[
https://issues.apache.org/jira/browse/TAJO-459?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13858116#comment-13858116
]
Min Zhou commented on TAJO-459:
-------------------------------
I'd like to discuss it since other partition type, hash/range/list can't
guarantee the value of partition columns in the same partition are the same.
At least I think we should distinguish regular result columns and partitioned
columns. I have no idea that what will going on if we insert into a column
partitioned table but with no partitioned key selected like the way below
{noformat}
insert overwrite into tbl(col1, col2) select l_orderkey, l_partkey, l_quantity
from lineitem
{noformat}
> Target tuples shouldn't include the partition keys
> --------------------------------------------------
>
> Key: TAJO-459
> URL: https://issues.apache.org/jira/browse/TAJO-459
> Project: Tajo
> Issue Type: Sub-task
> Components: planner/optimizer
> Affects Versions: 0.8-incubating
> Reporter: Min Zhou
> Fix For: 0.8-incubating
>
>
> Currently, if you create a column partitioned table
> {noformat}
> create table tbl (col1 int4, col2 int4, null_col int4) partition by
> column(key float8)
> {noformat}
> and later insert into it
> {noformat}
> insert overwrite into tbl(col1, col2, key) select l_orderkey, l_partkey,
> l_quantity from lineitem
> {noformat}
> From the code of ColumnPartitionedTableStoreExec.java, it seems that we
> didn't distinguish between the real result columns (col1 and col2 here) and
> the column partition keys. Just store the result including the partition
> keys. Hive doesn't save them like this way,cuz the value of partition
> columns in the same partition are the same.
> {noformat}
> public Tuple next() throws IOException {
> StringBuilder sb = new StringBuilder();
> while((tuple = child.next()) != null) {
> // set subpartition directory name
> sb.delete(0, sb.length());
> if (partitionColumnIndices != null) {
> for(int i = 0; i < partitionColumnIndices.length; i++) {
> Datum datum = tuple.get(partitionColumnIndices[i]);
> if(i > 0)
> sb.append("/");
> sb.append(partitionColumnNames[i]).append("=");
> sb.append(datum.asChars());
> }
> }
> // add tuple
> Appender appender = getAppender(sb.toString());
> appender.addTuple(tuple); // this line add the whole tuple into result
> record include the partition keys
> }
> ...
> }
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
