python/compact_metadata_v1.py

#!/usr/bin/env python
"""Compact compact-v1 all_revisions Parquet metadata files.

This rewrites table-level Parquet fragments under
``data/v1/all_revisions/<table>/`` into fewer ``compact-*.parquet`` files.
Raw ``archives/`` are never touched, and ``latest_revisions`` is intentionally
left for ``rebuild_latest_snapshot.py`` to regenerate after compaction. Row
order is preserved in source-file order, not globally resorted; consumers must
use explicit sort keys when order matters.
"""

from __future__ import annotations

import argparse
import concurrent.futures
import json
import os
import shutil
import sys
import time
import uuid
from pathlib import Path
from typing import Any

import pyarrow as pa
import pyarrow.dataset as ds
import pyarrow.parquet as pq
from tqdm.auto import tqdm

from parquet_writer import (
    PARQUET_WRITE_KWARGS,
    TableSchema,
    _atomic_write_parquet,
    load_schemas,
)


EXCLUDED_TABLES = {"latest_revisions"}
DEFAULT_COMPACT_WORKERS = min(4, max(1, os.cpu_count() or 1))


def _log(msg: str) -> None:
    print(msg, file=sys.stderr, flush=True)


def _parquet_files(root: Path) -> list[Path]:
    return sorted(p for p in root.rglob("*.parquet") if p.is_file())


def _count_rows(files: list[Path]) -> int:
    if not files:
        return 0
    return int(ds.dataset([str(p) for p in files], format="parquet").count_rows())


def _fallback_schema(files: list[Path]) -> pa.Schema:
    if not files:
        raise ValueError("cannot infer schema without parquet files")
    return pq.read_schema(files[0])


def _align_table(table: pa.Table, schema: pa.Schema) -> pa.Table:
    for field in schema:
        if field.name in table.column_names:
            continue
        table = table.append_column(
            field.name,
            pa.nulls(table.num_rows, type=field.type),
        )
    table = table.select(schema.names)
    return table.cast(schema, safe=False)


def _write_compacted_table(
    *,
    table_name: str,
    table_dir: Path,
    target_dir: Path,
    table_schema: TableSchema | None,
    target_rows: int,
    batch_size: int,
) -> dict[str, Any]:
    source_files = _parquet_files(table_dir)
    source_rows = _count_rows(source_files)
    file_schema = table_schema.file_schema() if table_schema else _fallback_schema(source_files)

    target_dir.mkdir(parents=True, exist_ok=True)
    dataset = ds.dataset([str(p) for p in source_files], format="parquet")
    scanner = dataset.scanner(batch_size=batch_size)

    written_files = 0
    written_rows = 0
    buffered: list[pa.RecordBatch] = []
    buffered_rows = 0

    def flush() -> None:
        nonlocal written_files, written_rows, buffered, buffered_rows
        if not buffered:
            return
        table = pa.Table.from_batches(buffered)
        table = _align_table(table, file_schema)
        target = target_dir / f"compact-{written_files:05d}.parquet"
        _atomic_write_parquet(table, target, **PARQUET_WRITE_KWARGS)
        written_files += 1
        written_rows += table.num_rows
        buffered = []
        buffered_rows = 0

    for batch in scanner.to_batches():
        if batch.num_rows <= 0:
            continue
        buffered.append(batch)
        buffered_rows += batch.num_rows
        if buffered_rows >= target_rows:
            flush()
    flush()

    if written_rows != source_rows:
        raise RuntimeError(
            f"{table_name}: row count changed during compaction: "
            f"source={source_rows}, written={written_rows}"
        )

    return {
        "table": table_name,
        "source_files": len(source_files),
        "source_rows": source_rows,
        "written_files": written_files,
        "written_rows": written_rows,
        "compacted": True,
    }


def compact_metadata_v1(
    repo_root: Path,
    *,
    schema_version: str = "v1",
    schemas_dir: Path | None = None,
    target_rows: int = 1_000_000,
    batch_size: int = 65_536,
    min_files: int = 2,
    workers: int = DEFAULT_COMPACT_WORKERS,
    keep_backup: bool = False,
) -> dict[str, Any]:
    repo_root = repo_root.resolve()
    all_revisions = repo_root / "data" / schema_version / "all_revisions"
    if not all_revisions.exists():
        raise FileNotFoundError(all_revisions)

    schemas_path = schemas_dir or repo_root / "schemas" / schema_version
    schemas = load_schemas(schemas_path) if schemas_path.exists() else {}

    tx_id = f"metadata-compact-{int(time.time())}-{uuid.uuid4().hex[:8]}"
    tx_root = repo_root / ".scratch" / "metadata-compaction" / tx_id
    staging_root = tx_root / "staging" / "all_revisions"
    backup_root = tx_root / "backup" / "all_revisions"

    table_dirs = [
        p
        for p in sorted(all_revisions.iterdir())
        if p.is_dir() and p.name not in EXCLUDED_TABLES
    ]

    summaries: list[dict[str, Any]] = []
    compacted_tables: list[str] = []
    summary_by_table: dict[str, dict[str, Any]] = {}

