Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
Description
Hei
The to_unix_timestamp function is implemented by GenericUDFToUnixTimeStamp. It uses SimpleDateFormat to parse the time of the string type.
But SimpleDateFormat does not specify the Locale parameter, that is, the default locale of the jvm machine will be used. This will cause some non-English local machines to be unable to run similar sql like :
hive> select to_unix_timestamp('16/Mar/2017:12:25:01', 'dd/MMM/yyy:HH:mm:ss'); OK NULL hive> select unix_timestamp('16/Mar/2017:12:25:01', 'dd/MMM/yyy:HH:mm:ss'); OK NULL
At the same time, I found that in spark, to_unix_timestamp & unix_timestamp also use SimpleDateFormat, and spark uses Locale.US by default, but this will make it impossible to use local language syntax. For example, in the Chinese environment, I can parse this result correctly in hive,
hive> select to_unix_timestamp('16/三月/2017:12:25:01', 'dd/MMMM/yyy:HH:mm:ss'); OK 1489638301 Time taken: 0.147 seconds, Fetched: 1 row(s) OK
But spark will return Null.
Because English dates are more common dates, I think two SimpleDateFormats are needed. The new SimpleDateFormat is initialized with the Locale.ENGLISH parameter.
