fix: TestRecordBatchDecoding failing sporadically

[k-wall]

TestRecordBatchDecoding was sporadically failing for me on main (6d970c7). This test command normally brings out the failure:

go test -p 2  -run 'TestRecordBatch.*' -count 10
--- FAIL: TestRecordBatchDecoding (0.00s)
    record_test.go:291: invalid decode of snappy compressed record
        got {FirstOffset:0 PartitionLeaderEpoch:0 Version:2 Codec:snappy CompressionLevel:0 Control:false LogAppendTime:false LastOffsetDelta:0 FirstTimestamp:2016-11-22 20:49:55 +0000 GMT MaxTimestamp:1970-01-01 01:00:00 +0100 BST ProducerID:0 ProducerEpoch:0 FirstSequence:0 Records:[<*>(0xc00020e0e0){Headers:[<*>(0xc000210090){Key:[8 9 10] Value:[11 12]}] Attributes:0 TimestampDelta:5ms OffsetDelta:0 Key:[1 2 3 4] Value:[5 6 7] length:{startOffset:0 length:0}}] PartialTrailingRecord:false IsTransactional:false compressedRecords:<nil> recordsLen:21}
        wanted {FirstOffset:0 PartitionLeaderEpoch:0 Version:2 Codec:snappy CompressionLevel:0 Control:false LogAppendTime:false LastOffsetDelta:0 FirstTimestamp:2016-11-22 20:49:55 +0000 GMT MaxTimestamp:1970-01-01 01:00:00 +0100 BST ProducerID:0 ProducerEpoch:0 FirstSequence:0 Records:[<*>(0x17b9000){Headers:[<*>(0x17b4680){Key:[8 9 10] Value:[11 12]}] Attributes:0 TimestampDelta:5ms OffsetDelta:0 Key:[1 2 3 4] Value:[5 6 7] length:{startOffset:0 length:20}}] PartialTrailingRecord:false IsTransactional:false compressedRecords:<nil> recordsLen:21}
--- FAIL: TestRecordBatchDecoding (0.00s)
    record_test.go:291: invalid decode of snappy compressed record
        got {FirstOffset:0 PartitionLeaderEpoch:0 Version:2 Codec:snappy CompressionLevel:0 Control:false LogAppendTime:false LastOffsetDelta:0 FirstTimestamp:2016-11-22 20:49:55 +0000 GMT MaxTimestamp:1970-01-01 01:00:00 +0100 BST ProducerID:0 ProducerEpoch:0 FirstSequence:0 Records:[<*>(0xc0001d60e0){Headers:[<*>(0xc000c160c0){Key:[8 9 10] Value:[11 12]}] Attributes:0 TimestampDelta:5ms OffsetDelta:0 Key:[1 2 3 4] Value:[5 6 7] length:{startOffset:0 length:0}}] PartialTrailingRecord:false IsTransactional:false compressedRecords:<nil> recordsLen:21}
        wanted {FirstOffset:0 PartitionLeaderEpoch:0 Version:2 Codec:snappy CompressionLevel:0 Control:false LogAppendTime:false LastOffsetDelta:0 FirstTimestamp:2016-11-22 20:49:55 +0000 GMT MaxTimestamp:1970-01-01 01:00:00 +0100 BST ProducerID:0 ProducerEpoch:0 FirstSequence:0 Records:[<*>(0x17b9000){Headers:[<*>(0x17b4680){Key:[8 9 10] Value:[11 12]}] Attributes:0 TimestampDelta:5ms OffsetDelta:0 Key:[1 2 3 4] Value:[5 6 7] length:{startOffset:0 length:20}}] PartialTrailingRecord:false IsTransactional:false compressedRecords:<nil> recordsLen:21}
FAIL

The issues has been brought out by the switch to parallel tests. TestRecordBatchDecoding and TestRecordBatchEncoding use the same test data - which includes *Record references. If TestRecordBatchEncoding races with TestRecordBatchDecoding, the action of the encode() causes the mutation of Record.varintLengthField within the test data. This can then cause the test assertions made by TestRecordBatchDecoding to fail.

[dnwe]

Thanks! Good catch