Files
govoplan-dataflow/src/govoplan_dataflow/backend/graph.py
T

1644 lines
58 KiB
Python

from __future__ import annotations
import hashlib
import json
import re
from collections import deque
from dataclasses import dataclass, field
from typing import Any
from govoplan_core.core.definition_graphs import (
DefinitionEdge,
DefinitionNode,
validate_definition_graph,
)
from govoplan_dataflow.backend.node_library import (
DATAFLOW_GRAPH_LIBRARY,
NODE_TYPES,
node_definition,
)
from govoplan_dataflow.backend.expressions import (
ExpressionError,
infer_expression_type,
parse_expression,
)
from govoplan_dataflow.backend.operator_registry import OPERATOR_REGISTRY
from govoplan_dataflow.backend.subflows import (
SubflowParameterError,
substitute_parameters,
)
from govoplan_dataflow.backend.schemas import DataflowDiagnostic, GraphEdge, GraphNode, PipelineGraph
SUPPORTED_NODE_TYPES = frozenset(NODE_TYPES)
FILTER_OPERATORS = frozenset(
{"eq", "ne", "gt", "gte", "lt", "lte", "contains", "is_null", "not_null"}
)
AGGREGATE_FUNCTIONS = frozenset({"count", "sum", "avg", "min", "max"})
DERIVE_OPERATIONS = frozenset(
{
"copy",
"upper",
"lower",
"trim",
"concat",
"coalesce",
"add",
"subtract",
"multiply",
"divide",
}
)
JOIN_TYPES = frozenset({"inner", "left", "right", "full"})
DATA_TYPES = frozenset(
{"string", "integer", "number", "boolean", "date", "datetime"}
)
QUALITY_OPERATORS = frozenset(
{"not_null", "type", "min", "max", "allowed", "unique"}
)
def canonical_graph_payload(graph: PipelineGraph) -> dict[str, Any]:
return graph.model_dump(mode="json", exclude_none=True)
def definition_hash(graph: PipelineGraph, sql_text: str | None = None) -> str:
payload = {
"graph": canonical_graph_payload(graph),
"sql_text": (sql_text or "").strip() or None,
}
encoded = json.dumps(payload, sort_keys=True, separators=(",", ":"), ensure_ascii=True)
return hashlib.sha256(encoded.encode("utf-8")).hexdigest()
def preserve_compatible_graph_layout(
reference: PipelineGraph,
compiled: PipelineGraph,
) -> PipelineGraph:
"""Keep graph identity and layout when SQL preserves the same topology."""
reference_keys = _structural_node_keys(reference)
compiled_keys = _structural_node_keys(compiled)
if reference_keys is None or compiled_keys is None:
return compiled
if (
len(set(reference_keys.values())) != len(reference_keys)
or len(set(compiled_keys.values())) != len(compiled_keys)
or set(reference_keys.values()) != set(compiled_keys.values())
):
return compiled
compiled_by_key = {
key: next(node for node in compiled.nodes if node.id == node_id)
for node_id, key in compiled_keys.items()
}
reference_edges = {
_structural_edge_key(edge, reference_keys): edge
for edge in reference.edges
}
compiled_edges = {
_structural_edge_key(edge, compiled_keys): edge
for edge in compiled.edges
}
if (
len(reference_edges) != len(reference.edges)
or len(compiled_edges) != len(compiled.edges)
or set(reference_edges) != set(compiled_edges)
):
return compiled
nodes = []
for reference_node in reference.nodes:
node = compiled_by_key[reference_keys[reference_node.id]]
nodes.append(
node.model_copy(
update={
"id": reference_node.id,
"label": reference_node.label,
"position": reference_node.position.model_copy(deep=True),
},
deep=True,
)
)
edges = []
for reference_edge in reference.edges:
edge_key = _structural_edge_key(reference_edge, reference_keys)
edge = compiled_edges[edge_key]
edges.append(
edge.model_copy(
update={
"id": reference_edge.id,
"source": reference_edge.source,
"target": reference_edge.target,
},
deep=True,
)
)
return compiled.model_copy(
update={"nodes": nodes, "edges": edges},
deep=True,
)
def validate_graph(graph: PipelineGraph) -> list[DataflowDiagnostic]:
diagnostics = [
DataflowDiagnostic(
severity=item.severity,
code=item.code,
message=item.message,
node_id=item.node_id,
field=item.field,
)
for item in validate_definition_graph(
DATAFLOW_GRAPH_LIBRARY,
nodes=tuple(DefinitionNode(id=node.id, type=node.type) for node in graph.nodes),
edges=tuple(
DefinitionEdge(
id=edge.id,
source=edge.source,
target=edge.target,
source_port=edge.source_port,
target_port=edge.target_port,
)
for edge in graph.edges
),
)
]
nodes = {node.id: node for node in graph.nodes}
incoming: dict[str, list[GraphEdge]] = {node_id: [] for node_id in nodes}
outgoing: dict[str, list[str]] = {node_id: [] for node_id in nodes}
for edge in graph.edges:
if edge.source not in nodes or edge.target not in nodes or edge.source == edge.target:
continue
source_definition = node_definition(nodes[edge.source].type)
target_definition = node_definition(nodes[edge.target].type)
if (
source_definition is None
or target_definition is None
or edge.source_port not in {port.id for port in source_definition.output_ports}
or edge.target_port not in {port.id for port in target_definition.input_ports}
):
continue
outgoing[edge.source].append(edge.target)
incoming[edge.target].append(edge)
source_nodes = [node for node in graph.nodes if node.type.startswith("source.")]
