Skip to main content

databricks.labs.dqx.actions.conditions

Safe AST-based condition evaluator for DQX action gating.

Conditions are arithmetic/boolean expressions evaluated against a metrics dict (e.g. "error_row_count > 0 or warning_row_count > 0"). They are parsed with ast.parse in "eval" mode and walked by a restricted visitor that allows only a small, safe subset of AST nodes. Any node outside that allowlist — including ast.Call, ast.Attribute, ast.Subscript, lambdas, and all comprehensions — raises InvalidConditionError.

Security: eval() / exec() / compile(mode="exec") are never used. The evaluator is purely structural, operating on the parsed AST.

ConditionEvaluator Objects​

class ConditionEvaluator()

Safe evaluator for DQX action-gating condition expressions.

Conditions are simple arithmetic/boolean expressions of the form "error_row_count > 0 or warning_row_count > 0". They are parsed via ast.parse (mode="eval") and evaluated by a restricted AST walker that allows only:

  • Literals: ast.Constant with int, float, bool, or str
  • Names: ast.Name — resolved from the metrics dict at evaluate time; numeric strings are coerced to float automatically
  • Boolean ops: and, or
  • Unary ops: not, unary minus, unary plus
  • Binary ops: add, subtract, multiply, divide, floor-divide, modulo, power
  • Comparisons: lt, le, gt, ge, eq, ne

Any other node type (calls, attribute access, subscripts, lambdas, comprehensions, …) raises InvalidConditionError.

A full-tree structural pre-pass is performed unconditionally before any evaluation, so short-circuit evaluation cannot bypass the allowlist.

Usage:

ConditionEvaluator.validate("error_row_count > 0")
result = ConditionEvaluator.evaluate("error_row_count > 0", {"error_row_count": 5})

validate​

@staticmethod
def validate(condition: str) -> None

Validate condition syntax and structure without requiring metrics.

Parses and walks every node in the AST, rejecting disallowed node types and syntax errors. Name resolution is not performed — unknown metric names are only caught at evaluate time. Call this method at DQAction construction time to surface malformed conditions early.

Arguments:

  • condition - The condition expression string to validate.

Raises:

  • InvalidConditionError - If condition has a syntax error or contains an AST node type that is not allowed.

evaluate​

@staticmethod
def evaluate(condition: str, metrics: dict[str, object]) -> bool

Evaluate condition against metrics and return a bool result.

Parses condition, performs a full-tree structural validation pass, resolves ast.Name nodes from metrics (coercing numeric strings to float), and returns the final truth value.

Arguments:

  • condition - The condition expression string to evaluate.
  • metrics - Mapping of metric name to value. Values that are numeric strings are coerced to float for arithmetic/comparison.

Returns:

Boolean result of the evaluated condition.

Raises:

  • InvalidConditionError - If condition has a syntax error, contains a disallowed AST node, or references a name not present in metrics.