Open GoogleCodeExporter opened 8 years ago
Great test. This could be a problem within Hadoop itself, since the replicator
deposited the value correctly in th staging table. What happens if you insert
such a value directly into hive and then select it back? Ideally you should
select using a key as this will force map/reduce and run the date through the
full process we are using to generate materialized views.
(It is possible Hadoop has some of the exact time time issues we have been
correcting the replicator.)
Original comment by robert.h...@continuent.com
on 6 Dec 2014 at 4:31
This is an issue with Hive re-interpreting the timestamp value during the
map/reduce process.
During Map/Reduce we effectively do an INSERT INTO ... SELECT FROM
With Hive, that means we interpret the table data according to the set
datatype, then insert the interpreted value back into the new table.
The easiest change is to alter our velocity templates used by dllscan and set
TIMESTAMP to be treated as STRING types. This will prevent the interpretation.
Longer term, we should change the map/reduce to explicitly interpret TIMESTAMP
values as strings (which side-steps the interpretation), but this needs to be
considered alongside any changes we make to the supported target databases in
Hadoop; HBase and Impala probably do not suffer the same problems.
Original comment by mc.br...@continuent.com
on 9 Dec 2014 at 3:51
Original comment by linas.vi...@continuent.com
on 11 Dec 2014 at 2:43
Original comment by linas.vi...@continuent.com
on 19 Jan 2015 at 2:18
Original issue reported on code.google.com by
g.maxia
on 6 Dec 2014 at 3:06