[
https://issues.apache.org/jira/browse/HIVE-3279?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13420819#comment-13420819
]
Travis Crawford commented on HIVE-3279:
---------------------------------------
Looking at
[Table.getEmptyTable|https://github.com/apache/hive/blob/trunk/ql/src/java/org/apache/hadoop/hive/ql/metadata/Table.java#L121]
we see:
{code}
sd.setCols(new ArrayList<FieldSchema>());
{code}
What I believe happens is a new empty table is created, which initializes an
empty list of columns. No columns are actually set, because the serde reports
them at runtime.
Later, when initializing a Partition we check if the storage descriptor has
null for its columns. It actually has the empty list copied from the empty
table (not null) and we do not copy the table schema into the partition.
Typically tables/partitions have an explicitly defined schema so maybe this
use-case just hasn't come up? If you explicitly define the schema things work
as expected.
> Table schema not being copied to Partitions with no columns
> -----------------------------------------------------------
>
> Key: HIVE-3279
> URL: https://issues.apache.org/jira/browse/HIVE-3279
> Project: Hive
> Issue Type: Bug
> Reporter: Travis Crawford
> Assignee: Travis Crawford
>
> Hive has a feature where {{Partition}}'s without any defined columns use the
> {{Table}} schema. This happens in
> {{[Partition.initialize|https://github.com/apache/hive/blob/trunk/ql/src/java/org/apache/hadoop/hive/ql/metadata/Partition.java#L167]}}
> {code}
> // set default if columns are not set
> if (tPartition.getSd().getCols() == null) {
> if (table.getCols() != null) {
> tPartition.getSd().setCols(table.getCols());
> }
> }
> {code}
> There's an issue though, because
> {{[Table.getEmptyTable|https://github.com/apache/hive/blob/trunk/ql/src/java/org/apache/hadoop/hive/ql/metadata/Table.java#L121]}}
> initializes cols to an empty array, which of course is not null, causing the
> above feature to not work as expected.
> I'm not sure of the fix - is there a case where cols can indeed be null? I
> think the best thing to do here is:
> {code}
> - if (tPartition.getSd().getCols() == null) {
> + if (tPartition.getSd().getCols() == null ||
> tPartition.getSd().getCols().size() == 0) {
> {code}
> Thoughts?
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
