SCD2 support

README.md

@@ -113,4 +113,4 @@ The dlt project is quickly growing, and we're excited to have you join our commu
 
 ## License
 
-DLT is released under the [Apache 2.0 License](LICENSE.txt).
+`dlt` is released under the [Apache 2.0 License](LICENSE.txt).

dlt/cli/_dlt.py

@@ -5,7 +5,7 @@
 import click
 
 from dlt.version import __version__
-from dlt.common import json
+from dlt.common.json import json
 from dlt.common.schema import Schema
 from dlt.common.typing import DictStrAny
 from dlt.common.runners import Venv

dlt/cli/config_toml_writer.py

@@ -4,7 +4,7 @@
 from tomlkit.container import Container as TOMLContainer
 from collections.abc import Sequence as C_Sequence
 
-from dlt.common import pendulum
+from dlt.common.pendulum import pendulum
 from dlt.common.configuration.specs import (
     BaseConfiguration,
     is_base_configuration_inner_hint,

dlt/cli/pipeline_command.py

@@ -3,7 +3,7 @@
 import dlt
 from dlt.cli.exceptions import CliCommandException
 
-from dlt.common import json
+from dlt.common.json import json
 from dlt.common.pipeline import resource_state, get_dlt_pipelines_dir, TSourceState
 from dlt.common.destination.reference import TDestinationReferenceArg
 from dlt.common.runners import Venv

dlt/cli/utils.py

@@ -1,11 +1,8 @@
 import ast
 import os
-import tempfile
 from typing import Callable
 
-from dlt.common import git
 from dlt.common.reflection.utils import set_ast_parents
-from dlt.common.storages import FileStorage
 from dlt.common.typing import TFun
 from dlt.common.configuration import resolve_configuration
 from dlt.common.configuration.specs import RunConfiguration

dlt/common/configuration/providers/google_secrets.py

@@ -1,9 +1,8 @@
 import base64
 import string
 import re
-from typing import Tuple
 
-from dlt.common import json
+from dlt.common.json import json
 from dlt.common.configuration.specs import GcpServiceAccountCredentials
 from dlt.common.exceptions import MissingDependencyException
 from .toml import VaultTomlProvider

dlt/common/configuration/providers/toml.py

@@ -6,7 +6,7 @@
 from tomlkit.container import Container as TOMLContainer
 from typing import Any, Dict, Optional, Tuple, Type, Union
 
-from dlt.common import pendulum
+from dlt.common.pendulum import pendulum
 from dlt.common.configuration.paths import get_dlt_settings_dir, get_dlt_data_dir
 from dlt.common.configuration.utils import auto_cast
 from dlt.common.configuration.specs import known_sections

dlt/common/configuration/specs/azure_credentials.py

@@ -1,7 +1,6 @@
 from typing import Optional, Dict, Any
 
-from dlt.common import pendulum
-from dlt.common.exceptions import MissingDependencyException
+from dlt.common.pendulum import pendulum
 from dlt.common.typing import TSecretStrValue
 from dlt.common.configuration.specs import (
     CredentialsConfiguration,

dlt/common/configuration/specs/gcp_credentials.py

@@ -2,7 +2,8 @@
 import sys
 from typing import Any, ClassVar, Final, List, Tuple, Union, Dict
 
-from dlt.common import json, pendulum
+from dlt.common.json import json
+from dlt.common.pendulum import pendulum
 from dlt.common.configuration.specs.api_credentials import OAuth2Credentials
 from dlt.common.configuration.specs.exceptions import (
     InvalidGoogleNativeCredentialsType,

dlt/common/configuration/utils.py

@@ -5,7 +5,7 @@
 from typing import Any, Dict, Mapping, NamedTuple, Optional, Tuple, Type, Sequence
 from collections.abc import Mapping as C_Mapping
 
-from dlt.common import json
+from dlt.common.json import json
 from dlt.common.typing import AnyType, TAny
 from dlt.common.data_types import coerce_value, py_type_to_sc_type
 from dlt.common.configuration.providers import EnvironProvider
@@ -122,8 +122,6 @@ def log_traces(
     default_value: Any,
     traces: Sequence[LookupTrace],
 ) -> None:
-    from dlt.common import logger
-
     # if logger.is_logging() and logger.log_level() == "DEBUG" and config:
     #     logger.debug(f"Field {key} with type {hint} in {type(config).__name__} {'NOT RESOLVED' if value is None else 'RESOLVED'}")
     # print(f"Field {key} with type {hint} in {type(config).__name__} {'NOT RESOLVED' if value is None else 'RESOLVED'}")

dlt/common/data_types/type_helpers.py

@@ -3,13 +3,13 @@
 import dataclasses
 import datetime  # noqa: I251
 from collections.abc import Mapping as C_Mapping, Sequence as C_Sequence
-from typing import Any, Type, Literal, Union, cast
+from typing import Any, Type, Union
 from enum import Enum
 
