pixeltable 0.3.15__py3-none-any.whl → 0.4.0rc1__py3-none-any.whl

pixeltable/io/table_data_conduit.py

@@ -23,7 +23,6 @@ from .utils import normalize_schema_names
 
 _logger = logging.getLogger('pixeltable')
 
-# ---------------------------------------------------------------------------------------------------------
 
 if TYPE_CHECKING:
     import datasets  # type: ignore[import-untyped]
@@ -46,9 +45,6 @@ class TableDataConduitFormat(str, enum.Enum):
         return False
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 @dataclass
 class TableDataConduit:
     source: TableDataSource
@@ -129,9 +125,6 @@ class TableDataConduit:
             raise excs.Error(f'Missing required column(s) ({", ".join(missing_cols)})')
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class DFTableDataConduit(TableDataConduit):
     pxt_df: pxt.DataFrame = None
 
@@ -155,9 +148,6 @@ class DFTableDataConduit(TableDataConduit):
         self.check_source_columns_are_insertable(self.pxt_df.schema.keys())
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class RowDataTableDataConduit(TableDataConduit):
     raw_rows: Optional[RowData] = None
     disable_mapping: bool = True
@@ -235,9 +225,6 @@ class RowDataTableDataConduit(TableDataConduit):
             yield self.valid_rows
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class PandasTableDataConduit(TableDataConduit):
     pd_df: pd.DataFrame = None
     batch_count: int = 0
@@ -293,9 +280,6 @@ class PandasTableDataConduit(TableDataConduit):
             yield self.valid_rows
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class CSVTableDataConduit(TableDataConduit):
     @classmethod
     def from_tds(cls, tds: TableDataConduit) -> 'PandasTableDataConduit':
@@ -307,9 +291,6 @@ class CSVTableDataConduit(TableDataConduit):
         return PandasTableDataConduit.from_tds(t)
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class ExcelTableDataConduit(TableDataConduit):
     @classmethod
     def from_tds(cls, tds: TableDataConduit) -> 'PandasTableDataConduit':
@@ -321,9 +302,6 @@ class ExcelTableDataConduit(TableDataConduit):
         return PandasTableDataConduit.from_tds(t)
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class JsonTableDataConduit(TableDataConduit):
     @classmethod
     def from_tds(cls, tds: TableDataConduit) -> RowDataTableDataConduit:
@@ -346,9 +324,6 @@ class JsonTableDataConduit(TableDataConduit):
         return t2
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class HFTableDataConduit(TableDataConduit):
     hf_ds: Optional[Union[datasets.Dataset, datasets.DatasetDict]] = None
     column_name_for_split: Optional[str] = None
@@ -478,9 +453,6 @@ class HFTableDataConduit(TableDataConduit):
                 yield batch
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class ParquetTableDataConduit(TableDataConduit):
     pq_ds: Optional[ParquetDataset] = None
 
@@ -542,9 +514,6 @@ class ParquetTableDataConduit(TableDataConduit):
             raise e
 
 
-# ---------------------------------------------------------------------------------------------------------
-
-
 class UnkTableDataConduit(TableDataConduit):
     """Source type is not known at the time of creation"""
 

pixeltable/metadata/__init__.py

@@ -16,7 +16,7 @@ _console_logger = ConsoleLogger(logging.getLogger('pixeltable'))
 
 
 # current version of the metadata; this is incremented whenever the metadata schema changes
-VERSION = 35
+VERSION = 36
 
 
 def create_system_info(engine: sql.engine.Engine) -> None:

pixeltable/metadata/converters/convert_13.py

@@ -12,9 +12,9 @@ _logger = logging.getLogger('pixeltable')
 @register_converter(version=13)
 def _(engine: sql.engine.Engine) -> None:
     with engine.begin() as conn:
-        for row in conn.execute(sql.select(Table)):
+        for row in conn.execute(sql.select(Table.id, Table.md)):
             id = row[0]
-            md = row[2]
+            md = row[1]
             updated_md = __update_md(md)
             if updated_md != md:
                 _logger.info(f'Updating schema for table: {id}')

pixeltable/metadata/converters/convert_30.py

@@ -1,33 +1,28 @@
 import copy
+from uuid import UUID
 
 import sqlalchemy as sql
 
 from pixeltable.metadata import register_converter
 from pixeltable.metadata.converters.util import (
-    convert_table_record,
+    convert_table_md,
     convert_table_schema_version_record,
     convert_table_version_record,
 )
-from pixeltable.metadata.schema import Table, TableSchemaVersion, TableVersion
+from pixeltable.metadata.schema import TableSchemaVersion, TableVersion
 
