Skip to content

[Bug] hll_* errors surface as Comet errors instead of HLL_INVALID_INPUT_SKETCH_BUFFER / HLL_UNION_DIFFERENT_LG_K #7549

Open

Description

@tlfua

Describe the bug

When the native HLL expressions fail, Comet raises a plain Comet execution error instead of Spark's SparkRuntimeException with the HLL error condition. Both engines fail on the same inputs, but the error class, SQLSTATE and message differ.

Expression Spark error condition (SQLSTATE)
hll_sketch_estimate, hll_union, hll_union_agg on an invalid buffer HLL_INVALID_INPUT_SKETCH_BUFFER (22546)
hll_union, hll_union_agg with different lgConfigK and allowDifferentLgConfigK = false HLL_UNION_DIFFERENT_LG_K (22000)

native/common/src/error.rs SparkError has no HLL variant, so these errors cannot reach SparkErrorConverter / ShimSparkErrorConverter.

Steps to reproduce

SELECT hll_sketch_estimate(X'0102030405');

Spark 4.x raises [HLL_INVALID_INPUT_SKETCH_BUFFER] Invalid call to hll_sketch_estimate; only valid HLL sketch buffers are supported as inputs ... with SQLSTATE 22546. Comet raises a Comet execution error with a different message.

Expected behavior

Comet raises the same error condition and SQLSTATE as Spark. A likely fix, following #5072 :

  • add SparkError variants for the two conditions in native/common/src/error.rs;
  • map them in the spark-4.x ShimSparkErrorConverter to QueryExecutionErrors.hllInvalidInputSketchBuffer(function) and QueryExecutionErrors.hllUnionDifferentLgK(left, right, function);
  • cover them with expect_error_class SQL tests or checkSparkError.

Additional context

Surfaced by the HLL audit for #7396 . The comparison experiment confirmed both engines fail on the same inputs ( #7396 (comment) ). The serdes currently list this as an incompatible reason that is never returned; the audit PR moves it to a compatibility note that links here.

Willingness to contribute

I can contribute a fix for this bug independently

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    area:expressionsExpression evaluationbugSomething isn't workingpriority:lowMinor issues, test failures, tooling, cosmetic

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions