private void SerializeRecordBatch(ReusableMemoryStream stream, ISerializer keySerializer,
ISerializer valueSerializer)
{
// Starting Produce request V3, messages are encoded in the new RecordBatch.
var batch = new RecordBatch
{
CompressionCodec = CompressionCodec,
Records = Messages.Select(message =>
{
SerializeMessageIfNotSized(ref message, keySerializer, valueSerializer, stream.Pool);
return(new Record
{
Key = message.Key,
Value = message.Value,
Headers = message.Headers,
Timestamp = message.TimeStamp,
// If the serializer is not compatible, we already resolved this
// previously, so it's ok if the cast returns null
KeySerializer = keySerializer as ISizableSerializer,
ValueSerializer = valueSerializer as ISizableSerializer,
SerializedKeyValue = message.SerializedKeyValue,
});
}),
};
Basics.WriteWithSize(stream, batch.Serialize);
}
+ 4; // batchLength
/// <remarks>
/// From the official protocol documentation available at
/// https://cwiki.apache.org/confluence/display/KAFKA/A+Guide+To+The+Kafka+Protocol#AGuideToTheKafkaProtocol-FetchAPI
/// "As an optimization the server is allowed to return a partial message at the end of the message set. Clients should handle this case."
/// If the end of the RecordBatch exceeds the length of the whole response (= endOfAllBatches), we should discard this RecordBatch.
/// </remarks>
public static RecordBatch Deserialize(ReusableMemoryStream input, Deserializers deserializers, long endOfAllBatches)
{
var recordBatch = new RecordBatch();
if (input.Position + BytesNecessaryToGetLength > endOfAllBatches)
{
throw new ProtocolException(
$"Trying to read a batch record at {input.Position} and the end of all batches is {endOfAllBatches}."
+ $" There is not enough bytes remaining to even read the first fields...");
}
recordBatch.BaseOffset = BigEndianConverter.ReadInt64(input);
var batchLength = BigEndianConverter.ReadInt32(input);
var endOfBatch = input.Position + batchLength;
if (endOfAllBatches < endOfBatch)
{
// Partial message, CRCs won't match, return here so the CRC check doesn't throw
return(null);
}
recordBatch.PartitionLeaderEpoch = BigEndianConverter.ReadInt32(input);
var magic = input.ReadByte();
// Current magic value is 2
if ((uint)magic != 2)
{
throw new UnsupportedMagicByteVersion((byte)magic, "2");
}
var crc = (uint)BigEndianConverter.ReadInt32(input);
var afterCrcPosition = input.Position; // The crc is calculated starting from this position
Crc32.CheckCrcCastagnoli((int)crc, input, afterCrcPosition, endOfBatch - afterCrcPosition);
var attributes = BigEndianConverter.ReadInt16(input);
recordBatch.CompressionCodec = (CompressionCodec)(attributes & CompressionCodecMask);
recordBatch.IsTransactional = (attributes & TransactionalFlagMask) != 0;
recordBatch.IsControl = (attributes & ControlFlagMask) != 0;
recordBatch.TimestampType = (attributes & TimestampTypeMask) > 0
? TimestampType.LogAppendTime
: TimestampType.CreateTime;
var lastOffsetDelta = BigEndianConverter.ReadInt32(input);
var firstTimestamp = BigEndianConverter.ReadInt64(input);
var maxTimestamp = BigEndianConverter.ReadInt64(input);
recordBatch.ProducerId = BigEndianConverter.ReadInt64(input);
recordBatch.ProducerEpoch = BigEndianConverter.ReadInt16(input);
recordBatch.BaseSequence = BigEndianConverter.ReadInt32(input);
var numberOfRecords = BigEndianConverter.ReadInt32(input);
recordBatch.Records = DeserializeRecords(recordBatch, input, numberOfRecords, endOfBatch, firstTimestamp, deserializers);
return(recordBatch);
}
internal static List <ResponseMessage> DeserializeRecordBatch(ReusableMemoryStream stream,
Deserializers deserializers)
{
var list = ResponseMessageListPool.Reserve();
// First of all, we need to check if it is really a recordBatch (i.e. magic byte = 2)
// For that, we need to fast-forward to the magic byte. Depending on its value, we go back
// and deserialize following the corresponding format. (Kafka is amazing).
var startOfBatchOffset = stream.Position;
stream.Position += HeaderSize;
var magic = stream.ReadByte();
stream.Position = startOfBatchOffset;
if (magic < 2)
{
// We were mistaken, this is not a recordBatch but a message set!
