf-8"
)
(output_dir / "url_cache.json").write_text(
json.dumps(new_cache, ensure_ascii=False), encoding="utf-8"
)
_emit(
{
"type": "report_ready",
"score": report["score"],
"confidence": report["confidence"],
"source_diversity": report["source_diversity"],
"synthesis": report["synthesis"],
"page_count": report["page_count"],
"chunk_count": report["chunk_count"],
}
)
return report
def main(argv: list) -> int:
if len(argv) != 3:
sys.stderr.write("usage: deepsearch.worker <payload_json> <output_dir>\n")
return 2
payload = json.loads(Path(argv[1]).read_text(encoding="utf-8"))
output_dir = Path(argv[2])
output_dir.mkdir(parents=True, exist_ok=True)
try:
asyncio.run(_run(payload, output_dir))
except Exception as exc:
_emit({"type": "error", "message": str(exc)[:500]})
return 1
return 0
if __name__ == "__main__":
raise SystemExit(main(sys.argv))