prestodb/presto · critical · PrestoException

DRUID_SEGMENT_LOAD_ERROR

DRUID_SEGMENT_LOAD_ERROR

Error message

failed to load druid segment

What it means

DruidSegmentReader's constructor opens a Druid segment file, loads its columns, and creates ColumnValueSelectors wrapped in column readers. Any IOException during segment loading is rethrown as DRUID_SEGMENT_LOAD_ERROR with the generic message 'failed to load druid segment', so the connector cannot read the segment's data.

Source

Thrown at presto-druid/src/main/java/com/facebook/presto/druid/segment/DruidSegmentReader.java:66

    public DruidSegmentReader(SegmentIndexSource segmentIndexSource, List<ColumnHandle> columns)
    {
        try {
            queryableIndex = segmentIndexSource.loadIndex(columns);
            totalRowCount = queryableIndex.getNumRows();
            ImmutableMap.Builder<String, ColumnReader> selectorsBuilder = ImmutableMap.builder();
            for (ColumnHandle column : columns) {
                DruidColumnHandle druidColumn = (DruidColumnHandle) column;
                String columnName = druidColumn.getColumnName();
                Type type = druidColumn.getColumnType();
                BaseColumn baseColumn = queryableIndex.getColumnHolder(columnName).getColumn();
                ColumnValueSelector<?> valueSelector = baseColumn.makeColumnValueSelector(new SimpleReadableOffset());
                selectorsBuilder.put(columnName, createColumnReader(type, valueSelector));
            }
            columnValueSelectors = selectorsBuilder.build();
        }
        catch (IOException e) {
            throw new PrestoException(DRUID_SEGMENT_LOAD_ERROR, "failed to load druid segment");
        }
    }

    @Override
    public int nextBatch()
    {
        // TODO: dynamic batch sizing
        currentBatchSize = toIntExact(min(BATCH_SIZE, totalRowCount - currentPosition));
        currentPosition += currentBatchSize;
        return currentBatchSize;
    }

    @Override
    public Block readBlock(Type type, String columnName)
    {
        return columnValueSelectors.get(columnName).readBlock(type, currentBatchSize);
    }
}

View on GitHub (pinned to 55bb57d202)

Solutions

  1. Inspect the underlying IOException cause (enable debug logging) and verify the segment file exists and is intact at the deep-storage path returned by DruidSegmentInfo.getDeepStoragePath().
  2. Verify the segment in Druid: query the Druid metadata for the segment and re-fetch/re-download it from deep storage; kill and re-ingest the segment if corrupted.
  3. Check filesystem/network access from the Presto coordinator/worker to deep storage (S3/HDFS/GS credentials, connectivity).
  4. Clear the segment cache if present and retry, so the segment is freshly pulled from deep storage.

Example fix

// before (read without checking segment availability)
DruidSegmentReader reader = new DruidSegmentReader(segmentFile, columns, types);

// after (validate the segment is loadable first)
if (!segmentFile.exists() || segmentFile.length() == 0) {
    throw new PrestoException(DRUID_SEGMENT_LOAD_ERROR, "Segment file missing or empty: " + segmentFile);
}
DruidSegmentReader reader = new DruidSegmentReader(segmentFile, columns, types);
Defensive patterns

Strategy: try-catch

Validate before calling

Path segmentPath = Paths.get(segmentFileLocation);
if (!Files.isReadable(segmentPath) || Files.size(segmentPath) == 0) {
    throw new PrestoException(DRUID_SEGMENT_LOAD_ERROR, "Segment file missing, unreadable, or empty: " + segmentFileLocation);
}
// optionally: validate magic/header bytes of the segment file before loading

Try / catch

try {
    DruidSegmentReader reader = new DruidSegmentReader(file, columns, types);
} catch (PrestoException e) {
    if (DRUID_SEGMENT_LOAD_ERROR.getCode().equals(e.getErrorCode())) {
        log.error(e.getCause(), "Segment load failed; invalidating cache for %s", file);
        // evict cached copy, re-fetch from deep storage, retry once
    } else {
        throw e;
    }
}

Prevention

When it happens

Trigger: Constructing DruidSegmentReader over a segment file that cannot be read: corrupted/truncated segment file, missing file at the resolved deep-storage path, network/filesystem failure while streaming from S3/HDFS/GS, or an IO error while building column value selectors.

Common situations: Incomplete or corrupted segments in deep storage (e.g. failed Druid ingestion left partial files); deep-storage path misconfiguration so the local cache points at a nonexistent file; HDFS/S3 outage or credentials issue while fetching; segment deleted between planning and read.

Related errors

AI-assisted analysis of prestodb/presto@55bb57d202 (2026-09-04). Data as JSON: /api/errors/98836ba8d9754109. Report an issue: GitHub.