return(DeserializeMessageSet(stream, deserializers));
}
// Now we know what we received is a proper recordBatch
var size = BigEndianConverter.ReadInt32(stream);
var endOfAllBatches = stream.Position + size;
if (stream.Length < endOfAllBatches)
{
throw new ProtocolException($"Fetch response advertise record batches of total size {size},"
+ $" but the stream only contains {stream.Length - stream.Position} byte remaining");
}
while (stream.Position < endOfAllBatches)
{
var recordBatch = RecordBatch.Deserialize(stream, deserializers, endOfAllBatches);
if (recordBatch == null)
{
// We received a partial record batch, discard it
stream.Position = endOfAllBatches;
break;
}
list.AddRange(recordBatch.Records.Select(record => new ResponseMessage
{
Message = new Message {
Key = record.Key, Value = record.Value, TimeStamp = record.Timestamp
},
Offset = record.Offset
}));
}
return(list);
}
public static IEnumerable <Record> DeserializeRecords(RecordBatch recordBatch, ReusableMemoryStream input, int numberOfRecords, long endOfBatch,
long firstTimeStamp, Deserializers deserializers)
{
if (recordBatch.CompressionCodec == CompressionCodec.None)
{
return(DeserializeRecordsUncompressed(recordBatch, input, numberOfRecords, endOfBatch, firstTimeStamp,
deserializers));
}
using (var uncompressedStream = input.Pool.Reserve())
{
Basics.Uncompress(uncompressedStream, input.GetBuffer(), (int)input.Position,
(int)(endOfBatch - input.Position), recordBatch.CompressionCodec);
input.Position = endOfBatch;
return(new List <Record>(DeserializeRecordsUncompressed(recordBatch, uncompressedStream, numberOfRecords, endOfBatch, firstTimeStamp,
deserializers))); // We use a list here to force iteration to take place, so that we can release uncompressedStream
}
}
// Used only in tests, and we only support byte array data
public void Serialize(ReusableMemoryStream stream, object extra, Basics.ApiVersion version)
{
BigEndianConverter.Write(stream, Partition);
BigEndianConverter.Write(stream, (short)ErrorCode);
BigEndianConverter.Write(stream, HighWatermarkOffset);
if (version >= Basics.ApiVersion.V4)
{
BigEndianConverter.Write(stream, LastStableOffset);
Basics.WriteArray(stream, AbortedTransactions);
}
if (version >= Basics.ApiVersion.V5)
{
BigEndianConverter.Write(stream, LogStartOffset);
}
if (version >= Basics.ApiVersion.V3)
{
var batch = new RecordBatch
{
CompressionCodec = Compression,
Records = Messages.Select(responseMessage => new Record
{
Key = responseMessage.Message.Key,
Value = responseMessage.Message.Value,
Timestamp = responseMessage.Message.TimeStamp,
KeySerializer = null,
ValueSerializer = null
}),
};
Basics.WriteWithSize(stream, batch.Serialize);
return;
}
Basics.WriteWithSize(stream, Messages, (s, l) =>
{
foreach (var m in l)
{
BigEndianConverter.Write(s, m.Offset);
Basics.WriteWithSize(s, m.Message,
(st, msg) =>
msg.Serialize(st, CompressionCodec.None, extra as Serializers,
version == Basics.ApiVersion.V2 ? MessageVersion.V1 : MessageVersion.V0));
}
});
}
private void SerializeRecordBatch(ReusableMemoryStream stream, ISerializer keySerializer,
ISerializer valueSerializer)
{
// Starting Produce request V3, messages are encoded in the new RecordBatch.
var batch = new RecordBatch
{
CompressionCodec = CompressionCodec,
Records = Messages.Select(message => new Record
{
Key = EnsureSizedSerializable(message.Key, keySerializer),
Value = EnsureSizedSerializable(message.Value, valueSerializer),
Headers = message.Headers,
Timestamp = message.TimeStamp,
// If the serializer is not compatible, we already resolved this
// previously, so it's ok if the cast returns null
KeySerializer = keySerializer as ISizableSerializer,
ValueSerializer = valueSerializer as ISizableSerializer
}),
};
Basics.WriteSizeInBytes(stream, batch.Serialize);
}
public static IEnumerable <Record> DeserializeRecordsUncompressed(RecordBatch recordBatch, ReusableMemoryStream input, int numberOfRecords, long endOfBatch,
long firstTimeStamp, Deserializers deserializers)
{
for (var i = 0; i < numberOfRecords; i++)
{
var length = VarIntConverter.ReadAsInt32(input);
if (input.Length - input.Position < length)
{
throw new ProtocolException(
$"Record said it was of length {length}, but actually only {input.Length - input.Position} bytes remain");
}
var attributes = input.ReadByte(); // ignored for now
var timeStampDelta = VarIntConverter.ReadAsInt64(input);
var offsetDelta = VarIntConverter.ReadAsInt32(input);
var keyLength = VarIntConverter.ReadAsInt32(input);
var key = keyLength == -1 ? null : deserializers.Item1.Deserialize(input, keyLength);
var valueLength = VarIntConverter.ReadAsInt32(input);
var value = valueLength == -1 ? null : deserializers.Item1.Deserialize(input, valueLength);
var headersCount = VarIntConverter.ReadAsInt32(input);
var headers = new List <KafkaRecordHeader>(headersCount);
for (var j = 0; j < headersCount; j++)
{
headers.Add(Record.DeserializeHeader(input));
}
yield return(new Record
{
Headers = headers,
Key = key,
Timestamp = firstTimeStamp + timeStampDelta,
Value = value,
Offset = recordBatch.BaseOffset + offsetDelta
});
}
}