output_nodes = [node for node in graph.nodes if node.type == "output"]
source_names = [
str(node.config.get("source_name", "")).strip()
for node in source_nodes
if _non_empty_text(node.config.get("source_name"))
]
duplicate_source_names = sorted(
{
name
for name in source_names
if sum(candidate.casefold() == name.casefold() for candidate in source_names) > 1
},
key=str.casefold,
)
if duplicate_source_names:
diagnostics.append(
_error(
"source.duplicate_name",
"Logical source names must be unique: "
f"{', '.join(duplicate_source_names)}.",
)
)
for node in graph.nodes:
if node.type not in SUPPORTED_NODE_TYPES:
continue
validator = OPERATOR_REGISTRY.config_validator(node.type)
if validator is None:
diagnostics.append(
_error(
"node.validator_missing",
f"Node type {node.type!r} has no registered validator.",
node_id=node.id,
)
)
continue
diagnostics.extend(validator(node))
ordered, cyclic = topological_order(graph)
if not cyclic and source_nodes and len(output_nodes) == 1:
reachable: set[str] = set()
for source in source_nodes:
reachable.update(_reachable_from(source.id, outgoing))
if len(reachable) != len(nodes):
diagnostics.append(
_error("graph.disconnected", "Every node must be connected to a pipeline source.")
)
reverse_adjacency = {
node_id: [edge.source for edge in edges]
for node_id, edges in incoming.items()
}
reaches_output = _reachable_from(output_nodes[0].id, reverse_adjacency)
if len(reaches_output) != len(nodes):
diagnostics.append(
_error("graph.dead_end", "Every node must lead to the pipeline output.")
)
if ordered and ordered[-1] != output_nodes[0].id:
diagnostics.append(
_error("graph.output_not_terminal", "The output node must be the terminal transform.")
)
diagnostics.extend(_validate_graph_schemas(graph, ordered=ordered))
return _dedupe_diagnostics(diagnostics)
def _structural_node_keys(
graph: PipelineGraph,
) -> dict[str, tuple[object, ...]] | None:
node_by_id = {node.id: node for node in graph.nodes}
if (
len(node_by_id) != len(graph.nodes)
or any(
edge.source not in node_by_id or edge.target not in node_by_id
for edge in graph.edges
)
):
return None
ordered, cyclic = topological_order(graph)
if cyclic:
return None
incoming: dict[str, list[GraphEdge]] = {
node.id: []
for node in graph.nodes
}
for edge in graph.edges:
if edge.target in incoming:
incoming[edge.target].append(edge)
keys: dict[str, tuple[object, ...]] = {}
for node_id in ordered:
node = node_by_id[node_id]
input_keys: list[tuple[object, ...]] = []
for edge in incoming[node_id]:
source_key = keys.get(edge.source)
if source_key is None:
return None
input_keys.append(
(
edge.target_port,
edge.source_port,
source_key,
)
)
if node.type != "combine.union":
input_keys.sort(key=repr)
source_name = (
str(node.config.get("source_name", "")).strip().casefold()
if node.type.startswith("source.")
else None
)
keys[node_id] = (
node.type,
source_name,
tuple(input_keys),
)
return keys
def _structural_edge_key(
edge: GraphEdge,
node_keys: dict[str, tuple[object, ...]],
) -> tuple[object, ...]:
return (
node_keys[edge.source],
node_keys[edge.target],
edge.source_port,
edge.target_port,
)
def _dedupe_diagnostics(
diagnostics: list[DataflowDiagnostic],
) -> list[DataflowDiagnostic]:
seen: set[tuple[str, str | None, str | None]] = set()
result: list[DataflowDiagnostic] = []
for item in diagnostics:
key = (item.code, item.node_id, item.field)
if key in seen:
continue
seen.add(key)
result.append(item)
return result
def topological_order(graph: PipelineGraph) -> tuple[list[str], bool]:
node_ids = [node.id for node in graph.nodes]
incoming_count = {node_id: 0 for node_id in node_ids}
outgoing: dict[str, list[str]] = {node_id: [] for node_id in node_ids}
for edge in graph.edges:
if edge.source in outgoing and edge.target in incoming_count and edge.source != edge.target:
outgoing[edge.source].append(edge.target)
incoming_count[edge.target] += 1
ready = deque(node_id for node_id in node_ids if incoming_count[node_id] == 0)
ordered: list[str] = []
while ready:
node_id = ready.popleft()
ordered.append(node_id)
for target in outgoing[node_id]:
incoming_count[target] -= 1
if incoming_count[target] == 0:
ready.append(target)
return ordered, len(ordered) != len(node_ids)
def graph_inputs_by_port(graph: PipelineGraph) -> dict[str, dict[str, list[str]]]:
result: dict[str, dict[str, list[str]]] = {}
for edge in graph.edges:
result.setdefault(edge.target, {}).setdefault(edge.target_port, []).append(edge.source)
return result
def _reachable_from(start: str, adjacency: dict[str, list[str]]) -> set[str]:
seen: set[str] = set()
pending = [start]
while pending:
node_id = pending.pop()
if node_id in seen:
continue
seen.add(node_id)
pending.extend(adjacency.get(node_id, ()))
return seen
@dataclass(frozen=True)
class _SchemaState:
columns: frozenset[str]
open: bool = False
types: dict[str, str] = field(default_factory=dict)
def knows(self, column: str) -> bool:
