lamindb_setup 1.15.0__py3-none-any.whl → 1.15.1__py3-none-any.whl

lamindb_setup/__init__.py

@@ -35,7 +35,7 @@ Migration management
 
 """
 
-__version__ = "1.15.0"  # denote a release candidate for 0.1.0 with 0.1rc1
+__version__ = "1.15.1"  # denote a release candidate for 0.1.0 with 0.1rc1
 
 import os
 import warnings

lamindb_setup/core/_aws_options.py

@@ -20,6 +20,7 @@ lamin_env = os.getenv("LAMIN_ENV")
 if lamin_env is None or lamin_env == "prod":
     HOSTED_BUCKETS = tuple([f"s3://lamin-{region}" for region in HOSTED_REGIONS])
 else:
+    logger.warning("loaded LAMIN_ENV: staging")
     HOSTED_BUCKETS = ("s3://lamin-hosted-test",)  # type: ignore
 
 

lamindb_setup/io.py

@@ -1,11 +1,14 @@
 from __future__ import annotations
 
+import io
 import json
 import warnings
+from concurrent.futures import ProcessPoolExecutor, as_completed
 from importlib import import_module
 from pathlib import Path
 from typing import TYPE_CHECKING
 
+import numpy as np
 import pandas as pd
 from django.db import models, transaction
 from rich.progress import Progress
@@ -43,21 +46,101 @@ def _get_registries(module_name: str) -> list[str]:
     ]
 
 
-def _export_registry_to_parquet(registry: type[models.Model], directory: Path) -> None:
-    """Export a single registry table to parquet."""
+def _export_full_table(
+    registry_info: tuple[str, str, str | None],
+    directory: Path,
+    chunk_size: int,
+) -> list[tuple[str, Path]] | str:
+    """Export a registry table to parquet.
+
+    For PostgreSQL, uses COPY TO which streams the table directly to CSV format,
+    bypassing query planner overhead and row-by-row conversion (10-50x faster than SELECT).
+
+    For SQLite with large tables, reads in chunks to avoid memory issues when tables exceed available RAM.
+
+    Args:
+        registry_info: Tuple of (module_name, model_name, field_name) where field_name
+            is None for regular tables or the field name for M2M link tables.
+        directory: Output directory for parquet files.
+        chunk_size: Maximum rows per chunk for SQLite large tables.
+
+    Returns:
+        String identifier for single-file exports, or list of (table_name, chunk_path) tuples for chunked exports that need merging.
+    """
+    from django.db import connection
+
     import lamindb_setup as ln_setup
 
+    module_name, model_name, field_name = registry_info
+    schema_module = import_module(module_name)
+    registry = getattr(schema_module, model_name)
+
+    if field_name:
+        registry = getattr(registry, field_name).through
+
     table_name = registry._meta.db_table
-    with warnings.catch_warnings():
-        warnings.filterwarnings("ignore", message="Skipped unsupported reflection")
-        df = pd.read_sql_table(table_name, ln_setup.settings.instance.db)
-    df.to_parquet(directory / f"{table_name}.parquet", compression=None)
+
+    try:
+        if ln_setup.settings.instance.dialect == "postgresql":
+            buffer = io.StringIO()
+            with connection.cursor() as cursor:
+                cursor.copy_expert(
+                    f'COPY "{table_name}" TO STDOUT WITH (FORMAT CSV, HEADER TRUE)',
+                    buffer,
+                )
+            buffer.seek(0)
+            df = pd.read_csv(buffer)
+            df.to_parquet(directory / f"{table_name}.parquet", compression=None)
+            return (
+                f"{module_name}.{model_name}.{field_name}"
+                if field_name
+                else f"{module_name}.{model_name}"
+            )
+        else:
+            with warnings.catch_warnings():
+                warnings.filterwarnings(
+                    "ignore", message="Skipped unsupported reflection"
+                )
+                row_count = pd.read_sql(
+                    f"SELECT COUNT(*) as count FROM {table_name}",
+                    ln_setup.settings.instance.db,
+                ).iloc[0]["count"]
+
+                if row_count > chunk_size:
+                    chunk_files = []
+                    num_chunks = (row_count + chunk_size - 1) // chunk_size
+                    for chunk_id in range(num_chunks):
+                        offset = chunk_id * chunk_size
+                        df = pd.read_sql(
+                            f"SELECT * FROM {table_name} LIMIT {chunk_size} OFFSET {offset}",
+                            ln_setup.settings.instance.db,
+                        )
+                        chunk_file = (
+                            directory / f"{table_name}_chunk_{chunk_id}.parquet"
+                        )
+                        df.to_parquet(chunk_file, compression=None)
+                        chunk_files.append((table_name, chunk_file))
+                    return chunk_files
+                else:
+                    df = pd.read_sql_table(table_name, ln_setup.settings.instance.db)
+                    df.to_parquet(directory / f"{table_name}.parquet", compression=None)
+                    return (
+                        f"{module_name}.{model_name}.{field_name}"
+                        if field_name
+                        else f"{module_name}.{model_name}"
+                    )
+    except (ValueError, pd.errors.DatabaseError):
+        raise ValueError(
+            f"Table '{table_name}' was not found. The instance might need to be migrated."
+        ) from None
 