    try:
        work_items: list[tuple[Path, list[Path]]] = []
        for table_dir in table_dirs:
            files = _parquet_files(table_dir)
            if len(files) < min_files:
                rows = _count_rows(files)
                summary_by_table[table_dir.name] = {
                    "table": table_dir.name,
                    "source_files": len(files),
                    "source_rows": rows,
                    "written_files": len(files),
                    "written_rows": rows,
                    "compacted": False,
                }
            else:
                work_items.append((table_dir, files))

        work_items.sort(
            key=lambda item: sum(p.stat().st_size for p in item[1]),
            reverse=True,
        )
        max_workers = min(max(1, workers), max(len(work_items), 1))
        desc = f"compacting all_revisions ({max_workers} workers)"
        bar = tqdm(
            total=len(work_items),
            desc=desc,
            unit="table",
            file=sys.stderr,
            mininterval=1.0,
            dynamic_ncols=True,
        )
        try:
            if max_workers == 1:
                for table_dir, _files in work_items:
                    summary = _write_compacted_table(
                        table_name=table_dir.name,
                        table_dir=table_dir,
                        target_dir=staging_root / table_dir.name,
                        table_schema=schemas.get(table_dir.name),
                        target_rows=target_rows,
                        batch_size=batch_size,
                    )
                    summary_by_table[table_dir.name] = summary
                    bar.update(1)
            else:
                with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as ex:
                    futures = {
                        ex.submit(
                            _write_compacted_table,
                            table_name=table_dir.name,
                            table_dir=table_dir,
                            target_dir=staging_root / table_dir.name,
                            table_schema=schemas.get(table_dir.name),
                            target_rows=target_rows,
                            batch_size=batch_size,
                        ): table_dir.name
                        for table_dir, _files in work_items
                    }
                    for future in concurrent.futures.as_completed(futures):
                        table_name = futures[future]
                        summary_by_table[table_name] = future.result()
                        bar.update(1)
        finally:
            bar.close()

        summaries = [
            summary_by_table[table_dir.name]
            for table_dir in table_dirs
            if table_dir.name in summary_by_table
        ]
        compacted_tables = [
            summary["table"]
            for summary in summaries
            if summary.get("compacted")
        ]

        for table in compacted_tables:
            source = all_revisions / table
            backup = backup_root / table
            staged = staging_root / table
            backup.parent.mkdir(parents=True, exist_ok=True)
            if backup.exists():
                shutil.rmtree(backup)
            source.rename(backup)
            staged.rename(source)

        total_before = sum(int(s["source_files"]) for s in summaries)
        total_after = sum(int(s["written_files"]) for s in summaries)
        total_rows = sum(int(s["written_rows"]) for s in summaries)
        result = {
            "ok": True,
            "tx_id": tx_id,
            "tables": summaries,
            "compacted_tables": compacted_tables,
            "files_before": total_before,
            "files_after": total_after,
            "rows": total_rows,
            "backup_path": str(backup_root) if keep_backup else None,
        }
        if keep_backup:
            shutil.rmtree(staging_root.parent, ignore_errors=True)
        else:
            shutil.rmtree(tx_root, ignore_errors=True)
        return result
    except Exception:
        _log(f"compaction failed; transaction scratch kept at {tx_root}")
        raise


def parse_args(argv: list[str] | None = None) -> argparse.Namespace:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--repo-root", default=".")
    parser.add_argument("--schema-version", default="v1")
    parser.add_argument("--schemas-dir", default=None)
    parser.add_argument("--target-rows", type=int, default=1_000_000)
    parser.add_argument("--batch-size", type=int, default=65_536)
    parser.add_argument("--min-files", type=int, default=2)
    parser.add_argument(
        "--workers",
        type=int,
        default=DEFAULT_COMPACT_WORKERS,
        help=(
            "Parallel table compaction workers. Defaults to min(4, CPU count), "
            f"currently {DEFAULT_COMPACT_WORKERS}."
        ),
    )
    parser.add_argument("--keep-backup", action="store_true")
    parser.add_argument("--json", action="store_true")
    return parser.parse_args(argv)


def main(argv: list[str] | None = None) -> int:
    args = parse_args(argv)
    summary = compact_metadata_v1(
        Path(args.repo_root),
        schema_version=args.schema_version,
        schemas_dir=Path(args.schemas_dir) if args.schemas_dir else None,
        target_rows=max(1, args.target_rows),
        batch_size=max(1, args.batch_size),
        min_files=max(1, args.min_files),
        workers=max(1, args.workers),
        keep_backup=args.keep_backup,
    )
    if args.json:
        print(json.dumps(summary, indent=2, sort_keys=True))
    else:
        print(
            "compaction: "
            f"files_before={summary['files_before']} "
            f"files_after={summary['files_after']} "
            f"rows={summary['rows']}"
        )
        for table in summary["tables"]:
            if table["compacted"]:
                print(
                    f"  {table['table']}: "
                    f"{table['source_files']} -> {table['written_files']} files, "
                    f"{table['written_rows']} rows"
                )
    return 0


if __name__ == "__main__":
    raise SystemExit(main())