What causes "Non-supported bytesWidth: " + bytesWidth" in apache/iceberg?

Rewrite the Parquet files with a standard writer (Spark/Iceberg) so integer widths are 1, 2, or 4 bytes.

apache/iceberg · error · RuntimeException

Non-supported bytesWidth: " + bytesWidth

Error message

Non-supported bytesWidth: " + bytesWidth

What it means

readIntLittleEndianPaddedOnBitWidth only supports values packed into 1, 2, or 4 bytes; any other bytesWidth indicates a decoding mode this vectorized RLE reader cannot handle, so it throws a RuntimeException.

Source

Thrown at arrow/src/main/java/org/apache/iceberg/arrow/vectorized/parquet/BaseVectorizedParquetValuesReader.java:176

      case 2:
        {
          int ch2 = inputStream.read();
          int ch1 = inputStream.read();
          return (ch1 << 8) + ch2;
        }
      case 3:
        {
          int ch3 = inputStream.read();
          int ch2 = inputStream.read();
          int ch1 = inputStream.read();
          return (ch1 << 16) + (ch2 << 8) + ch3;
        }
      case 4:
        {
          return readIntLittleEndian();
        }
    }
    throw new RuntimeException("Non-supported bytesWidth: " + bytesWidth);
  }

  /** Reads the next group. */
  void readNextGroup() {
    try {
      int header = readUnsignedVarInt();
      this.mode = (header & 1) == 0 ? Mode.RLE : Mode.PACKED;
      switch (mode) {
        case RLE:
          this.currentCount = header >>> 1;
          this.currentValue = readIntLittleEndianPaddedOnBitWidth();
          return;
        case PACKED:
          int numGroups = header >>> 1;
          this.currentCount = numGroups * 8;
          if (this.packedValuesBuffer.length < this.currentCount) {
            this.packedValuesBuffer = new int[this.currentCount];
          }

View on GitHub (pinned to 86d9c8fc54)

Solutions

Rewrite the Parquet files with a standard writer (Spark/Iceberg) so integer widths are 1, 2, or 4 bytes.
Use the non-vectorized Parquet reader (disable vectorized reads) which supports the full decoding spec.
Verify the file is valid with parquet-tools; replace the corrupt source file.

Example fix

// before
.option("vectorization-enabled", "true")
// after
spark.read.format("iceberg").option("vectorization-enabled", "false").load("t")

Defensive patterns

Strategy: fallback

Try / catch

// catch (RuntimeException e) {
//   if (e.getMessage() != null && e.getMessage().contains("Non-supported bytesWidth")) {
//     retryWithVectorizationDisabled();
//   } else throw e;
// }

Prevention

Write Parquet files with standard producers (Spark/Iceberg).
Validate source files with parquet-tools before ingesting.
Fall back to non-vectorized reads for externally-produced files.

When it happens

Trigger: Decoding an RLE/bit-packed hybrid INTEGER column whose width resolves to a byte size other than 1, 2, or 4 inside readNextGroup.

Common situations: Reading a Parquet file written by an unusual producer or with an unexpected physical type width; corrupted or hand-crafted Parquet metadata.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

AI-assisted analysis of apache/iceberg@86d9c8fc54 (2026-09-12). Data as JSON: /api/errors/8cd5c69e10ebb731. Report an issue: GitHub.