 
 def export_db(
     module_names: Sequence[str] | None = None,
     *,
     output_dir: str | Path = "./lamindb_export/",
+    max_workers: int = 8,
+    chunk_size: int = 500_000,
 ) -> None:
     """Export registry tables and many-to-many link tables to parquet files.
 
@@ -67,26 +150,57 @@ def export_db(
         module_names: Module names to export (e.g., ["lamindb", "bionty", "wetlab"]).
             Defaults to "lamindb" if not provided.
         output_dir: Directory path for exported parquet files.
+        max_workers: Number of parallel processes.
+        chunk_size: Number of rows per chunk for large tables.
     """
     directory = Path(output_dir)
     directory.mkdir(parents=True, exist_ok=True)
 
     module_names = module_names or ["lamindb"]
     modules = {name: _get_registries(name) for name in module_names}
-    total_models = sum(len(models) for models in modules.values())
+
+    tasks = []
+    for module_name, model_names in modules.items():
+        schema_module = import_module(module_name)
+        for model_name in model_names:
+            registry = getattr(schema_module, model_name)
+            tasks.append((module_name, model_name, None))
+            for field in registry._meta.many_to_many:
+                tasks.append((module_name, model_name, field.name))
+
+    chunk_files_by_table: dict[str, list[Path]] = {}
 
     with Progress() as progress:
-        task = progress.add_task("Exporting", total=total_models)
-        for module_name, model_names in modules.items():
-            schema_module = import_module(module_name)
-            for model_name in model_names:
-                progress.update(task, description=f"[cyan]{module_name}.{model_name}")
-                registry = getattr(schema_module, model_name)
-                _export_registry_to_parquet(registry, directory)
-                for field in registry._meta.many_to_many:
-                    link_orm = getattr(registry, field.name).through
-                    _export_registry_to_parquet(link_orm, directory)
-                progress.advance(task)
+        task_id = progress.add_task("Exporting", total=len(tasks))
+        with ProcessPoolExecutor(max_workers=max_workers) as executor:
+            futures = {
+                executor.submit(_export_full_table, task, directory, chunk_size): task
+                for task in tasks
+            }
+
+            for future in as_completed(futures):
+                result = future.result()
+                if isinstance(result, list):
+                    for table_name, chunk_file in result:
+                        chunk_files_by_table.setdefault(table_name, []).append(
+                            chunk_file
+                        )
+                progress.advance(task_id)
+
+    for table_name, chunk_files in chunk_files_by_table.items():
+        merged_df = pd.concat([pd.read_parquet(f) for f in sorted(chunk_files)])
+        merged_df.to_parquet(directory / f"{table_name}.parquet", compression=None)
+        for chunk_file in chunk_files:
+            chunk_file.unlink()
+
+
+def _serialize_value(val):
+    """Convert value to JSON string if it's a dict, list, or numpy array, otherwise return as-is."""
+    if isinstance(val, (dict, list, np.ndarray)):
+        return json.dumps(
+            val, default=lambda o: o.tolist() if isinstance(o, np.ndarray) else None
+        )
+    return val
 
 
 def _import_registry(
@@ -96,8 +210,14 @@ def _import_registry(
 ) -> None:
     """Import a single registry table from parquet.
 