return self.open or column in self.columns
def type_of(self, column: str) -> str:
return self.types.get(column, "unknown")
def _validate_graph_schemas(
graph: PipelineGraph,
*,
ordered: list[str],
) -> list[DataflowDiagnostic]:
diagnostics: list[DataflowDiagnostic] = []
node_by_id = {node.id: node for node in graph.nodes}
inputs = graph_inputs_by_port(graph)
schemas: dict[str, _SchemaState] = {}
for node_id in ordered:
node = node_by_id[node_id]
node_inputs = inputs.get(node.id, {})
input_states = [
schemas[source_id]
for port_sources in node_inputs.values()
for source_id in port_sources
if source_id in schemas
]
input_state = input_states[0] if input_states else _SchemaState(frozenset(), open=True)
if node.type == "source.inline":
rows = node.config.get("rows")
schemas[node.id] = _inline_schema(rows)
continue
if node.type == "source.reference":
schemas[node.id] = _configured_source_schema(
node.config.get("source_columns")
)
continue
if node.type == "combine.union":
if len(input_states) > 1:
closed_shapes = {
state.columns
for state in input_states
if not state.open
}
if len(closed_shapes) > 1:
diagnostics.append(
_warning(
"union.schema_mismatch",
"Appended inputs use different columns; missing values will be null.",
node_id=node.id,
)
)
schemas[node.id] = _SchemaState(
frozenset().union(*(state.columns for state in input_states)),
open=any(state.open for state in input_states),
types=_merged_schema_types(input_states),
)
continue
if node.type == "combine.join":
left_state = _port_schema(node_inputs, schemas, "left")
right_state = _port_schema(node_inputs, schemas, "right")
_validate_columns(
diagnostics,
node=node,
state=left_state,
columns=node.config.get("left_keys"),
field="left_keys",
)
_validate_columns(
diagnostics,
node=node,
state=right_state,
columns=node.config.get("right_keys"),
field="right_keys",
)
prefix = str(node.config.get("right_prefix", "right_"))
prefixed_right = {f"{prefix}{column}" for column in right_state.columns}
collisions = left_state.columns & prefixed_right
if collisions:
diagnostics.append(
_error(
"join.output_collision",
f"Join output columns collide: {', '.join(sorted(collisions))}.",
node_id=node.id,
field="right_prefix",
)
)
schemas[node.id] = _SchemaState(
frozenset(left_state.columns | prefixed_right),
open=left_state.open or right_state.open,
types={
**left_state.types,
**{
f"{prefix}{column}": right_state.type_of(column)
for column in right_state.columns
},
},
)
continue
if node.type == "filter":
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=[node.config.get("column")],
field="column",
)
schemas[node.id] = input_state
continue
if node.type == "filter.expression":
parsed = _parsed_node_expression(
diagnostics,
node=node,
field_name="expression",
)
if parsed is not None:
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=list(parsed.columns),
field="expression",
)
schemas[node.id] = input_state
continue
if node.type == "distinct":
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=node.config.get("columns"),
field="columns",
)
schemas[node.id] = input_state
continue
if node.type == "select":
fields = node.config.get("fields")
selected_columns: list[str] = []
output_columns: list[str] = []
if isinstance(fields, list):
for field in fields:
if isinstance(field, str):
selected_columns.append(field)
output_columns.append(field)
elif isinstance(field, dict):
column = field.get("column")
alias = field.get("alias") or column
if isinstance(column, str):
selected_columns.append(column)
if isinstance(alias, str):
output_columns.append(alias)
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=selected_columns,
field="fields",
)
_validate_output_names(
diagnostics,
node=node,
columns=output_columns,
field="fields",
)
schemas[node.id] = _SchemaState(
frozenset(output_columns),
types={
output: input_state.type_of(source)
for source, output in zip(
selected_columns,
output_columns,
strict=False,
)
},
)
continue
if node.type == "derive":
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=node.config.get("source_columns"),
field="source_columns",
)
target = node.config.get("target_column")
if isinstance(target, str) and target:
if target in input_state.columns:
diagnostics.append(
_warning(
"derive.overwrites_column",
f"Derived column {target!r} replaces an existing value.",
node_id=node.id,
field="target_column",
)
)
schemas[node.id] = _SchemaState(
input_state.columns | frozenset((target,)),
open=input_state.open,
types={
**input_state.types,
target: _derive_result_type(
str(node.config.get("operation") or ""),
[
input_state.type_of(column)
for column in _text_items(
node.config.get("source_columns")
)
],
),
},
)
else:
schemas[node.id] = input_state
continue
if node.type == "expression":
parsed = _parsed_node_expression(
diagnostics,
node=node,
field_name="expression",
)
target = str(node.config.get("target_column") or "")
result_type = "unknown"
