openmetadata-ingestion 1.5.0.0rc2__py3-none-any.whl → 1.5.2.0__py3-none-any.whl

metadata/ingestion/source/database/saphana/cdata_parser.py

@@ -0,0 +1,772 @@
+#  Copyright 2021 Collate
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#  http://www.apache.org/licenses/LICENSE-2.0
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+"""
+Parse CDATA XMLs from SAP Hana
+"""
+import itertools
+import re
+import traceback
+import xml.etree.ElementTree as ET
+from collections import defaultdict
+from enum import Enum
+from functools import lru_cache
+from typing import Dict, Iterable, List, NewType, Optional, Set
+
+from pydantic import Field, computed_field
+from typing_extensions import Annotated
+
+from metadata.generated.schema.api.lineage.addLineage import AddLineageRequest
+from metadata.generated.schema.entity.data.table import Table
+from metadata.generated.schema.entity.services.ingestionPipelines.status import (
+    StackTraceError,
+)
+from metadata.generated.schema.type.basic import FullyQualifiedEntityName
+from metadata.generated.schema.type.entityLineage import (
+    ColumnLineage,
+    EntitiesEdge,
+    LineageDetails,
+    Source,
+)
+from metadata.generated.schema.type.entityReference import EntityReference
+from metadata.ingestion.api.models import Either
+from metadata.ingestion.lineage.sql_lineage import get_column_fqn
+from metadata.ingestion.models.custom_pydantic import BaseModel
+from metadata.ingestion.ometa.ometa_api import OpenMetadata
+from metadata.ingestion.source.database.saphana.models import (
+    SYS_BIC_SCHEMA_NAME,
+    ViewType,
+)
+from metadata.utils import fqn
+from metadata.utils.constants import ENTITY_REFERENCE_TYPE_MAP
+from metadata.utils.dispatch import enum_register
+from metadata.utils.logger import ingestion_logger
+
+logger = ingestion_logger()
+
+
+class CDATAParsingError(Exception):
+    """Error parsing CDATA XML"""
+
+
+XSI_NS = {"xsi": "http://www.w3.org/2001/XMLSchema-instance"}
+
+NAMESPACE_DICT = {
+    ViewType.ANALYTIC_VIEW.value: {
+        "Cube": "http://www.sap.com/ndb/BiModelCube.ecore",
+        **XSI_NS,
+    },
+    ViewType.CALCULATION_VIEW.value: {
+        "Calculation": "http://www.sap.com/ndb/BiModelCalculation.ecore",
+        **XSI_NS,
+    },
+    ViewType.ATTRIBUTE_VIEW.value: {
+        "Dimension": "http://www.sap.com/ndb/BiModelDimension.ecore",
+        **XSI_NS,
+    },
+}
+
+FORMULA_PATTERN = re.compile(r"\"(.*?)\"")
+
+
+class CDATAKeys(Enum):
+    """Keys to access data in CDATA XML files"""
+
+    COLUMN_OBJECT_NAME = "columnObjectName"
+    COLUMN_OBJECT = "columnObject"
+    COLUMN_NAME = "columnName"
+    SCHEMA_NAME = "schemaName"
+    ATTRIBUTES = "attributes"
+    ATTRIBUTE = "attribute"
+    KEY_MAPPING = "keyMapping"
+    MAPPING = "mapping"
+    SOURCE = "source"
+    TARGET = "target"
+    NODE = "node"
+    TYPE = "type"
+    INPUT = "input"
+    CALCULATION_VIEWS = "calculationViews"
+    CALCULATION_VIEW = "calculationView"
+    CALCULATION_VIEW_ATTRIBUTES = "calculatedViewAttributes"
+    CALCULATION_VIEW_ATTRIBUTE = "calculatedViewAttribute"
+    RESOURCE_URI = "resourceUri"
+    CALCULATED_ATTRS = "calculatedAttributes"
+    KEY_CALCULATION = "keyCalculation"
+    FORMULA = "formula"
+    BASE_MEASURES = "baseMeasures"
+    MEASURE = "measure"
+    MEASURE_MAPPING = "measureMapping"
+    PRIVATE_MEASURE_GROUP = "privateMeasureGroup"
+    LOGICAL_MODEL = "logicalModel"
+    DATA_SOURCES = "dataSources"
+    DATA_SOURCE = "DataSource"  # yes, with capital D
+    ID = "id"
+
+
+class CalculatedAttrKey(Enum):
+    CALCULATED_ATTRIBUTE = "calculatedAttribute"
+    CALCULATED_VIEW_ATTRIBUTE = "calculatedViewAttribute"
+
+
+class ParentSource(BaseModel):
+    """Parent Source of a given column"""
+
+    # TODO: Multiple sources from the same parent should be possible
+    source: Annotated[
+        str, Field(..., description="Column name in the parent Data Source")
+    ]
+    parent: Annotated[str, Field(..., description="Parent ID")]
+
+
+class DataSourceMapping(BaseModel):
+    """Column Mapping of DataSources and Logical Calculated Views"""
+
+    target: Annotated[
+        str, Field(..., description="Column name in the provided Data Source")
+    ]
+    parents: Annotated[
+        List[ParentSource], Field(..., description="Parent Sources for a target col")