-    Uses raw SQL export instead of django to later circumvent FK constraints.
+    For PostgreSQL, uses COPY FROM which bypasses SQL parsing and writes directly to
+    table pages (20-50x faster than multi-row INSERTs).
+
+    For SQLite, uses multi-row INSERTs with dynamic chunking to stay under the 999
+    variable limit (2-5x faster than single-row INSERTs).
     """
+    from django.db import connection
+
     table_name = registry._meta.db_table
     parquet_file = directory / f"{table_name}.parquet"
 
@@ -113,13 +233,46 @@ def _import_registry(
 
     for col in df.columns:
         if df[col].dtype == "object":
-            df[col] = df[col].apply(
-                lambda x: json.dumps(x) if isinstance(x, (dict, list)) else x
-            )
+            mask = df[col].apply(lambda x: isinstance(x, (dict, list, np.ndarray)))
+            if mask.any():
+                df.loc[mask, col] = df.loc[mask, col].map(_serialize_value)
 
-    from django.db import connection
+    if df.empty:
+        return
 
-    df.to_sql(table_name, connection.connection, if_exists=if_exists, index=False)
+    if connection.vendor == "postgresql":
+        columns = df.columns.tolist()
+        column_names = ", ".join(f'"{col}"' for col in columns)
+
+        buffer = io.StringIO()
+        df.to_csv(buffer, index=False, header=False, sep="\t", na_rep="\\N")
+        buffer.seek(0)
+
+        with connection.cursor() as cursor:
+            if if_exists == "replace":
+                cursor.execute(f'DELETE FROM "{table_name}"')
+            elif if_exists == "fail":
+                cursor.execute(f'SELECT COUNT(*) FROM "{table_name}"')
+                if cursor.fetchone()[0] > 0:
+                    raise ValueError(f"Table {table_name} already contains data")
+
+            cursor.copy_expert(
+                f"COPY \"{table_name}\" ({column_names}) FROM STDIN WITH (FORMAT CSV, DELIMITER E'\\t', NULL '\\N')",
+                buffer,
+            )
+    else:
+        num_cols = len(df.columns)
+        max_vars = 900  # SQLite has a limit of 999 variables per statement
+        chunksize = max(1, max_vars // num_cols)
+
+        df.to_sql(
+            table_name,
+            connection.connection,
+            if_exists=if_exists,
+            index=False,
+            method="multi",
+            chunksize=chunksize,
+        )
 
 
 def import_db(
@@ -157,38 +310,45 @@ def import_db(
     modules = {name: _get_registries(name) for name in module_names}
     total_models = sum(len(models) for models in modules.values())
 