if parsed is not None:
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=list(parsed.columns),
field="expression",
)
result_type = infer_expression_type(
parsed,
{
name: input_state.type_of(name) # type: ignore[dict-item]
for name in input_state.columns
},
)
expected = str(node.config.get("result_type") or "unknown")
if (
expected != "unknown"
and result_type not in {"unknown", "null", expected}
):
diagnostics.append(
_warning(
"expression.type_mismatch",
f"Expression infers {result_type}, not {expected}.",
node_id=node.id,
field="result_type",
)
)
schemas[node.id] = _SchemaState(
input_state.columns | ({target} if target else set()),
open=input_state.open,
types={
**input_state.types,
**(
{target: expected if expected != "unknown" else result_type}
if target
else {}
),
},
)
continue
if node.type in {"convert", "replace"}:
source = str(node.config.get("source_column") or "")
target = str(node.config.get("target_column") or "")
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=[source],
field="source_column",
)
target_type = (
str(node.config.get("target_type") or "unknown")
if node.type == "convert"
else input_state.type_of(source)
)
schemas[node.id] = _SchemaState(
input_state.columns | ({target} if target else set()),
open=input_state.open,
types={
**input_state.types,
**({target: target_type} if target else {}),
},
)
continue
if node.type == "aggregate":
group_by = _text_items(node.config.get("group_by"))
aggregates = node.config.get("aggregates")
aggregate_columns = [
str(item.get("column"))
for item in aggregates
if isinstance(aggregates, list)
and isinstance(item, dict)
and item.get("column") not in (None, "", "*")
] if isinstance(aggregates, list) else []
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=[*group_by, *aggregate_columns],
field="aggregates",
)
aliases = [
str(item.get("alias"))
for item in aggregates
if isinstance(aggregates, list)
and isinstance(item, dict)
and item.get("alias")
] if isinstance(aggregates, list) else []
output_columns = [*group_by, *aliases]
_validate_output_names(
diagnostics,
node=node,
columns=output_columns,
field="aggregates",
)
aggregate_types = {
str(item.get("alias")): (
"integer"
if item.get("function") == "count"
else input_state.type_of(str(item.get("column") or ""))
)
for item in aggregates
if isinstance(aggregates, list)
and isinstance(item, dict)
and item.get("alias")
} if isinstance(aggregates, list) else {}
schemas[node.id] = _SchemaState(
frozenset(output_columns),
types={
**{
column: input_state.type_of(column)
for column in group_by
},
**aggregate_types,
},
)
continue
if node.type == "sort":
fields = node.config.get("fields")
columns = [
str(item.get("column"))
for item in fields
if isinstance(fields, list)
and isinstance(item, dict)
and item.get("column")
] if isinstance(fields, list) else []
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=columns,
field="fields",
)
schemas[node.id] = input_state
continue
if node.type == "quality.rules":
rules = node.config.get("rules")
_validate_columns(
diagnostics,
node=node,
state=input_state,
columns=[
str(rule.get("column") or "")
for rule in rules
if isinstance(rules, list) and isinstance(rule, dict)
] if isinstance(rules, list) else [],
field="rules",
)
if node.config.get("action", "annotate") == "annotate":
schemas[node.id] = _SchemaState(
input_state.columns
| frozenset(("_quality_valid", "_quality_errors")),
open=input_state.open,
types={
**input_state.types,
"_quality_valid": "boolean",
"_quality_errors": "array",
},
)
else:
schemas[node.id] = input_state
continue
if node.type == "reconcile.compare":
left_state = _port_schema(node_inputs, schemas, "left")
right_state = _port_schema(node_inputs, schemas, "right")
_validate_columns(
diagnostics,
node=node,
state=left_state,
columns=node.config.get("left_keys"),
field="left_keys",
)
_validate_columns(
diagnostics,
node=node,
state=right_state,
columns=node.config.get("right_keys"),
field="right_keys",
)
comparison_columns = node.config.get("compare_columns")
if isinstance(comparison_columns, list):
left_columns: list[str] = []
right_columns: list[str] = []
for item in comparison_columns:
if isinstance(item, str):
left_columns.append(item)
right_columns.append(item)
elif isinstance(item, dict):
left_columns.append(
str(item.get("left") or item.get("column") or "")
)
right_columns.append(
str(
item.get("right")
or item.get("left")
or item.get("column")
or ""
)
)
_validate_columns(
diagnostics,
node=node,
state=left_state,
columns=left_columns,
field="compare_columns",
)
_validate_columns(
diagnostics,
node=node,
state=right_state,
columns=right_columns,
field="compare_columns",
)
prefix = str(node.config.get("right_prefix") or "observed_")
schemas[node.id] = _SchemaState(
left_state.columns
| frozenset(f"{prefix}{column}" for column in right_state.columns)
| frozenset(
(
"_reconciliation_status",
"_reconciliation_differences",