+    ]
+
+
+class DataSource(BaseModel):
+    """Data source from CDATA XML"""
+
+    name: Annotated[str, Field(..., description="Data Source name")]
+    location: Annotated[
+        Optional[str], Field(None, description="Schema or project for the Data Source")
+    ]
+    source_type: Annotated[
+        Optional[ViewType],
+        Field(..., description="Data Source type"),
+    ]
+    mapping: Annotated[
+        Optional[Dict[str, DataSourceMapping]],
+        Field(
+            None,
+            description="Logical source column mapping. Key: source column; value: mapping",
+        ),
+    ]
+
+    def get_entity(
+        self,
+        metadata: OpenMetadata,
+        service_name: str,
+    ) -> Table:
+        """Build the Entity Reference for this DataSource"""
+
+        if self.source_type == ViewType.LOGICAL:
+            raise CDATAParsingError(
+                f"We could not find the logical DataSource origin for {self.name}"
+            )
+
+        if self.source_type == ViewType.DATA_BASE_TABLE:
+            # The source is a table, so the location is the schema
+            fqn_ = fqn.build(
+                metadata=metadata,
+                entity_type=Table,
+                service_name=service_name,
+                database_name=None,  # TODO: Can we assume HXE?
+                schema_name=self.location,
+                table_name=self.name,
+            )
+        else:
+            # The source is a CalculationView, AttributeView or AnalyticView
+            # package from <resourceUri>/SFLIGHT.MODELING/calculationviews/CV_SFLIGHT_SBOOK</resourceUri>
+            package = self.location.split("/")[1]
+            fqn_ = fqn.build(
+                metadata=metadata,
+                entity_type=Table,
+                service_name=service_name,
+                database_name=None,
+                schema_name=SYS_BIC_SCHEMA_NAME,
+                table_name=f"{package}/{self.name}",
+            )
+
+        return metadata.get_by_name(entity=Table, fqn=fqn_)
+
+    def __hash__(self):
+        return hash(self.location) + hash(self.name) + hash(self.source_type)
+
+
+# Given the DataSource ID, get the DataSource from the CDATA XML
+DataSourceMap = NewType("DataSourceMap", Dict[str, DataSource])
+
+
+class ColumnMapping(BaseModel):
+    """Column Mapping from CDATA XML"""
+
+    data_source: Annotated[DataSource, Field(..., description="Source table name")]
+    sources: Annotated[List[str], Field(..., description="Source column names")]
+    target: Annotated[str, Field(..., description="Destination column name")]
+    formula: Annotated[
+        Optional[str], Field(None, description="Formula used to derive the column")
+    ]
+
+
+class ParsedLineage(BaseModel):
+    """Parsed Lineage from CDATA XML. For each view, we'll parse the sources"""
+
+    mappings: Annotated[
+        Optional[List[ColumnMapping]], Field([], description="Column mappings")
+    ]
+
+    @computed_field
+    @property
+    def sources(self) -> Set[DataSource]:
+        """Get all the different source tables we'll need to iterate over"""
+        return {mapping.data_source for mapping in self.mappings}
+
+    @lru_cache(maxsize=256)
+    def find_target(self, column: str) -> Optional[ColumnMapping]:
+        """Find the column mapping based on the target column"""
+        return next(
+            (mapping for mapping in self.mappings if mapping.target == column), None
+        )
+
+    def __add__(self, other: "ParsedLineage") -> "ParsedLineage":
+        """Merge two parsed lineages"""
+        return ParsedLineage(mappings=self.mappings + other.mappings)
+
+    def __hash__(self):
+        """
+        Note that the LRU Cache require us to implement the __hash__ method, otherwise
+        the BaseModel is not hashable. Since we just want a per-instance cache, we'll use the id
+        """
+        return id(self)
+
+    def to_request(
+        self, metadata: OpenMetadata, service_name: str, to_entity: Table
+    ) -> Iterable[Either[AddLineageRequest]]:
+        """Given the target entity, build the AddLineageRequest based on the sources in `self`"""
+        for source in self.sources:
+            try:
+                source_table = source.get_entity(
+                    metadata=metadata, service_name=service_name