-    # Disable FK constraints to allow insertion in arbitrary order
-    if ln_setup.settings.instance.dialect == "sqlite":
+    is_sqlite = ln_setup.settings.instance.dialect == "sqlite"
+
+    try:
         with connection.cursor() as cursor:
             if ln_setup.settings.instance.dialect == "postgresql":
                 cursor.execute("SET session_replication_role = 'replica'")
-            elif ln_setup.settings.instance.dialect == "sqlite":
+            elif is_sqlite:
                 cursor.execute("PRAGMA foreign_keys = OFF")
-
-    with transaction.atomic():
-        if ln_setup.settings.instance.dialect == "postgresql":
-            with connection.cursor() as cursor:
-                cursor.execute("SET CONSTRAINTS ALL DEFERRED")
-
-        with Progress() as progress:
-            task = progress.add_task("Importing", total=total_models)
-            for module_name, model_names in modules.items():
-                schema_module = import_module(module_name)
-                for model_name in model_names:
-                    progress.update(
-                        task, description=f"[cyan]{module_name}.{model_name}"
-                    )
-                    registry = getattr(schema_module, model_name)
-                    _import_registry(registry, directory, if_exists=if_exists)
-                    for field in registry._meta.many_to_many:
-                        link_orm = getattr(registry, field.name).through
-                        _import_registry(link_orm, directory, if_exists=if_exists)
-                    progress.advance(task)
-
-    # Re-enable FK constraints again
-    if ln_setup.settings.instance.dialect == "sqlite":
+                # Disables fsync - OS buffers writes to disk, 10-50x faster but can corrupt DB on crash
+                cursor.execute("PRAGMA synchronous = OFF")
+                # Keeps rollback journal in RAM - 2-5x faster but cannot rollback on crash
+                cursor.execute("PRAGMA journal_mode = MEMORY")
+                # 64MB page cache for better performance on large imports
+                cursor.execute("PRAGMA cache_size = -64000")
+
+        with transaction.atomic():
+            if ln_setup.settings.instance.dialect == "postgresql":
+                with connection.cursor() as cursor:
+                    cursor.execute("SET CONSTRAINTS ALL DEFERRED")
+
+            with Progress() as progress:
+                task = progress.add_task("Importing", total=total_models)
+                for module_name, model_names in modules.items():
+                    schema_module = import_module(module_name)
+                    for model_name in model_names:
+                        progress.update(
+                            task, description=f"[cyan]{module_name}.{model_name}"
+                        )
+                        registry = getattr(schema_module, model_name)
+                        _import_registry(registry, directory, if_exists=if_exists)
+                        for field in registry._meta.many_to_many:
+                            link_orm = getattr(registry, field.name).through
+                            _import_registry(link_orm, directory, if_exists=if_exists)
+                        progress.advance(task)
+    finally:
         with connection.cursor() as cursor:
             if ln_setup.settings.instance.dialect == "postgresql":
                 cursor.execute("SET session_replication_role = 'origin'")
-            elif ln_setup.settings.instance.dialect == "sqlite":
+            elif is_sqlite:
+                cursor.execute("PRAGMA synchronous = FULL")
+                cursor.execute("PRAGMA journal_mode = DELETE")
                 cursor.execute("PRAGMA foreign_keys = ON")

{lamindb_setup-1.15.0.dist-info → lamindb_setup-1.15.1.dist-info}/METADATA

@@ -1,10 +1,11 @@
-Metadata-Version: 2.3
+Metadata-Version: 2.4
 Name: lamindb_setup
-Version: 1.15.0
+Version: 1.15.1
 Summary: Setup & configure LaminDB.
 Author-email: Lamin Labs <open-source@lamin.ai>
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
+License-File: LICENSE
 Requires-Dist: lamin_utils>=0.3.3
 Requires-Dist: django>=5.2,<5.3
 Requires-Dist: dj_database_url>=1.3.0,<3.0.0