)
),
open=left_state.open or right_state.open,
types={
**left_state.types,
**{
f"{prefix}{column}": right_state.type_of(column)
for column in right_state.columns
},
"_reconciliation_status": "string",
"_reconciliation_differences": "array",
},
)
continue
if node.type == "subflow":
output_schema = _configured_source_schema(
node.config.get("output_schema")
)
schemas[node.id] = (
output_schema
if output_schema.columns
else _SchemaState(frozenset(), open=True)
)
continue
schemas[node.id] = input_state
return diagnostics
def _configured_source_schema(value: object) -> _SchemaState:
if not isinstance(value, list):
return _SchemaState(frozenset(), open=True)
columns = {
item if isinstance(item, str) else str(item.get("name"))
for item in value
if (
isinstance(item, str) and item
or isinstance(item, dict) and item.get("name")
)
}
types = {
str(item.get("name")): str(
item.get("data_type") or item.get("type") or "unknown"
)
for item in value
if isinstance(item, dict) and item.get("name")
}
return _SchemaState(
frozenset(columns),
open=not columns,
types=types,
)
def _inline_schema(value: object) -> _SchemaState:
if not isinstance(value, list):
return _SchemaState(frozenset(), open=True)
columns = {
str(column)
for row in value
if isinstance(row, dict)
for column in row
}
types: dict[str, str] = {}
for column in columns:
observed = {
_schema_value_type(row.get(column))
for row in value
if isinstance(row, dict) and row.get(column) is not None
}
types[column] = observed.pop() if len(observed) == 1 else "unknown"
return _SchemaState(
frozenset(columns),
open=not columns,
types=types,
)
def _schema_value_type(value: object) -> str:
if isinstance(value, bool):
return "boolean"
if isinstance(value, int):
return "integer"
if isinstance(value, float):
return "number"
if isinstance(value, str):
return "string"
if isinstance(value, list):
return "array"
if isinstance(value, dict):
return "object"
return "unknown"
def _merged_schema_types(states: list[_SchemaState]) -> dict[str, str]:
columns = frozenset().union(*(state.columns for state in states))
types: dict[str, str] = {}
for column in columns:
observed = {
state.type_of(column)
for state in states
if column in state.columns and state.type_of(column) != "unknown"
}
types[column] = observed.pop() if len(observed) == 1 else "unknown"
return types
def _derive_result_type(operation: str, source_types: list[str]) -> str:
if operation in {"upper", "lower", "trim", "concat"}:
return "string"
if operation in {"add", "subtract", "multiply", "divide"}:
return "number" if "number" in source_types or operation == "divide" else "integer"
if operation in {"copy", "coalesce"}:
concrete = {
item for item in source_types if item not in {"unknown", "null"}
}
return concrete.pop() if len(concrete) == 1 else "unknown"
return "unknown"
def _parsed_node_expression(
diagnostics: list[DataflowDiagnostic],
*,
node: GraphNode,
field_name: str,
):
try:
return parse_expression(str(node.config.get(field_name) or ""))
except ExpressionError as exc:
diagnostics.append(
_error(
"expression.invalid",
str(exc),
node_id=node.id,
field=field_name,
)
)
return None
def _port_schema(
inputs: dict[str, list[str]],
schemas: dict[str, _SchemaState],
port: str,
) -> _SchemaState:
source_ids = inputs.get(port, [])
return schemas.get(source_ids[0], _SchemaState(frozenset(), open=True)) if source_ids else _SchemaState(frozenset(), open=True)
def _validate_columns(
diagnostics: list[DataflowDiagnostic],
*,
node: GraphNode,
state: _SchemaState,
columns: object,
field: str,
) -> None:
for column in _text_items(columns):
if not state.knows(column):
diagnostics.append(
_error(
"schema.unknown_column",
f"Column {column!r} is not available at this node.",
node_id=node.id,
field=field,
)
)
def _validate_output_names(
diagnostics: list[DataflowDiagnostic],
*,
node: GraphNode,
columns: list[str],
field: str,
) -> None:
duplicates = sorted({column for column in columns if columns.count(column) > 1})
if duplicates:
diagnostics.append(
_error(
"schema.duplicate_output",
f"Output column names must be unique: {', '.join(duplicates)}.",
node_id=node.id,
field=field,
)
)
def _text_items(value: object) -> list[str]:
if not isinstance(value, list):
return []
return [item for item in value if isinstance(item, str) and item]
def _validate_node_config(node: GraphNode) -> list[DataflowDiagnostic]:
config = node.config
diagnostics: list[DataflowDiagnostic] = []
if node.type in {"source.inline", "source.reference"}:
source_name = config.get("source_name")
if not isinstance(source_name, str) or not source_name.strip():
diagnostics.append(
_error(
"source.name_required",
"A source needs a logical SQL name.",
node_id=node.id,
field="source_name",
)
)
elif re.fullmatch(r"[A-Za-z_][A-Za-z0-9_]*", source_name.strip()) is None:
diagnostics.append(
_error(
"source.name_invalid",
"Logical source names must be SQL identifiers, such as monthly_cases.",