+                )
+                if not source_table:
+                    logger.warning(f"Can't find table for source [{source}]")
+                    continue
+                yield Either(
+                    right=AddLineageRequest(
+                        edge=EntitiesEdge(
+                            fromEntity=EntityReference(
+                                id=source_table.id,
+                                type=ENTITY_REFERENCE_TYPE_MAP[Table.__name__],
+                            ),
+                            toEntity=EntityReference(
+                                id=to_entity.id,
+                                type=ENTITY_REFERENCE_TYPE_MAP[Table.__name__],
+                            ),
+                            lineageDetails=LineageDetails(
+                                source=Source.ViewLineage,
+                                columnsLineage=[
+                                    ColumnLineage(
+                                        fromColumns=[
+                                            FullyQualifiedEntityName(
+                                                get_column_fqn(
+                                                    table_entity=source_table,
+                                                    column=source_col,
+                                                )
+                                            )
+                                            for source_col in mapping.sources
+                                        ],
+                                        toColumn=FullyQualifiedEntityName(
+                                            get_column_fqn(
+                                                table_entity=to_entity,
+                                                column=mapping.target,
+                                            )
+                                        ),
+                                        function=mapping.formula
+                                        if mapping.formula
+                                        else None,
+                                    )
+                                    for mapping in self.mappings
+                                    if mapping.data_source == source
+                                ],
+                            ),
+                        )
+                    )
+                )
+            except Exception as exc:
+                yield Either(
+                    left=StackTraceError(
+                        name=to_entity.fullyQualifiedName.root,
+                        error=f"Error trying to get lineage for [{source}] due to [{exc}]",
+                        stackTrace=traceback.format_exc(),
+                    )
+                )
+
+
+def _get_column_datasources(
+    entry: ET.Element, datasource_map: Optional[DataSourceMap] = None
+) -> Set[DataSource]:
+    """Read a DataSource from the CDATA XML"""
+    if (
+        datasource_map
+        and entry.get(CDATAKeys.COLUMN_OBJECT_NAME.value) in datasource_map
+    ):
+        # If the datasource is in the map, we'll traverse all intermediate logical
+        # datasources until we arrive to a table or view.
+        # Note that we can have multiple sources for a single column, e.g., columns
+        # coming from a JOIN
+        return set(
+            _traverse_ds(
+                current_column=entry.get(CDATAKeys.COLUMN_NAME.value),
+                ds_origin_list=[],
+                current_ds=datasource_map[
+                    entry.get(CDATAKeys.COLUMN_OBJECT_NAME.value)
+                ],
+                datasource_map=datasource_map,
+            )
+        )
+
+    # If we don't have any logical sources (projections, aggregations, etc.) We'll stick to
+    # a single table origin
+    return {
+        DataSource(
+            name=entry.get(CDATAKeys.COLUMN_OBJECT_NAME.value),
+            location=entry.get(CDATAKeys.SCHEMA_NAME.value),
+            source_type=ViewType.DATA_BASE_TABLE,
+        )
+    }
+
+
+def _traverse_ds(
+    current_column: str,
+    ds_origin_list: List[DataSource],
+    current_ds: DataSource,
+    datasource_map: Optional[DataSourceMap],
+) -> List[DataSource]:
+    """
+    Traverse the ds dict jumping from target -> source columns and getting the right parent.
