Package org.elasticsearch.xpack.esql.datasources.spi

Class RecordCappingInputStream

java.lang.Object

java.io.InputStream

java.io.FilterInputStream

org.elasticsearch.xpack.esql.datasources.spi.RecordCappingInputStream

All Implemented Interfaces:

Closeable, AutoCloseable

public abstract class RecordCappingInputStream
extends FilterInputStream

Stream wrapper that enforces the max_record_size byte cap on every record without decoding characters or visiting bytes outside the bulk-read fast path. The byte counter starts at zero, accumulates every byte (including line terminators), and resets when an unquoted record terminator is observed: '\n', '\r', or '\r\n'. The cap matches the terminator-inclusive byte length used by record splitters so an oversized record fails at every layer with the same threshold.

Format-specific subclasses provide the exception factory; the rest of the state machine — single-byte / bulk read accounting, CRLF spanning across reads, mark/reset lockout — is shared. This keeps the CSV and NDJSON cap streams from drifting in lockstep when one gets a CRLF or boundary-handling fix.

The hot path is read(byte[], int, int): it delegates a single bulk read to the underlying stream and then sweeps the freshly populated region in one tight loop. read() is wired to the same accounting but is not on the throughput-critical path (the downstream parser buffers in bulk).

mark/reset are disabled to keep the byte counter monotonic with the bytes the downstream parser consumes; a rewind would skew the cap relative to the actual record boundaries.

Semantics under lenient error policies: a thrown IOException from a bulk read leaves the underlying stream in an undefined position (the parser may have buffered partial bytes ahead of the trip), so the wrapper cannot resume after a cap failure. Callers that need row-level recovery under lenient policy must keep this wrapper out of those paths.

Field Summary

Fields inherited from class java.io.FilterInputStream

in

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	RecordCappingInputStream(InputStream in, int maxRecordBytes)

Method Summary

Modifier and Type	Method	Description
final void	mark(int readLimit)
final boolean	markSupported()
final int	read()
final int	read(byte[] buf, int off, int len)
protected abstract IOException	recordTooLarge()	Format-specific exception thrown when a record exceeds the byte cap.
final void	reset()

Methods inherited from class java.io.FilterInputStream

available, close, read, skip

Methods inherited from class java.io.InputStream

nullInputStream, readAllBytes, readNBytes, readNBytes, skipNBytes, transferTo

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

RecordCappingInputStream

protected RecordCappingInputStream(InputStream in,
 int maxRecordBytes)

Method Details

recordTooLarge

protected abstract IOException recordTooLarge()

Format-specific exception thrown when a record exceeds the byte cap. Subclasses should return an exception with the same "record exceeded max_record_size [N]" message shape used by the matching record splitter so log lines stay consistent across enforcement points.

read

public final int read()
               throws IOException

Overrides:

read in class FilterInputStream

Throws:

IOException

read

public final int read(byte[] buf,
 int off,
 int len)
               throws IOException

Overrides:

read in class FilterInputStream

Throws:

IOException

markSupported

public final boolean markSupported()

Overrides:

markSupported in class FilterInputStream

mark

public final void mark(int readLimit)

Overrides:

mark in class FilterInputStream

reset

public final void reset()
                 throws IOException

Overrides:

reset in class FilterInputStream

Throws:

IOException