 
 @register_converter(version=30)
 def _(engine: sql.engine.Engine) -> None:
-    convert_table_record(engine, table_record_updater=__update_table_record)
+    convert_table_md(engine, table_md_updater=__update_table_md)
     convert_table_version_record(engine, table_version_record_updater=__update_table_version_record)
     convert_table_schema_version_record(
         engine, table_schema_version_record_updater=__update_table_schema_version_record
     )
 
 
-def __update_table_record(record: Table) -> None:
-    """
-    Update TableMd with table_id
-    """
-    assert isinstance(record.md, dict)
-    md = copy.copy(record.md)
-    md['tbl_id'] = str(record.id)
-    record.md = md
+def __update_table_md(md: dict, tbl_id: UUID) -> None:
+    md['tbl_id'] = str(tbl_id)
 
 
 def __update_table_version_record(record: TableVersion) -> None:

pixeltable/metadata/converters/convert_35.py

@@ -0,0 +1,9 @@
+import sqlalchemy as sql
+
+from pixeltable.metadata import register_converter
+
+
+@register_converter(version=35)
+def _(engine: sql.engine.Engine) -> None:
+    with engine.begin() as conn:
+        conn.execute(sql.text('ALTER TABLE tables ADD COLUMN lock_dummy int8'))

pixeltable/metadata/converters/util.py

@@ -33,9 +33,10 @@ def convert_table_md(
             the original entry will be replaced, and the traversal will continue with `v'`.
     """
     with engine.begin() as conn:
-        for row in conn.execute(sql.select(Table)):
+        # avoid a SELECT * here, which breaks when we add new columns to Table
+        for row in conn.execute(sql.select(Table.id, Table.md)):
             tbl_id = row[0]
-            table_md = row[2]
+            table_md = row[1]
             assert isinstance(table_md, dict)
             updated_table_md = copy.deepcopy(table_md)
             if table_md_updater is not None:
@@ -145,13 +146,6 @@ def __update_schema_column(table_schema_version_md: dict, schema_column_updater:
         schema_column_updater(schema_col)
 
 
-def convert_table_record(engine: sql.engine.Engine, table_record_updater: Optional[Callable[[Table], None]]) -> None:
-    with sql.orm.Session(engine, future=True) as session:
-        for record in session.query(Table).all():
-            table_record_updater(record)
-        session.commit()
-
-
 def convert_table_version_record(
     engine: sql.engine.Engine, table_version_record_updater: Optional[Callable[[TableVersion], None]]
 ) -> None:

pixeltable/metadata/notes.py

@@ -2,6 +2,7 @@
 # rather than as a comment, so that the existence of a description can be enforced by
 # the unit tests when new versions are added.
 VERSION_NOTES = {
+    36: 'Added Table.lock_dummy',
     35: 'Track reference_tbl in ColumnRef',
     34: 'Set default value for is_pk field in column metadata to False',
     33: 'Add is_replica field to table metadata',

pixeltable/metadata/schema.py

@@ -84,7 +84,8 @@ class Dir(Base):
     )
     parent_id: orm.Mapped[uuid.UUID] = orm.mapped_column(UUID(as_uuid=True), ForeignKey('dirs.id'), nullable=True)
     md: orm.Mapped[dict[str, Any]] = orm.mapped_column(JSONB, nullable=False)  # DirMd
-    # This field is updated to synchronize database operations across multiple sessions
+
+    # used to force acquisition of an X-lock via an Update stmt
     lock_dummy: orm.Mapped[int] = orm.mapped_column(BigInteger, nullable=True)
 
 
@@ -200,6 +201,9 @@ class Table(Base):
     dir_id: orm.Mapped[uuid.UUID] = orm.mapped_column(UUID(as_uuid=True), ForeignKey('dirs.id'), nullable=False)
     md: orm.Mapped[dict[str, Any]] = orm.mapped_column(JSONB, nullable=False)  # TableMd
 
+    # used to force acquisition of an X-lock via an Update stmt
+    lock_dummy: orm.Mapped[int] = orm.mapped_column(BigInteger, nullable=True)
+
 
 @dataclasses.dataclass
 class TableVersionMd:

pixeltable/plan.py

@@ -289,7 +289,7 @@ class Planner:
 