+    We keep inspecting current datasources and will append to the origin list the ones
+    that are not LOGICAL
+    """
+    if current_ds.source_type != ViewType.LOGICAL:
+        ds_origin_list.append(current_ds)
+
+    else:
+        # Based on our current column, find the parents from the mappings in the current_ds
+        current_ds_mapping: DataSourceMapping = current_ds.mapping.get(current_column)
+
+        if current_ds_mapping:
+            for parent in current_ds_mapping.parents:
+                parent_ds = datasource_map.get(parent.parent)
+                if not parent_ds:
+                    raise CDATAParsingError(
+                        f"Can't find parent [{parent.parent}] for column [{current_column}]"
+                    )
+
+                # Traverse from the source column in the parent mapping
+                _traverse_ds(
+                    current_column=parent.source,
+                    ds_origin_list=ds_origin_list,
+                    current_ds=parent_ds,
+                    datasource_map=datasource_map,
+                )
+        else:
+            logger.info(
+                f"Can't find mapping for column [{current_column}] in [{current_ds}]. "
+                f"We still have to implement `calculatedViewAttributes`."
+            )
+
+    return ds_origin_list
+
+
+def _read_attributes(
+    tree: ET.Element, ns: dict, datasource_map: Optional[DataSourceMap] = None
+) -> ParsedLineage:
+    """Compute the lineage based from the attributes"""
+    lineage = ParsedLineage()
+    attribute_list = tree.find(CDATAKeys.ATTRIBUTES.value, ns) if tree else None
+    if not attribute_list:
+        return lineage
+
+    for attribute in attribute_list.findall(CDATAKeys.ATTRIBUTE.value, ns):
+        key_mapping = attribute.find(CDATAKeys.KEY_MAPPING.value, ns)
+        data_sources = _get_column_datasources(
+            entry=key_mapping, datasource_map=datasource_map
+        )
+        attr_lineage = ParsedLineage(
+            mappings=[
+                ColumnMapping(
+                    data_source=ds,
+                    sources=[key_mapping.get(CDATAKeys.COLUMN_NAME.value)],
+                    target=attribute.get(CDATAKeys.ID.value),
+                )
+                for ds in data_sources
+            ]
+        )
+        lineage += attr_lineage
+
+    return lineage
+
+
+def _read_calculated_attributes(
+    tree: ET.Element,
+    ns: dict,
+    base_lineage: ParsedLineage,
+    key: CalculatedAttrKey = CalculatedAttrKey.CALCULATED_ATTRIBUTE,
+) -> ParsedLineage:
+    """Compute the lineage based on the calculated attributes"""
+    lineage = ParsedLineage()
+
+    calculated_attrs = tree.find(CDATAKeys.CALCULATED_ATTRS.value, ns)
+    if not calculated_attrs:
+        return lineage
+
+    for calculated_attr in calculated_attrs.findall(key.value, ns):
+        formula = (
+            calculated_attr.find(CDATAKeys.KEY_CALCULATION.value, ns)
+            .find(CDATAKeys.FORMULA.value, ns)
+            .text
+        )
+        lineage += _explode_formula(
+            target=calculated_attr.get(CDATAKeys.ID.value),
+            formula=formula,
+            base_lineage=base_lineage,
+        )
+
+    return lineage
+
+
+def _read_base_measures(
+    tree: ET.Element, ns: dict, datasource_map: Optional[DataSourceMap] = None
+) -> ParsedLineage:
+    """
+    Compute the lineage based on the base measures.
+    For CalculationViews, we have a dictionary of pre-defined DataSources. For the rest,
+    we'll default to Table DataSources with the given information in the measure.
+
+    See examples cdata_calculation_view.xml and cdata_attribute_view.xml in test resources.
