# File Objective  : Orchestrate Phase 2 end-to-end ingestion into Qdrant.
# Scope           : Dev script — run once (or re-run idempotently) to populate the collection.
# What it does    : 1. Streams Documents → chunks → embeds in batches → upserts dense+sparse.
#                   2. Respects INGEST_LIMIT env var (default 2000) for fast sample runs.
#                   3. Prints a single progress line and a final summary.
# What it does not: 1. Build graph (Phase 7) or compute gaps (Phase 6).

from __future__ import annotations

import os
from pathlib import Path

from vantage_core.adapters.embedder_st import STEmbedder
from vantage_core.adapters.vector_qdrant import QdrantVectorRepo
from vantage_core.chunker import chunk_document
from vantage_core.loader import load_huffpost

DATA_PATH  = Path("data/news_category.json")
BATCH_SIZE = 64     # chunks per upsert — balances memory vs network round-trips


def main() -> None:
    limit = int(os.getenv("INGEST_LIMIT", "2000"))
    print(f"\n── Vantage Ingest ──────────────────────────────────────────")
    print(f"   dataset : {DATA_PATH}  |  limit : {limit} docs  |  batch : {BATCH_SIZE}")

    repo     = QdrantVectorRepo()
    embedder = STEmbedder()
    repo.ensure_collection()

    buf: list = []
    total_docs = total_chunks = 0

    for doc in load_huffpost(DATA_PATH, limit=limit):
        buf.extend(chunk_document(doc))
        total_docs += 1

        if len(buf) >= BATCH_SIZE:
            _flush(buf, embedder, repo)
            total_chunks += len(buf)
            print(f"  docs={total_docs:>5}  chunks={total_chunks:>5}", end="\r")
            buf = []

    if buf:                         # flush remainder
        _flush(buf, embedder, repo)
        total_chunks += len(buf)

    final = repo.count()
    print(f"\n── Done ────────────────────────────────────────────────────")
    print(f"   docs={total_docs}  chunks_sent={total_chunks}  qdrant_total={final}\n")


def _flush(chunks: list, embedder: STEmbedder, repo: QdrantVectorRepo) -> None:
    import time
    texts   = [c.chunk_text for c in chunks]
    vectors = embedder.embed(texts)
    for attempt in range(3):
        try:
            repo.upsert(chunks, vectors)
            return
        except Exception:
            if attempt == 2:
                raise
            time.sleep(8 * (attempt + 1))   # 8s, 16s — survive Qdrant Cloud free-tier disconnects


if __name__ == "__main__":
    main()