[ https://issues.apache.org/jira/browse/PARQUET-1065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16318436#comment-16318436 ]
Zoltan Ivanfi commented on PARQUET-1065: ---------------------------------------- The Parquet specification does not talk about endianness (which is something that I think should be addressed), but it defines data in terms of Thrift structures and the language bindings (at least parquet-mr) directly use these Thrift structures for reading and writing. Based on the Thrift specification (and some actual data files as well), these Thrift structures have a big-endian byte order. To quote from the [Integer encoding|https://github.com/apache/thrift/blob/master/doc/specs/thrift-binary-protocol.md#integer-encoding] section of the Thrift specification: {quote}In the binary protocol integers are encoded with the most significant byte first (big endian byte order, aka network order). An int8 needs 1 byte, an int16 2, an int32 4 and an int64 needs 8 bytes.{quote} However, please note that there is no int96 type here, so that really should be specified in Parquet Format, but given that all other int types have a big-endian byte order, I don't think any other choice would make sense for int96. (Parquet-tools already interperts int96 values according to this ordering). Impala, however, simply writes the 12 bytes of it's little-endian in-memory representation into the consecutive bytes of an int96, so the values are meaningless for less-than or greater-than comparisons. > Deprecate type-defined sort ordering for INT96 type > --------------------------------------------------- > > Key: PARQUET-1065 > URL: https://issues.apache.org/jira/browse/PARQUET-1065 > Project: Parquet > Issue Type: Bug > Reporter: Zoltan Ivanfi > Assignee: Zoltan Ivanfi > > [parquet.thrift in > parquet-format|https://github.com/apache/parquet-format/blob/041708da1af52e7cb9288c331b542aa25b68a2b6/src/main/thrift/parquet.thrift#L37] > defines the the sort order for INT96 to be signed. > [ParquetMetadataConverter.java in > parquet-mr|https://github.com/apache/parquet-mr/blob/352b906996f392030bfd53b93e3cf4adb78d1a55/parquet-hadoop/src/main/java/org/apache/parquet/format/converter/ParquetMetadataConverter.java#L422] > uses unsigned ordering instead. In practice, INT96 is only used for > timestamps and neither signed nor unsigned ordering of the numeric values is > correct for this purpose. For this reason, the INT96 sort order should be > specified as undefined. > (As a special case, min == max signifies that all values are the same, and > can be considered valid even for undefined orderings.) -- This message was sent by Atlassian JIRA (v6.4.14#64029)