+    """
+    lineage = ParsedLineage()
+
+    base_measures = tree.find(CDATAKeys.BASE_MEASURES.value, ns)
+    if not base_measures:
+        return lineage
+
+    for measure in base_measures.findall(CDATAKeys.MEASURE.value, ns):
+        measure_mapping = measure.find(CDATAKeys.MEASURE_MAPPING.value, ns)
+        data_sources = _get_column_datasources(
+            entry=measure_mapping, datasource_map=datasource_map
+        )
+        measure_lineage = ParsedLineage(
+            mappings=[
+                ColumnMapping(
+                    data_source=ds,
+                    sources=[measure_mapping.get(CDATAKeys.COLUMN_NAME.value)],
+                    target=measure.get(CDATAKeys.ID.value),
+                )
+                for ds in data_sources
+            ]
+        )
+        lineage += measure_lineage
+
+    return lineage
+
+
+def _explode_formula(
+    target: str, formula: str, base_lineage: ParsedLineage
+) -> ParsedLineage:
+    """
+    Explode the formula and extract the columns
+    Args:
+        formula: formula to extract involved columns from
+        base_lineage: parsed lineage of the main attributes. We'll use this to pick up the original lineage columns
+    Returns:
+        Parsed Lineage from the formula
+    """
+    column_ds = {
+        match.group(1): base_lineage.find_target(match.group(1)).data_source
+        for match in FORMULA_PATTERN.finditer(formula)
+    }
+
+    # Group every datasource (key) with a list of the involved columns (values)
+    ds_columns = defaultdict(list)
+    for column, ds in column_ds.items():
+        ds_columns[ds].append(column)
+
+    return ParsedLineage(
+        mappings=[
+            ColumnMapping(
+                # We get the source once we find the mapping of the target
+                data_source=data_source,
+                sources=columns,
+                target=target,
+                formula=formula,
+            )
+            for data_source, columns in ds_columns.items()
+        ]
+    )
+
+
+parse_registry = enum_register()
+
+
+@parse_registry.add(ViewType.ANALYTIC_VIEW.value)
+def _(cdata: str) -> ParsedLineage:
+    """Parse the CDATA XML for Analytics View"""
+    ns = NAMESPACE_DICT[ViewType.ANALYTIC_VIEW.value]
+    tree = ET.fromstring(cdata)
+    measure_group = tree.find(CDATAKeys.PRIVATE_MEASURE_GROUP.value, ns)
+    # TODO: Handle lineage from calculatedMeasures, restrictedMeasures and sharedDimensions
+    return _read_attributes(measure_group, ns)
+
+
+@parse_registry.add(ViewType.ATTRIBUTE_VIEW.value)
+def _(cdata: str) -> ParsedLineage:
+    """Parse the CDATA XML for Analytics View"""
+    ns = NAMESPACE_DICT[ViewType.ATTRIBUTE_VIEW.value]
+    tree = ET.fromstring(cdata)
+    attribute_lineage = _read_attributes(tree=tree, ns=ns)
+    calculated_attrs_lineage = _read_calculated_attributes(
+        tree=tree, ns=ns, base_lineage=attribute_lineage
+    )
+    base_measure_lineage = _read_base_measures(tree=tree, ns=ns, datasource_map=None)
+
+    return attribute_lineage + calculated_attrs_lineage + base_measure_lineage
+
+
+@parse_registry.add(ViewType.CALCULATION_VIEW.value)
+def _(cdata: str) -> ParsedLineage:
+    """
+    Parse the CDATA XML for Calculation View
+
+    We can think of the DataSources in a CV as:
+    - "real" sources: they are tables or other views that are used in the calculation
+    - "logical" sources: as internal projections, aggregations, etc.
+
+    The behavior is:
+    - The LogicalModel attributes can be linked - transitively - to either a "real" or a "logical" source.
+    - The "logical" sources are defined inside each `calculationView`
+    - Each `calculationView` can have an `input` node that can either be "real" or "logical"
+    - We can identify "real" inputs as their `<input node="#<ID>">` will match the <DataSource id="ID" type="...">
+
+    When building the dataSources here we need to ensure we iterate over the calculationViews, and during
+    the _read_ds we'll iteratively traverse the dataSources to find the actual table.
+
+    Internally, we'll identify "logical" dataSources by giving them a list of column mappings, which we'll
+    use to identify the actual source.