-from dlt.common import pendulum, json, Decimal, Wei
 from dlt.common.json import custom_pua_remove, json
 from dlt.common.json._simplejson import custom_encode as json_custom_encode
-from dlt.common.arithmetics import InvalidOperation
+from dlt.common.wei import Wei
+from dlt.common.arithmetics import InvalidOperation, Decimal
 from dlt.common.data_types.typing import TDataType
 from dlt.common.time import (
     ensure_pendulum_datetime,

dlt/common/data_writers/escape.py

@@ -4,6 +4,8 @@
 from datetime import date, datetime, time  # noqa: I251
 
 from dlt.common.json import json
+from dlt.common.pendulum import pendulum
+from dlt.common.time import reduce_pendulum_datetime_precision
 
 # use regex to escape characters in single pass
 SQL_ESCAPE_DICT = {"'": "''", "\\": "\\\\", "\n": "\\n", "\r": "\\r"}
@@ -152,3 +154,17 @@ def escape_databricks_literal(v: Any) -> Any:
         return "NULL"
 
     return str(v)
+
+
+def format_datetime_literal(v: pendulum.DateTime, precision: int = 6, no_tz: bool = False) -> str:
+    """Converts `v` to ISO string, optionally without timezone spec (in UTC) and with given `precision`"""
+    if no_tz:
+        v = v.in_timezone(tz="UTC").replace(tzinfo=None)
+    v = reduce_pendulum_datetime_precision(v, precision)
+    # yet another precision translation
+    timespec: str = "microseconds"
+    if precision < 6:
+        timespec = "milliseconds"
+    elif precision < 3:
+        timespec = "seconds"
+    return v.isoformat(sep=" ", timespec=timespec)

dlt/common/data_writers/writers.py

@@ -17,7 +17,7 @@
     TypeVar,
 )
 
-from dlt.common import json
+from dlt.common.json import json
 from dlt.common.configuration import configspec, known_sections, with_config
 from dlt.common.configuration.specs import BaseConfiguration
 from dlt.common.data_writers.exceptions import DataWriterNotFound, InvalidDataItem
@@ -176,6 +176,9 @@ def writer_spec(cls) -> FileWriterSpec:
 
 class InsertValuesWriter(DataWriter):
     def __init__(self, f: IO[Any], caps: DestinationCapabilitiesContext = None) -> None:
+        assert (
+            caps is not None
+        ), "InsertValuesWriter requires destination capabilities to be present"
         super().__init__(f, caps)
         self._chunks_written = 0
         self._headers_lookup: Dict[str, int] = None
@@ -272,7 +275,7 @@ def __init__(
         coerce_timestamps: Optional[Literal["s", "ms", "us", "ns"]] = None,
         allow_truncated_timestamps: bool = False,
     ) -> None:
-        super().__init__(f, caps)
+        super().__init__(f, caps or DestinationCapabilitiesContext.generic_capabilities("parquet"))
         from dlt.common.libs.pyarrow import pyarrow
 
         self.writer: Optional[pyarrow.parquet.ParquetWriter] = None
@@ -287,7 +290,15 @@ def __init__(
         self.allow_truncated_timestamps = allow_truncated_timestamps
 
     def _create_writer(self, schema: "pa.Schema") -> "pa.parquet.ParquetWriter":
-        from dlt.common.libs.pyarrow import pyarrow
+        from dlt.common.libs.pyarrow import pyarrow, get_py_arrow_timestamp
+
+        # if timestamps are not explicitly coerced, use destination resolution
+        # TODO: introduce maximum timestamp resolution, using timestamp_precision too aggressive
+        # if not self.coerce_timestamps:
+        #     self.coerce_timestamps = get_py_arrow_timestamp(
+        #         self._caps.timestamp_precision, "UTC"
+        #     ).unit
+        #     self.allow_truncated_timestamps = True
 
         return pyarrow.parquet.ParquetWriter(
             self._f,
@@ -331,7 +342,9 @@ def write_data(self, rows: Sequence[Any]) -> None:
         for key in self.complex_indices:
             for row in rows:
                 if (value := row.get(key)) is not None:
-                    row[key] = json.dumps(value)
+                    # TODO: make this configurable
+                    if value is not None and not isinstance(value, str):
+                        row[key] = json.dumps(value)
 