         # create InMemoryDataNode for 'rows'
         plan: exec.ExecNode = exec.InMemoryDataNode(
-            TableVersionHandle(tbl.id, tbl.effective_version), rows, row_builder, tbl.next_rowid
+            TableVersionHandle(tbl.id, tbl.effective_version), rows, row_builder, tbl.next_row_id
         )
 
         media_input_col_info = [
@@ -385,7 +385,7 @@ class Planner:
 
         cls.__check_valid_columns(tbl.tbl_version.get(), recomputed_cols, 'updated in')
 
-        recomputed_base_cols = {col for col in recomputed_cols if col.tbl == tbl.tbl_version}
+        recomputed_base_cols = {col for col in recomputed_cols if col.tbl.id == tbl.tbl_version.id}
         copied_cols = [
             col
             for col in target.cols_by_id.values()
@@ -409,7 +409,7 @@ class Planner:
         for i, col in enumerate(all_base_cols):
             plan.row_builder.add_table_column(col, select_list[i].slot_idx)
         recomputed_user_cols = [c for c in recomputed_cols if c.name is not None]
-        return plan, [f'{c.tbl.get().name}.{c.name}' for c in updated_cols + recomputed_user_cols], recomputed_user_cols
+        return plan, [f'{c.tbl.name}.{c.name}' for c in updated_cols + recomputed_user_cols], recomputed_user_cols
 
     @classmethod
     def __check_valid_columns(
@@ -465,7 +465,7 @@ class Planner:
         recomputed_cols.update(idx_val_cols)
         # we only need to recompute stored columns (unstored ones are substituted away)
         recomputed_cols = {c for c in recomputed_cols if c.is_stored}
-        recomputed_base_cols = {col for col in recomputed_cols if col.tbl == target}
+        recomputed_base_cols = {col for col in recomputed_cols if col.tbl.id == target.id}
         copied_cols = [
             col
             for col in target.cols_by_id.values()

pixeltable/share/packager.py

@@ -7,6 +7,7 @@ import urllib.request
 import uuid
 from pathlib import Path
 from typing import Any, Iterator, Optional
+from uuid import UUID
 
 import more_itertools
 import pyarrow as pa
@@ -51,7 +52,7 @@ class TablePackager:
         self.media_files = {}
 
         # Load metadata
-        with Env.get().begin_xact():
+        with catalog.Catalog.get().begin_xact(for_write=False):
             tbl_md = catalog.Catalog.get().load_replica_md(table)
             self.md = {
                 'pxt_version': pxt.__version__,
@@ -66,15 +67,15 @@ class TablePackager:
         Export the table to a tarball containing Parquet tables and media files.
         """
         assert not self.tmp_dir.exists()  # Packaging can only be done once per TablePackager instance
-        _logger.info(f"Packaging table '{self.table._path}' and its ancestors in: {self.tmp_dir}")
+        _logger.info(f"Packaging table '{self.table._path()}' and its ancestors in: {self.tmp_dir}")
         self.tmp_dir.mkdir()
         with open(self.tmp_dir / 'metadata.json', 'w', encoding='utf8') as fp:
             json.dump(self.md, fp)
         self.tables_dir = self.tmp_dir / 'tables'
         self.tables_dir.mkdir()
-        with Env.get().begin_xact():
+        with catalog.Catalog.get().begin_xact(for_write=False):
             for tv in self.table._tbl_version_path.get_tbl_versions():
-                _logger.info(f"Exporting table '{tv.get().name}:{tv.get().version}'.")
+                _logger.info(f"Exporting table '{tv.get().versioned_name}'.")
                 self.__export_table(tv.get())
         _logger.info('Building archive.')
         bundle_path = self.__build_tarball()
@@ -253,13 +254,26 @@ class TableRestorer:
         tbl_md = [schema.FullTableMd.from_dict(t) for t in self.md['md']['tables']]
 
         # Create the replica table
-        # TODO: This needs to be made concurrency-safe.
-        replica_tbl = catalog.Catalog.get().create_replica(catalog.Path(self.tbl_path), tbl_md)
-        assert replica_tbl._tbl_version.get().is_snapshot
+        # The logic here needs to be completely restructured in order to make it concurrency-safe.
+        # - Catalog.create_replica() needs to write the metadata and also create the physical store tables
+        #   and populate them, otherwise concurrent readers will see an inconsistent state (table metadata w/o
+        #   an actual table)
+        # - this could be done one replica at a time (instead of the entire hierarchy)
+        cat = catalog.Catalog.get()
+        cat.create_replica(catalog.Path(self.tbl_path), tbl_md)
+        # don't call get_table() until after the calls to create_replica() and __import_table() below;
+        # the TV instances created by get_table() would be replaced by create_replica(), which creates duplicate
+        # TV instances for the same replica version, which then leads to failures when constructing queries
 