node_id=node.id,
field="source_name",
)
)
if node.type == "source.reference":
if not _non_empty_text(config.get("source_ref")):
diagnostics.append(
_error(
"source.reference_required",
"Choose a datasource.",
node_id=node.id,
field="source_ref",
)
)
expected_fingerprint = config.get("expected_fingerprint")
if expected_fingerprint is not None and not isinstance(expected_fingerprint, str):
diagnostics.append(
_error(
"source.fingerprint",
"The expected source fingerprint must be text.",
node_id=node.id,
field="expected_fingerprint",
)
)
return diagnostics
rows = config.get("rows")
if not isinstance(rows, list):
diagnostics.append(
_error("source.rows_required", "Inline source rows must be a list.", node_id=node.id, field="rows")
)
elif len(rows) > 250:
diagnostics.append(
_error("source.row_limit", "Inline sources are limited to 250 rows.", node_id=node.id, field="rows")
)
elif any(not isinstance(row, dict) for row in rows):
diagnostics.append(
_error("source.row_shape", "Every inline source row must be an object.", node_id=node.id, field="rows")
)
if len(json.dumps(config, default=str).encode("utf-8")) > 512_000:
diagnostics.append(
_error("source.byte_limit", "Inline source configuration is limited to 512 KiB.", node_id=node.id)
)
elif node.type == "filter":
if not _non_empty_text(config.get("column")):
diagnostics.append(_node_field_error(node, "filter.column", "Choose a column.", "column"))
operator = config.get("operator")
if operator not in FILTER_OPERATORS:
diagnostics.append(
_node_field_error(node, "filter.operator", "Choose a supported comparison operator.", "operator")
)
if operator not in {"is_null", "not_null"} and "value" not in config:
diagnostics.append(_node_field_error(node, "filter.value", "Enter a comparison value.", "value"))
elif node.type == "filter.expression":
if not _non_empty_text(config.get("expression")):
diagnostics.append(
_node_field_error(
node,
"expression.required",
"Enter a filter expression.",
"expression",
)
)
elif node.type == "distinct":
columns = config.get("columns", [])
if not isinstance(columns, list) or any(not _non_empty_text(item) for item in columns):
diagnostics.append(
_node_field_error(
node,
"distinct.columns",
"Distinct key columns must be named.",
"columns",
)
)
elif node.type == "combine.union":
if config.get("mode", "all") not in {"all", "distinct"}:
diagnostics.append(
_node_field_error(
node,
"union.mode",
"Choose whether duplicate rows are kept or removed.",
"mode",
)
)
elif node.type == "combine.join":
if config.get("join_type", "inner") not in JOIN_TYPES:
diagnostics.append(
_node_field_error(node, "join.type", "Choose a supported join type.", "join_type")
)
left_keys = config.get("left_keys")
right_keys = config.get("right_keys")
if (
not isinstance(left_keys, list)
or not left_keys
or any(not _non_empty_text(item) for item in left_keys)
):
diagnostics.append(
_node_field_error(node, "join.left_keys", "Add at least one left key.", "left_keys")
)
if (
not isinstance(right_keys, list)
or not right_keys
or any(not _non_empty_text(item) for item in right_keys)
):
diagnostics.append(
_node_field_error(node, "join.right_keys", "Add at least one right key.", "right_keys")
)
if isinstance(left_keys, list) and isinstance(right_keys, list) and len(left_keys) != len(right_keys):
diagnostics.append(
_node_field_error(
node,
"join.key_count",
"Left and right joins need the same number of keys.",
"right_keys",
)
)
right_prefix = config.get("right_prefix", "right_")
if (
not _non_empty_text(right_prefix)
or re.fullmatch(r"[A-Za-z_][A-Za-z0-9_]*_", str(right_prefix)) is None
):
diagnostics.append(
_node_field_error(
node,
"join.right_prefix",
"Enter an identifier prefix ending in an underscore, such as right_.",
"right_prefix",
)
)
elif node.type == "select":
fields = config.get("fields")
if not isinstance(fields, list) or not fields:
diagnostics.append(
_node_field_error(node, "select.fields", "Select at least one output column.", "fields")
)
else:
for field in fields:
column = field if isinstance(field, str) else field.get("column") if isinstance(field, dict) else None
if not _non_empty_text(column):
diagnostics.append(
_node_field_error(node, "select.field", "Selected fields need a source column.", "fields")
)
break
elif node.type == "aggregate":
group_by = config.get("group_by", [])
aggregates = config.get("aggregates")
if not isinstance(group_by, list) or any(not _non_empty_text(item) for item in group_by):
diagnostics.append(
_node_field_error(node, "aggregate.group_by", "Group-by columns must be named.", "group_by")
)
if not isinstance(aggregates, list) or not aggregates:
diagnostics.append(
_node_field_error(node, "aggregate.required", "Add at least one aggregate.", "aggregates")
)
else:
for aggregate in aggregates:
if not isinstance(aggregate, dict) or aggregate.get("function") not in AGGREGATE_FUNCTIONS:
diagnostics.append(
