prestodb/presto · warning · UncheckedIOException

java.io.IOException (wrapped in UncheckedIOException)

Error message

java.io.IOException (wrapped in UncheckedIOException)

What it means

MapBatchStreamReader.close wraps IOException raised while closing its direct/flat delegate readers in UncheckedIOException. It is a cleanup-time failure: closing the child map readers (and their nested key/value streams) hit an I/O problem, typically because the underlying source was already closed or a storage error occurred.

Source

Thrown at presto-orc/src/main/java/com/facebook/presto/orc/reader/MapBatchStreamReader.java:113

    }

    @Override
    public String toString()
    {
        return toStringHelper(this)
                .addValue(streamDescriptor)
                .toString();
    }

    @Override
    public void close()
    {
        try (Closer closer = Closer.create()) {
            closer.register(directReader::close);
            closer.register(flatReader::close);
        }
        catch (IOException e) {
            throw new UncheckedIOException(e);
        }
    }

    @Override
    public long getRetainedSizeInBytes()
    {
        return INSTANCE_SIZE + directReader.getRetainedSizeInBytes() + flatReader.getRetainedSizeInBytes();
    }
}

View on GitHub (pinned to 55bb57d202)

Solutions

Ensure the OrcDataSource outlives all readers; close readers first, source last.
Make close() idempotent in any custom stream implementations.
Inspect the wrapped cause — treat this as secondary to the original query failure.
Upgrade if the unwrap reveals a known double-close issue in the ORC driver.

Example fix

// before
dataSource.close();
mapReader.close();
// after
mapReader.close();
dataSource.close();

Defensive patterns

Strategy: try-catch

Try / catch

try {
    mapReader.close();
}
catch (UncheckedIOException e) {
    LOG.warn(e.getCause(), "Failed to close map batch reader delegates");
}

Prevention

Close readers before closing the OrcDataSource.
Use Closer.create() so all delegates still get closed even if one throws.
Keep the datasource open until all page sources are drained or cancelled.
Make close() idempotent in custom streams.

When it happens

Trigger: Calling close() after the OrcDataSource has already been closed elsewhere; storage-layer failure during stream teardown of key/value sub-readers.

Common situations: Query cancellation paths that close the datasource before page-source readers; double-close in custom readers; transient HDFS/S3 errors during teardown.

Related errors

AI-assisted analysis of prestodb/presto@55bb57d202 (2026-09-04). Data as JSON: /api/errors/b28d67e47db79431. Report an issue: GitHub.