What causes "Failed to read ORC file: %s" in prestodb/presto?

Read the `cause` of the HIVE_CURSOR_ERROR to get the real underlying problem

prestodb/presto · error · PrestoException

HIVE_CURSOR_ERROR

HIVE_CURSOR_ERROR

Error message

Failed to read ORC file: %s

What it means

OrcBatchPageSource.getNextPage converts OrcCorruptionException to HIVE_BAD_DATA, but any other IOException or RuntimeException during record-set advancement closes the source and is wrapped as PrestoException(HIVE_CURSOR_ERROR) with the ORC data source id. This is the connector's generic 'the cursor hit a failure while reading this file' signal.

Source

Thrown at presto-hive/src/main/java/com/facebook/presto/hive/orc/OrcBatchPageSource.java:214

                    blocks[fieldId] = constantBlocks[fieldId].getRegion(0, batchSize);
                }
                else {
                    blocks[fieldId] = new LazyBlock(batchSize, new OrcBlockLoader(hiveColumnIndexes[fieldId]));
                }
            }
            return new Page(batchSize, blocks);
        }
        catch (PrestoException e) {
            closeWithSuppression(e);
            throw e;
        }
        catch (OrcCorruptionException e) {
            closeWithSuppression(e);
            throw new PrestoException(HIVE_BAD_DATA, e);
        }
        catch (IOException | RuntimeException e) {
            closeWithSuppression(e);
            throw new PrestoException(HIVE_CURSOR_ERROR, format("Failed to read ORC file: %s", orcDataSource.getId()), e);
        }
    }

    @Override
    public void close()
    {
        // some hive input formats are broken and bad things can happen if you close them multiple times
        if (closed) {
            return;
        }
        closed = true;

        try {
            stats.addMaxCombinedBytesPerRow(recordReader.getMaxCombinedBytesPerRow());
            recordReader.close();
        }
        catch (IOException e) {
            throw new UncheckedIOException(e);

View on GitHub (pinned to 55bb57d202)

Solutions

Read the `cause` of the HIVE_CURSOR_ERROR to get the real underlying problem
Validate file integrity: hdfs fsck and try reading the file with a Hive/ORC tooling (orc-tools dump)
Re-run the query if the cause was transient HDFS I/O (coexists with HIVE_FILESYSTEM_ERROR causes)
Replace/regenerate corrupt or truncated files; rewrite with orc-tools if the writer produced bad metadata

Example fix

// before
SELECT * FROM t; -- HIVE_CURSOR_ERROR: Failed to read ORC file: hdfs://.../part-0 (cause: EOFException)

// after
-- java -jar orc-tools-*.jar dump /warehouse/t/part-0  # verify readability
-- regenerate part-0 from the upstream job
SELECT * FROM t;

Defensive patterns

Strategy: try-catch

Validate before calling

# verify file readability outside Presto first
hadoop jar orc-tools-*.jar meta /warehouse/t/part-0
hdfs fsck /warehouse/t/part-0 -blocks

Try / catch

try {
    return query(sql);
} catch (PrestoException e) {
    if (e.getErrorCode().getName().equals("HIVE_CURSOR_ERROR")) {
        // inspect e.getCause() for the real IOException/RuntimeException
        logRootCause(e.getCause());
        return query(sql); // retry only if cause was transient IO
    }
    throw e;
}

Prevention

Always check e.getCause() — HIVE_CURSOR_ERROR is a wrapper, not a diagnosis
Validate newly written ORC files with orc-tools before publishing partitions
Monitor for truncated files (writes interrupted) in table locations
Ensure worker memory is sufficient for the column widths being read

When it happens

Trigger: getNextPage -> recordSet.advanceNextPosition() throws IOException or RuntimeException that is not OrcCorruptionException (e.g., EOF from truncated file, HdfsOrcDataSource PrestoExceptions bubbling, NPEs, OOM).

Common situations: Truncated files whose postscript/stripes cut off; transient HDFS read failures during long scans; faulty writer output causing unexpected runtime errors; worker memory exhaustion reading wide rows.

Related errors

AI-assisted analysis of prestodb/presto@55bb57d202 (2026-09-04). Data as JSON: /api/errors/44151b764564ec6c. Report an issue: GitHub.