_node_field_error(
node,
"aggregate.function",
"Choose COUNT, SUM, AVG, MIN, or MAX.",
"aggregates",
)
)
break
if aggregate.get("function") != "count" and not _non_empty_text(aggregate.get("column")):
diagnostics.append(
_node_field_error(
node,
"aggregate.column",
"This aggregate needs a source column.",
"aggregates",
)
)
break
if not _non_empty_text(aggregate.get("alias")):
diagnostics.append(
_node_field_error(node, "aggregate.alias", "Every aggregate needs an alias.", "aggregates")
)
break
elif node.type == "derive":
if not _non_empty_text(config.get("target_column")):
diagnostics.append(
_node_field_error(
node,
"derive.target_column",
"Choose an output column.",
"target_column",
)
)
operation = config.get("operation")
if operation not in DERIVE_OPERATIONS:
diagnostics.append(
_node_field_error(
node,
"derive.operation",
"Choose a supported derive operation.",
"operation",
)
)
source_columns = config.get("source_columns")
if (
not isinstance(source_columns, list)
or not source_columns
or any(not _non_empty_text(item) for item in source_columns)
):
diagnostics.append(
_node_field_error(
node,
"derive.source_columns",
"Choose at least one source column.",
"source_columns",
)
)
if operation in {"copy", "upper", "lower", "trim"} and isinstance(source_columns, list) and len(source_columns) != 1:
diagnostics.append(
_node_field_error(
node,
"derive.source_count",
"This operation requires exactly one source column.",
"source_columns",
)
)
if operation in {"add", "subtract", "multiply", "divide"} and isinstance(source_columns, list) and len(source_columns) != 2:
diagnostics.append(
_node_field_error(
node,
"derive.numeric_source_count",
"Numeric operations require exactly two source columns.",
"source_columns",
)
)
elif node.type == "expression":
if not _non_empty_text(config.get("target_column")):
diagnostics.append(
_node_field_error(
node,
"expression.target_column",
"Choose an output column.",
"target_column",
)
)
if not _non_empty_text(config.get("expression")):
diagnostics.append(
_node_field_error(
node,
"expression.required",
"Enter an expression.",
"expression",
)
)
if str(config.get("result_type") or "unknown") not in {
"unknown",
*DATA_TYPES,
}:
diagnostics.append(
_node_field_error(
node,
"expression.result_type",
"Choose a supported expression result type.",
"result_type",
)
)
elif node.type == "convert":
for field_name, message in (
("source_column", "Choose a source column."),
("target_column", "Choose an output column."),
):
if not _non_empty_text(config.get(field_name)):
diagnostics.append(
_node_field_error(
node,
f"convert.{field_name}",
message,
field_name,
)
)
if config.get("target_type") not in DATA_TYPES:
diagnostics.append(
_node_field_error(
node,
"convert.target_type",
"Choose a supported conversion type.",
"target_type",
)
)
if config.get("on_error", "fail") not in {"fail", "null", "keep"}:
diagnostics.append(
_node_field_error(
node,
"convert.on_error",
"Choose how conversion errors are handled.",
"on_error",
)
)
elif node.type == "replace":
for field_name, message in (
("source_column", "Choose a source column."),
("target_column", "Choose an output column."),
):
if not _non_empty_text(config.get(field_name)):
diagnostics.append(
_node_field_error(
node,
f"replace.{field_name}",
message,
field_name,
)
)
if config.get("mode", "exact") not in {"exact", "text"}:
diagnostics.append(
_node_field_error(
node,
"replace.mode",
"Choose exact-value or text replacement.",
"mode",
)
)
elif node.type == "sort":
fields = config.get("fields")
if not isinstance(fields, list) or not fields:
diagnostics.append(_node_field_error(node, "sort.fields", "Add at least one sort field.", "fields"))
elif any(
not isinstance(item, dict)
or not _non_empty_text(item.get("column"))
or item.get("direction", "asc") not in {"asc", "desc"}
for item in fields
):
diagnostics.append(
_node_field_error(node, "sort.field", "Sort fields need a column and direction.", "fields")
)
elif node.type == "limit":
count = config.get("count")
if not isinstance(count, int) or isinstance(count, bool) or not 1 <= count <= 100_000:
diagnostics.append(
_node_field_error(node, "limit.count", "Limit must be between 1 and 100,000.", "count")
)
elif node.type == "quality.rules":
rules = config.get("rules")
if not isinstance(rules, list) or not rules or len(rules) > 100:
diagnostics.append(
_node_field_error(
node,
"quality.rules",
"Add between one and 100 quality rules.",
"rules",
)
)
else:
rule_ids: list[str] = []
for rule in rules:
if not isinstance(rule, dict):
diagnostics.append(
_node_field_error(
node,
"quality.rule_shape",
"Every quality rule must be an object.",
"rules",
)
)
break
rule_id = str(rule.get("id") or "").strip()
rule_ids.append(rule_id)
operator = rule.get("operator")
if not rule_id or not _non_empty_text(rule.get("column")):
diagnostics.append(
_node_field_error(
node,
"quality.rule_identity",