+    """
+    # TODO: Handle lineage from calculatedMeasure, restrictedMeasure and sharedDimesions
+    ns = NAMESPACE_DICT[ViewType.CALCULATION_VIEW.value]
+    tree = ET.fromstring(cdata)
+
+    # Prepare a dictionary of defined data sources
+    datasource_map = _parse_cv_data_sources(tree=tree, ns=ns)
+
+    # Iterate over the Logical Model attributes
+    logical_model = tree.find(CDATAKeys.LOGICAL_MODEL.value, ns)
+    attribute_lineage = _read_attributes(
+        tree=logical_model, ns=ns, datasource_map=datasource_map
+    )
+    calculated_attrs_lineage = _read_calculated_attributes(
+        tree=tree,
+        ns=ns,
+        base_lineage=attribute_lineage,
+        key=CalculatedAttrKey.CALCULATED_VIEW_ATTRIBUTE,
+    )
+    base_measure_lineage = _read_base_measures(
+        tree=logical_model, ns=ns, datasource_map=datasource_map
+    )
+
+    return attribute_lineage + calculated_attrs_lineage + base_measure_lineage
+
+
+def _parse_cv_data_sources(tree: ET.Element, ns: dict) -> DataSourceMap:
+    """
+    Parse the real and logical data sources of a CV
+
+    The logical (`calculationViews`) have the following shape:
+    ```
+    <calculationViews>
+    <calculationView xsi:type="Calculation:AggregationView" id="Aggregation_1">
+      <descriptions/>
+      <viewAttributes>
+        <viewAttribute id="MANDT"/>
+        ...
+      </viewAttributes>
+      <calculatedViewAttributes>
+        <calculatedViewAttribute datatype="INTEGER" id="USAGE_PCT" expressionLanguage="COLUMN_ENGINE">
+          <formula>&quot;SEATSOCC_ALL&quot;/&quot;SEATSMAX_ALL&quot;</formula>
+        </calculatedViewAttribute>
+        ...
+      </calculatedViewAttributes>
+      <input node="#AT_SFLIGHT">
+        <mapping xsi:type="Calculation:AttributeMapping" target="MANDT" source="MANDT"/>
+        ...
+      </input>
+      <input emptyUnionBehavior="NO_ROW" node="#Projection_1">
+        <mapping xsi:type="Calculation:ConstantAttributeMapping" target="CARRNAME" null="true" value=""/>
+        ...
+      </input>
+      ...
+    </calculationView>
+    ...
+    </calculationViews>
+    ```
+    """
+    datasource_map = DataSourceMap({})
+    for ds in tree.find(CDATAKeys.DATA_SOURCES.value, ns).findall(
+        CDATAKeys.DATA_SOURCE.value, ns
+    ):
+        column_object = ds.find(CDATAKeys.COLUMN_OBJECT.value, ns)
+        # we can't rely on the falsy value of the object even if present in the XML
+        # If columnObject is informed, we're talking about a table
+        if column_object is not None:
+            ds_value = DataSource(
+                name=column_object.get(CDATAKeys.COLUMN_OBJECT_NAME.value),
+                location=column_object.get(CDATAKeys.SCHEMA_NAME.value),
+                source_type=ViewType.DATA_BASE_TABLE,
+            )
+        # or a package object
+        else:
+            ds_value = DataSource(
+                name=ds.get(CDATAKeys.ID.value),
+                location=ds.find(CDATAKeys.RESOURCE_URI.value).text,
+                source_type=ViewType.__members__[ds.get(CDATAKeys.TYPE.value)],
+            )
+        datasource_map[ds.get(CDATAKeys.ID.value)] = ds_value
+
+    calculation_views = tree.find(CDATAKeys.CALCULATION_VIEWS.value, ns)
+    if calculation_views is None:
+        return datasource_map
+
+    for cv in calculation_views.findall(CDATAKeys.CALCULATION_VIEW.value, ns):
+        mappings = _build_mappings(calculation_view=cv, ns=ns)
+        datasource_map[cv.get(CDATAKeys.ID.value)] = DataSource(
+            name=cv.get(CDATAKeys.ID.value),
+            location=None,
+            mapping={mapping.target: mapping for mapping in mappings},
+            source_type=ViewType.LOGICAL,
+        )
+
+    return datasource_map
+
+
+def _build_mappings(calculation_view: ET.Element, ns: dict) -> List[DataSourceMapping]:
+    """
+    Build the DataSourceMappings from each `input` inside a Calculation View tree.
