[ https://issues.apache.org/jira/browse/PARQUET-1065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16318446#comment-16318446 ]
Zoltan Ivanfi commented on PARQUET-1065: ---------------------------------------- I think it is worth going through an example, the timestamp '2000-01-01 12:34:56' stored as an int96 and dumped with parquet-tools: $ parquet-tools dump hdfs://n1/user/hive/warehouse/test/11481f03a2ea6bed-b19656cd00000000_1937418586_data.0.parq | tail -n 1 value 1: R:0 D:1 V:117253024523396126668760320 Since 117253024523396126668760320 = 0x60FD4B3229000059682500, the 12 bytes are 00 60 FD 4B 32 29 00 00 | 59 68 25 00, where | shows the boundary between the time and the date parts. 00 60 FD 4B 32 29 00 00 is the time part, if we reverse the bytes we get 0x000029324BFD6000 = 45296 * 10^9 nanoseconds = 45296 seconds = 12 hours + 34 minutes + 56 seconds. 59 68 25 00 is the date part, if we reverse the bytes we get 0x00256859 = 2451545 as the Julian day number, which corresponds to 2000-01-01. For correct ordering based purely on numerical value, in comparisons the example above should not be interpreted as 0x0060FD4B3229000059682500 = 117253024523396126668760320 like it currently is, but as 0x00256859000029324BFD6000 = 45223023200227578716446720 instead. But since we do not want to introduce a new little-endian comparison order, we should just deprecate the ordering for this type. > 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)