         table = pyarrow.Table.from_pylist(rows, schema=self.schema)
         # Write

dlt/common/destination/reference.py

@@ -26,6 +26,7 @@
 
 from dlt.common import logger
 from dlt.common.schema import Schema, TTableSchema, TSchemaTables
+from dlt.common.schema.typing import MERGE_STRATEGIES
 from dlt.common.schema.exceptions import SchemaException
 from dlt.common.schema.utils import (
     get_write_disposition,
@@ -344,6 +345,12 @@ def _verify_schema(self) -> None:
                     table_name,
                     self.capabilities.max_identifier_length,
                 )
+            if table.get("write_disposition") == "merge":
+                if "x-merge-strategy" in table and table["x-merge-strategy"] not in MERGE_STRATEGIES:  # type: ignore[typeddict-item]
+                    raise SchemaException(
+                        f'"{table["x-merge-strategy"]}" is not a valid merge strategy. '  # type: ignore[typeddict-item]
+                        f"""Allowed values: {', '.join(['"' + s + '"' for s in MERGE_STRATEGIES])}."""
+                    )
             if has_column_with_prop(table, "hard_delete"):
                 if len(get_columns_names_with_prop(table, "hard_delete")) > 1:
                     raise SchemaException(

dlt/common/json/__init__.py

@@ -12,7 +12,7 @@
 except ImportError:
     PydanticBaseModel = None  # type: ignore[misc]
 
-from dlt.common import pendulum
+from dlt.common.pendulum import pendulum
 from dlt.common.arithmetics import Decimal
 from dlt.common.wei import Wei
 from dlt.common.utils import map_nested_in_place
@@ -99,19 +99,19 @@ def _datetime_decoder(obj: str) -> datetime:
         # Backwards compatibility for data encoded with previous dlt version
         # fromisoformat does not support Z suffix (until py3.11)
         obj = obj[:-1] + "+00:00"
-    return pendulum.DateTime.fromisoformat(obj)  # type: ignore[attr-defined, no-any-return]
+    return pendulum.DateTime.fromisoformat(obj)
 
 
 # define decoder for each prefix
 DECODERS: List[Callable[[Any], Any]] = [
     Decimal,
     _datetime_decoder,
-    pendulum.Date.fromisoformat,  # type: ignore[attr-defined]
+    pendulum.Date.fromisoformat,
     UUID,
     HexBytes,
     base64.b64decode,
     Wei,
-    pendulum.Time.fromisoformat,  # type: ignore[attr-defined]
+    pendulum.Time.fromisoformat,
 ]
 # how many decoders?
 PUA_CHARACTER_MAX = len(DECODERS)

dlt/common/libs/numpy.py

@@ -3,4 +3,4 @@
 try:
     import numpy
 except ModuleNotFoundError:
-    raise MissingDependencyException("DLT Numpy Helpers", ["numpy"])
+    raise MissingDependencyException("dlt Numpy Helpers", ["numpy"])

dlt/common/libs/pandas.py

@@ -4,7 +4,7 @@
 try:
     import pandas
 except ModuleNotFoundError:
-    raise MissingDependencyException("DLT Pandas Helpers", ["pandas"])
+    raise MissingDependencyException("dlt Pandas Helpers", ["pandas"])
 
 
 def pandas_to_arrow(df: pandas.DataFrame) -> Any:

dlt/common/libs/pyarrow.py

@@ -15,7 +15,7 @@
 )
 
 from dlt import version
-from dlt.common import pendulum
+from dlt.common.pendulum import pendulum
 from dlt.common.exceptions import MissingDependencyException
 from dlt.common.schema.typing import DLT_NAME_PREFIX, TTableSchemaColumns
 
@@ -119,7 +119,7 @@ def get_pyarrow_int(precision: Optional[int]) -> Any:
     return pyarrow.int64()
 
 
-def _get_column_type_from_py_arrow(dtype: pyarrow.DataType) -> TColumnType:
+def get_column_type_from_py_arrow(dtype: pyarrow.DataType) -> TColumnType:
     """Returns (data_type, precision, scale) tuple from pyarrow.DataType"""
     if pyarrow.types.is_string(dtype) or pyarrow.types.is_large_string(dtype):
         return dict(data_type="text")
@@ -226,14 +226,14 @@ def should_normalize_arrow_schema(
 ) -> Tuple[bool, Mapping[str, str], Dict[str, str], TTableSchemaColumns]:
     rename_mapping = get_normalized_arrow_fields_mapping(schema, naming)
     rev_mapping = {v: k for k, v in rename_mapping.items()}
-    dlt_table_prefix = naming.normalize_table_identifier(DLT_NAME_PREFIX)
+    dlt_tables = list(map(naming.normalize_table_identifier, ("_dlt_id", "_dlt_load_id")))
 
     # remove all columns that are dlt columns but are not present in arrow schema. we do not want to add such columns
     # that should happen in the normalizer
     columns = {
         name: column
         for name, column in columns.items()
-        if not name.startswith(dlt_table_prefix) or name in rev_mapping
+        if name not in dlt_tables or name in rev_mapping
     }
 
     # check if nothing to rename
@@ -322,7 +322,7 @@ def py_arrow_to_table_schema_columns(schema: pyarrow.Schema) -> TTableSchemaColu
         result[field.name] = {
             "name": field.name,
             "nullable": field.nullable,
-            **_get_column_type_from_py_arrow(field.type),
+            **get_column_type_from_py_arrow(field.type),
         }
     return result