{lamindb_setup-1.15.0.dist-info → lamindb_setup-1.15.1.dist-info}/RECORD

@@ -1,4 +1,4 @@
-lamindb_setup/__init__.py,sha256=jFfiLt0yIvOHoum_g8DmraECRdU-EOd-iSmAKTZyG3c,3215
+lamindb_setup/__init__.py,sha256=3Oaw6Mj_4lWEvxbDmE5Y6Q-hyP20r-8yeSKOpJXvjso,3215
 lamindb_setup/_cache.py,sha256=pGvDNVHGx4HWr_6w5ajqEJOdysmaGc6F221qFnXkT-k,2747
 lamindb_setup/_check.py,sha256=28PcG8Kp6OpjSLSi1r2boL2Ryeh6xkaCL87HFbjs6GA,129
 lamindb_setup/_check_setup.py,sha256=ToKMxsUq8dQBQh8baOrNVlSb1iC8h4zTg5dV8wMu0W4,6760
@@ -16,11 +16,11 @@ lamindb_setup/_set_managed_storage.py,sha256=y5YQASsWNYVWUYeLgh3N2YBETYP7mBtbpxe
 lamindb_setup/_setup_user.py,sha256=ojq7UP2Aia8GTCr6m8fylFx9VSuvGu0HmvIJ8RzymE0,6108
 lamindb_setup/_silence_loggers.py,sha256=AKF_YcHvX32eGXdsYK8MJlxEaZ-Uo2f6QDRzjKFCtws,1568
 lamindb_setup/errors.py,sha256=lccF3X3M2mcbHVG_0HxfuJRFFpUE-42paccIxFOfefQ,1958
-lamindb_setup/io.py,sha256=7wU3g2AH1h18wQerltA7p8VP0n47ZOsy32XlRcBbBAc,7321
+lamindb_setup/io.py,sha256=9LstFkIaki_m_oE7hFSPN8j1eXLjfEyarf13v2wcGso,13702
 lamindb_setup/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 lamindb_setup/types.py,sha256=fuQxZJnrGYe7a_Ju9n1RqO-HhkOAr1l1xjpAg9dmBu8,605
 lamindb_setup/core/__init__.py,sha256=gMqbkPeD-T6o_KykonvNiUJJQYz5SCk2yGgFNWFUBNc,603
-lamindb_setup/core/_aws_options.py,sha256=6SAs3dhKxNadQqt7Ce7uHSus0DsU_LcNZoCQWvhgBWA,9626
+lamindb_setup/core/_aws_options.py,sha256=9kQ5BB-cuJQrlJRGNqMRe1m48dP67xMbefOJP2c9OQw,9674
 lamindb_setup/core/_aws_storage.py,sha256=QEtV-riQrwfivcwqHnXBbkJ-9YyNEXL4fLoCmOHZ1BI,2003
 lamindb_setup/core/_clone.py,sha256=2NlXV04yykqg_k7z59C_kD1F1Hi4H-55H-JtNjhenQ0,3691
 lamindb_setup/core/_deprecated.py,sha256=M3vpM4fZPOncxY2qsXQAPeaEph28xWdv7tYaueaUyAA,2554
@@ -44,7 +44,7 @@ lamindb_setup/core/exceptions.py,sha256=qjMzqy_uzPA7mCOdnoWnS_fdA6OWbdZGftz-YYpl
 lamindb_setup/core/hashing.py,sha256=Y8Uc5uSGTfU6L2R_gb5w8DdHhGRog7RnkK-e9FEMjPY,3680
 lamindb_setup/core/types.py,sha256=T7NwspfRHgIIpYsXDcApks8jkOlGeGRW-YbVLB7jNIo,67
 lamindb_setup/core/upath.py,sha256=bi3k8AYeiGB_NtVTO9e9gHsfs2AFB4fXiVHcbNpnlpI,35780
-lamindb_setup-1.15.0.dist-info/LICENSE,sha256=UOZ1F5fFDe3XXvG4oNnkL1-Ecun7zpHzRxjp-XsMeAo,11324
-lamindb_setup-1.15.0.dist-info/WHEEL,sha256=CpUCUxeHQbRN5UGRQHYRJorO5Af-Qy_fHMctcQ8DSGI,82
-lamindb_setup-1.15.0.dist-info/METADATA,sha256=aVm2bsYK6bWwBDEcp1IJR3kp8xrRnJb5CYK_KT09HCA,1798
-lamindb_setup-1.15.0.dist-info/RECORD,,
+lamindb_setup-1.15.1.dist-info/licenses/LICENSE,sha256=UOZ1F5fFDe3XXvG4oNnkL1-Ecun7zpHzRxjp-XsMeAo,11324
+lamindb_setup-1.15.1.dist-info/WHEEL,sha256=G2gURzTEtmeR8nrdXUJfNiB3VYVxigPQ-bEQujpNiNs,82
+lamindb_setup-1.15.1.dist-info/METADATA,sha256=z9yk-pHFnYB7gv59trxqyhbbnpxzRC2XPuhMoTuujDE,1820
+lamindb_setup-1.15.1.dist-info/RECORD,,

{lamindb_setup-1.15.0.dist-info → lamindb_setup-1.15.1.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: flit 3.10.1
+Generator: flit 3.12.0
 Root-Is-Purelib: true
 Tag: py3-none-any