apache/druid · error · IllegalStateException

Column[ ] is not a multi-value string column, do not ask…

Error message

Column[%s] is not a multi-value string column, do not ask for a multi-value selector

What it means

A multi-value dimension vector selector can only be built on a column that actually exists and is a dictionary-encoded multi-value STRING column. When the requested column is missing, not string-typed, or single-valued, the factory throws this ISE to catch planning bugs where the vector engine asked for the wrong selector kind.

Solutions

Confirm the column is ingested as multi-value string (use_json_string_array / MV ingestion) and exists in the segment
Use a single-value selector/query path for single-valued columns, or use ARRAY functions for typed arrays
Check the segment schema (segment metadata query) for actual column type; re-ingest if types diverge
Disable vectorization ('vectorize':'false') as a workaround while fixing the spec

Example fix

// before
factory.makeMultiValueDimensionSelector(DefaultDimensionSpec.of("singleValuedCol"));
// after
factory.makeSingleValueDimensionSelector(DefaultDimensionSpec.of("singleValuedCol"));

Defensive patterns

Strategy: validation

Validate before calling

// check segment metadata first
ColumnIngestionPermits/dataset: run segment metadata query and verify capabilities:
boolean ok = holder != null && holder.getCapabilities().isDictionaryEncoded().isMaybeTrue()
          && holder.getCapabilities().is(ValueType.STRING)
          && holder.getCapabilities().hasMultipleValues().isMaybeTrue();

Type guard

boolean isMvString = caps.is(ValueType.STRING) && caps.hasMultipleValues().isMaybeTrue();

Try / catch

try { sel = factory.makeMultiValueDimensionSelector(spec); } catch (IllegalStateException e) { sel = factory.makeSingleValueDimensionSelector(spec); }

Prevention

Enforce consistent column types with an explicit dimensionsSpec at ingest
Run a segment metadata query to confirm column types before MV-specific queries
Use ARRAY-typed columns and ARRAY functions for typed arrays instead of MV strings

When it happens

Trigger: Vectorized query calls makeMultiValueDimensionSelector on a single-value string/long column, a non-existent column, or a non-dictionary-encoded column.

Common situations: Auto-detected column type changed between segments (some segments string MV, others long) causing one segment to fail; planning code assuming multi-value based on ingest-time schema; typo'd column name.

Understand the failure class

Background: "is not a compatible type" / "cannot merge" errors: when a value's type doesn't match what the library requires — this error's family across 65 libraries.

Related errors

AI-assisted analysis of apache/druid@9b90983fd2 (2026-09-07). Data as JSON: /api/errors/eec34c8c9ca39c3b. Report an issue: GitHub.

Appendix: source

Thrown at processing/src/main/java/org/apache/druid/segment/vector/QueryableIndexVectorColumnSelectorFactory.java:94

  }

  @Override
  public MultiValueDimensionVectorSelector makeMultiValueDimensionSelector(final DimensionSpec dimensionSpec)
  {
    if (!dimensionSpec.canVectorize()) {
      throw new ISE("DimensionSpec[%s] cannot be vectorized", dimensionSpec);
    }
    Function<DimensionSpec, MultiValueDimensionVectorSelector> mappingFunction = spec -> {
      if (virtualColumns.exists(spec.getDimension())) {
        return virtualColumns.makeMultiValueDimensionVectorSelector(dimensionSpec, this, columnSelector, offset);
      }

      final ColumnHolder holder = columnSelector.getColumnHolder(spec.getDimension());
      if (holder == null
          || holder.getCapabilities().isDictionaryEncoded().isFalse()
          || !holder.getCapabilities().is(ValueType.STRING)
          || holder.getCapabilities().hasMultipleValues().isFalse()) {
        throw new ISE(
            "Column[%s] is not a multi-value string column, do not ask for a multi-value selector",
            spec.getDimension()
        );
      }

      @SuppressWarnings("unchecked")
      final DictionaryEncodedColumn<String> dictionaryEncodedColumn =
          (DictionaryEncodedColumn<String>) holder.getColumn();

      // dictionaryEncodedColumn is not null because of holder null check above
      assert dictionaryEncodedColumn != null;
      final MultiValueDimensionVectorSelector selector = dictionaryEncodedColumn.makeMultiValueDimensionVectorSelector(
          offset
      );

      return spec.decorate(selector);
    };

View on GitHub (pinned to 9b90983fd2)