+
+    Note how we can have:
+    ```
+    <input emptyUnionBehavior="NO_ROW" node="#Aggregation_1">
+        <mapping xsi:type="Calculation:AttributeMapping" target="MANDT" source="MANDT"/>
+        ...
+    </input>
+    <input emptyUnionBehavior="NO_ROW" node="#Projection_1">
+        <mapping xsi:type="Calculation:AttributeMapping" target="MANDT" source="MANDT"/>
+        ...
+    </input>
+    ```
+    Where a single target column `MANDT` comes from multiple sources. We need to consider
+    this when building the `parent` field in DataSourceMapping.
+
+    1. First, create a single list of all the mappings from all the inputs independently
+    2. Then, group by `target` and listagg the `parent`s
+
+    TODO: We still need to take care of mappings without source, since those come from
+      `calculatedViewAttributes` where we should handle the formula. Check `cdata_calculation_view.xml`
+      and take the `USAGE_PCT` as an example.
+    """
+
+    input_mappings = _build_input_mappings(calculation_view=calculation_view, ns=ns)
+    # calculated_view_attrs = _build_cv_attributes(
+    #     calculation_view=calculation_view, ns=ns, input_mappings=input_mappings
+    # )
+
+    return input_mappings
+
+
+def _build_input_mappings(
+    calculation_view: ET.Element, ns: dict
+) -> List[DataSourceMapping]:
+    """Map input nodes"""
+    mappings = []
+    for input_node in calculation_view.findall(CDATAKeys.INPUT.value, ns):
+        for mapping in input_node.findall(CDATAKeys.MAPPING.value, ns):
+            if mapping.get(CDATAKeys.SOURCE.value) and mapping.get(
+                CDATAKeys.TARGET.value
+            ):
+                mappings.append(
+                    DataSourceMapping(
+                        target=mapping.get(CDATAKeys.TARGET.value),
+                        parents=[
+                            ParentSource(
+                                source=mapping.get(CDATAKeys.SOURCE.value),
+                                parent=input_node.get(CDATAKeys.NODE.value).replace(
+                                    "#", ""
+                                ),
+                            )
+                        ],
+                    )
+                )
+
+    return _group_mappings(mappings)
+
+
+def _build_cv_attributes(
+    calculation_view: ET.Element, ns: dict, input_mappings: List[DataSourceMapping]
+) -> List[DataSourceMapping]:
+    """Extract mapping from `calculatedViewAttribute` formulas"""
+    mappings = []
+    view_attrs = calculation_view.find(CDATAKeys.CALCULATION_VIEW_ATTRIBUTES.value, ns)
+    if view_attrs is None:
+        return mappings
+
+    for view_attr in view_attrs.findall(CDATAKeys.CALCULATION_VIEW_ATTRIBUTE.value, ns):
+        formula = (
+            view_attr.find(CDATAKeys.FORMULA.value, ns).text
+            if view_attr.find(CDATAKeys.FORMULA.value, ns) is not None
+            else None
+        )
+        if not formula:
+            logger.debug(f"Skipping formula without expression at {view_attr}")
+            continue
+
+        involved_columns = FORMULA_PATTERN.findall(formula)
+        for col in involved_columns:
+            # Find the mapping for the involved column
+            mapping = next(
+                (mapping for mapping in input_mappings if mapping.target == col), None
+            )
+            if not mapping:
+                logger.debug(
+                    f"Can't find mapping for column [{col}] in [{input_mappings}]"
+                )
+                continue
+
+            mappings.append(
+                DataSourceMapping(
+                    target=view_attr.get(CDATAKeys.ID.value),
+                    parents=mapping.parents,
+                )
+            )
+
+    return _group_mappings(mappings)
+
+
+def _group_mappings(mappings: List[DataSourceMapping]) -> List[DataSourceMapping]:
+    """Group the mappings by target column and listagg the parents"""
+    # Sort the data by the target field
+    mappings.sort(key=lambda x: x.target)
+
+    # Use groupby to group by the target field
+    grouped_data = [
+        DataSourceMapping(
+            target=target,
+            parents=list(itertools.chain.from_iterable(item.parents for item in group)),
+        )
+        for target, group in itertools.groupby(mappings, key=lambda x: x.target)
+    ]
+
+    return grouped_data