         # Now we need to instantiate and load data for replica_tbl and its ancestors, except that we skip
         # replica_tbl itself if it's a pure snapshot.
-        if replica_tbl._id != replica_tbl._tbl_version.id:
+        target_md = tbl_md[0]
+        is_pure_snapshot = (
+            target_md.tbl_md.view_md is not None
+            and target_md.tbl_md.view_md.predicate is None
+            and len(target_md.schema_version_md.columns) == 0
+        )
+        if is_pure_snapshot:
             ancestor_md = tbl_md[1:]  # Pure snapshot; skip replica_tbl
         else:
             ancestor_md = tbl_md  # Not a pure snapshot; include replica_tbl
@@ -273,7 +287,8 @@ class TableRestorer:
                 _logger.info(f'Importing table {tv.name!r}.')
                 self.__import_table(self.tmp_dir, tv, md)
 
-        return replica_tbl
+        with cat.begin_xact(for_write=False):
+            return cat.get_table_by_id(UUID(tbl_md[0].tbl_md.tbl_id))
 
     def __import_table(self, bundle_path: Path, tv: catalog.TableVersion, tbl_md: schema.FullTableMd) -> None:
         """

pixeltable/share/publish.py

@@ -35,7 +35,7 @@ def push_replica(dest_tbl_uri: str, src_tbl: pxt.Table) -> str:
     upload_id = response_json['upload_id']
     destination_uri = response_json['destination_uri']
 
-    Env.get().console_logger.info(f"Creating a snapshot of '{src_tbl._path}' at: {dest_tbl_uri}")
+    Env.get().console_logger.info(f"Creating a snapshot of '{src_tbl._path()}' at: {dest_tbl_uri}")
 
     bundle = packager.package()
 
@@ -117,7 +117,7 @@ def pull_replica(dest_path: str, src_tbl_uri: str) -> pxt.Table:
 
     restorer = TableRestorer(dest_path, response_json)
     tbl = restorer.restore(bundle_path)
-    Env.get().console_logger.info(f'Created local replica {tbl._path!r} from URI: {src_tbl_uri}')
+    Env.get().console_logger.info(f'Created local replica {tbl._path()!r} from URI: {src_tbl_uri}')
     return tbl
 
 

pixeltable/store.py

@@ -52,7 +52,8 @@ class StoreBase:
         # We need to declare a `base` variable here, even though it's only defined for instances of `StoreView`,
         # since it's referenced by various methods of `StoreBase`
         self.base = tbl_version.base.get().store_tbl if tbl_version.base is not None else None
-        self.create_sa_tbl()
+        # we're passing in tbl_version to avoid a circular call to TableVersionHandle.get()
+        self.create_sa_tbl(tbl_version)
 
     def system_columns(self) -> list[sql.Column]:
         return [*self._pk_cols, self.v_max_col]
@@ -77,11 +78,13 @@ class StoreBase:
         self._pk_cols = [*rowid_cols, self.v_min_col]
         return [*rowid_cols, self.v_min_col, self.v_max_col]
 
-    def create_sa_tbl(self) -> None:
+    def create_sa_tbl(self, tbl_version: Optional[catalog.TableVersion] = None) -> None:
         """Create self.sa_tbl from self.tbl_version."""
+        if tbl_version is None:
+            tbl_version = self.tbl_version.get()
         system_cols = self._create_system_columns()
         all_cols = system_cols.copy()
-        for col in [c for c in self.tbl_version.get().cols if c.is_stored]:
+        for col in [c for c in tbl_version.cols if c.is_stored]:
             # re-create sql.Column for each column, regardless of whether it already has sa_col set: it was bound
             # to the last sql.Table version we created and cannot be reused
             col.create_sa_cols()
@@ -99,16 +102,17 @@ class StoreBase:
         # - base x view joins can be executed as merge joins
         # - speeds up ORDER BY rowid DESC
         # - allows filtering for a particular table version in index scan
-        idx_name = f'sys_cols_idx_{self.tbl_version.id.hex}'
+        idx_name = f'sys_cols_idx_{tbl_version.id.hex}'
         idxs.append(sql.Index(idx_name, *system_cols))
 