"Quality rules need an ID and column.",
"rules",
)
)
break
if operator not in QUALITY_OPERATORS:
diagnostics.append(
_node_field_error(
node,
"quality.operator",
"Choose a supported quality operator.",
"rules",
)
)
break
if operator in {"type", "min", "max"} and "value" not in rule:
diagnostics.append(
_node_field_error(
node,
"quality.value",
"This quality rule requires a value.",
"rules",
)
)
break
if operator == "allowed" and not isinstance(rule.get("values"), list):
diagnostics.append(
_node_field_error(
node,
"quality.values",
"Allowed-value rules require a list of values.",
"rules",
)
)
break
duplicates = sorted(
{rule_id for rule_id in rule_ids if rule_ids.count(rule_id) > 1}
)
if duplicates:
diagnostics.append(
_node_field_error(
node,
"quality.duplicate_id",
f"Quality rule IDs must be unique: {', '.join(duplicates)}.",
"rules",
)
)
if config.get("action", "annotate") not in {"annotate", "drop", "fail"}:
diagnostics.append(
_node_field_error(
node,
"quality.action",
"Choose how invalid rows are handled.",
"action",
)
)
elif node.type == "reconcile.compare":
left_keys = config.get("left_keys")
right_keys = config.get("right_keys")
if (
not isinstance(left_keys, list)
or not left_keys
or any(not _non_empty_text(item) for item in left_keys)
):
diagnostics.append(
_node_field_error(
node,
"reconcile.left_keys",
"Add at least one expected-table key.",
"left_keys",
)
)
if (
not isinstance(right_keys, list)
or not right_keys
or any(not _non_empty_text(item) for item in right_keys)
):
diagnostics.append(
_node_field_error(
node,
"reconcile.right_keys",
"Add at least one observed-table key.",
"right_keys",
)
)
if (
isinstance(left_keys, list)
and isinstance(right_keys, list)
and len(left_keys) != len(right_keys)
):
diagnostics.append(
_node_field_error(
node,
"reconcile.key_count",
"Expected and observed keys must have the same length.",
"right_keys",
)
)
prefix = config.get("right_prefix", "observed_")
if (
not _non_empty_text(prefix)
or re.fullmatch(r"[A-Za-z_][A-Za-z0-9_]*_", str(prefix)) is None
):
diagnostics.append(
_node_field_error(
node,
"reconcile.right_prefix",
"Enter an identifier prefix ending in an underscore.",
"right_prefix",
)
)
elif node.type == "subflow":
if not _non_empty_text(config.get("template_ref")):
diagnostics.append(
_node_field_error(
node,
"subflow.template_ref",
"Choose a reusable template.",
"template_ref",
)
)
if not _non_empty_text(config.get("template_version")):
diagnostics.append(
_node_field_error(
node,
"subflow.template_version",
"Pin a template version.",
"template_version",
)
)
parameters = config.get("parameters")
if not isinstance(parameters, dict) or len(parameters) > 100:
diagnostics.append(
_node_field_error(
node,
"subflow.parameters",
"Subflow parameters must be an object with at most 100 entries.",
"parameters",
)
)
try:
nested_payload = substitute_parameters(
config.get("graph"),
parameters if isinstance(parameters, dict) else {},
)
nested = PipelineGraph.model_validate(nested_payload)
except (SubflowParameterError, ValueError) as exc:
diagnostics.append(
_node_field_error(
node,
"subflow.graph",
f"Pinned subflow graph is invalid: {exc}",
"graph",
)
)
else:
bound_inputs = [
item
for item in nested.nodes
if item.type == "source.inline"
and item.config.get("input_binding") is True
]
if len(bound_inputs) != 1:
diagnostics.append(
_node_field_error(
node,
"subflow.input_binding",
"Pinned subflows need exactly one inline input binding.",
"graph",
)
)
for nested_diagnostic in validate_graph(nested):
if nested_diagnostic.severity == "error":
diagnostics.append(
_node_field_error(
node,
"subflow.graph_validation",
f"Pinned subflow: {nested_diagnostic.message}",
"graph",
)
)
break
return diagnostics
def _non_empty_text(value: object) -> bool:
return isinstance(value, str) and bool(value.strip())
def _node_field_error(node: GraphNode, code: str, message: str, field: str) -> DataflowDiagnostic:
return _error(code, message, node_id=node.id, field=field)
def _error(
code: str,
message: str,
*,
node_id: str | None = None,
field: str | None = None,
) -> DataflowDiagnostic:
return DataflowDiagnostic(
severity="error",
code=code,
message=message,
node_id=node_id,
field=field,
)
def _warning(
code: str,
message: str,
*,
node_id: str | None = None,
field: str | None = None,
) -> DataflowDiagnostic:
return DataflowDiagnostic(
severity="warning",
code=code,
message=message,
node_id=node_id,
field=field,
)
def _register_config_validators() -> None:
for node_type in SUPPORTED_NODE_TYPES:
if OPERATOR_REGISTRY.config_validator(node_type) is None:
OPERATOR_REGISTRY.register_config_validator(
node_type,
_validate_node_config,
)
_register_config_validators()
__all__ = [
"AGGREGATE_FUNCTIONS",
"DERIVE_OPERATIONS",
"FILTER_OPERATORS",
"JOIN_TYPES",
"SUPPORTED_NODE_TYPES",
"canonical_graph_payload",
"definition_hash",
"graph_inputs_by_port",
"topological_order",
"validate_graph",
]