         # v_min/v_max indices: speeds up base table scans needed to propagate a base table insert or delete
-        idx_name = f'vmin_idx_{self.tbl_version.id.hex}'
+        idx_name = f'vmin_idx_{tbl_version.id.hex}'
         idxs.append(sql.Index(idx_name, self.v_min_col, postgresql_using=Env.get().dbms.version_index_type))
-        idx_name = f'vmax_idx_{self.tbl_version.id.hex}'
+        idx_name = f'vmax_idx_{tbl_version.id.hex}'
         idxs.append(sql.Index(idx_name, self.v_max_col, postgresql_using=Env.get().dbms.version_index_type))
 
         self.sa_tbl = sql.Table(self._storage_name(), self.sa_md, *all_cols, *idxs)
+        # _logger.debug(f'created sa tbl for {tbl_version.id!s} (sa_tbl={id(self.sa_tbl):x}, tv={id(tbl_version):x})')
 
     @abc.abstractmethod
     def _rowid_join_predicate(self) -> sql.ColumnElement[bool]:
@@ -285,7 +289,7 @@ class StoreBase:
                         else:
                             if col.col_type.is_image_type() and result_row.file_urls[value_expr_slot_idx] is None:
                                 # we have yet to store this image
-                                filepath = str(MediaStore.prepare_media_path(col.tbl.id, col.id, col.tbl.get().version))
+                                filepath = str(MediaStore.prepare_media_path(col.tbl.id, col.id, col.tbl.version))
                                 result_row.flush_img(value_expr_slot_idx, filepath)
                             val = result_row.get_stored_val(value_expr_slot_idx, col.sa_col.type)
                             if col.col_type.is_media_type():
@@ -415,9 +419,7 @@ class StoreBase:
             number of deleted rows
         """
         where_clause = sql.true() if where_clause is None else where_clause
-        where_clause = sql.and_(
-            self.v_min_col < current_version, self.v_max_col == schema.Table.MAX_VERSION, where_clause
-        )
+        version_clause = sql.and_(self.v_min_col < current_version, self.v_max_col == schema.Table.MAX_VERSION)
         rowid_join_clause = self._rowid_join_predicate()
         base_versions_clause = (
             sql.true() if len(base_versions) == 0 else self.base._versions_clause(base_versions, match_on_vmin)
@@ -428,10 +430,12 @@ class StoreBase:
             set_clause[index_info.undo_col.sa_col] = index_info.val_col.sa_col
             # set value column to NULL
             set_clause[index_info.val_col.sa_col] = None
+
         stmt = (
             sql.update(self.sa_tbl)
             .values(set_clause)
             .where(where_clause)
+            .where(version_clause)
             .where(rowid_join_clause)
             .where(base_versions_clause)
         )
@@ -528,10 +532,12 @@ class StoreComponentView(StoreView):
         self.rowid_cols.append(self.pos_col)
         return self.rowid_cols
 
-    def create_sa_tbl(self) -> None:
-        super().create_sa_tbl()
+    def create_sa_tbl(self, tbl_version: Optional[catalog.TableVersion] = None) -> None:
+        if tbl_version is None:
+            tbl_version = self.tbl_version.get()
+        super().create_sa_tbl(tbl_version)
         # we need to fix up the 'pos' column in TableVersion
-        self.tbl_version.get().cols_by_name['pos'].sa_col = self.pos_col
+        tbl_version.cols_by_name['pos'].sa_col = self.pos_col
 
     def _rowid_join_predicate(self) -> sql.ColumnElement[bool]:
         return sql.and_(

pixeltable/utils/dbms.py

@@ -35,7 +35,7 @@ class PostgresqlDbms(Dbms):
     """
 
     def __init__(self, db_url: URL):
-        super().__init__('postgresql', 'REPEATABLE READ', 'brin', db_url)
+        super().__init__('postgresql', 'SERIALIZABLE', 'brin', db_url)
 
     def drop_db_stmt(self, database: str) -> str:
         return f'DROP DATABASE {database}'

{pixeltable-0.3.15.dist-info → pixeltable-0.4.0rc1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: pixeltable
-Version: 0.3.15
+Version: 0.4.0rc1
 Summary: AI Data Infrastructure: Declarative, Multimodal, and Incremental
 License: Apache-2.0
 Keywords: data-science,machine-learning,database,ai,computer-vision,chatbot,ml,artificial-intelligence,feature-engineering,multimodal,mlops,feature-store,vector-database,llm,genai