[
https://issues.apache.org/jira/browse/HIVE-21798?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16850479#comment-16850479
]
Karel Kolman commented on HIVE-21798:
-------------------------------------
[~gopalv] thx for your reply, ill try to dig into the acid v1 / v2 differences
a bit more for our use case.
OrcInputFormat#ACIDSplitStrategy#getSplits code has what seems to me a "return
empty split list / throw exception code" only code
{noformat}
static class ACIDSplitStrategyimplements SplitStrategy<OrcSplit> {
....
@Override
public List<OrcSplit> getSplits() throws IOException {
List<OrcSplit> splits = Lists.newArrayList();
if (acidOperationalProperties != null &&
acidOperationalProperties.isSplitUpdate()) {
return Collections.emptyList();
}
if (!deltas.isEmpty()) {
//since HIVE-17089 if here, then it's not an acid table so there should
never be any deltas
throw new IllegalStateException("Found unexpected deltas: " + deltas +
" in " + dir);
}
return splits;
}
{noformat}
since
https://github.com/apache/hive/commit/34b0e07a3bd5002b197b749e3e5a7992e196c237#diff-ede47c0c518f8577dd3895984f019330L1037
that seems kind of weird.
> OrcInputFormat#generateSplitsInfo returns 0 splits for a transactional
> delta-only table
> ---------------------------------------------------------------------------------------
>
> Key: HIVE-21798
> URL: https://issues.apache.org/jira/browse/HIVE-21798
> Project: Hive
> Issue Type: Bug
> Affects Versions: 3.1.0
> Reporter: Karel Kolman
> Priority: Major
>
> OrcInputFormat#generateSplitsInfo returns 0 splits for acid table which has
> no base - deltas only.
>
> Table data (no base)
> {noformat}
> $ hadoop fs -ls /apps/hive/warehouse/test_table
> drwxr-xr-x - hive hdfs 0 2019-05-28 13:13
> /apps/hive/warehouse/test_table/delta_0000001_0000001_0000
> $ hadoop fs -ls -r /apps/hive/warehouse/test_table/delta_0000001_0000001_0000
> Found 4 items
> -rw-r--r-- 3 hive hdfs 1302 2019-05-28 13:13
> /apps/hive/warehouse/test_table/delta_0000001_0000001_0000/bucket_00002
> -rw-r--r-- 3 hive hdfs 1250 2019-05-28 13:13
> /apps/hive/warehouse/test_table/delta_0000001_0000001_0000/bucket_00001
> -rw-r--r-- 3 hive hdfs 1215 2019-05-28 13:13
> /apps/hive/warehouse/test_table/delta_0000001_0000001_0000/bucket_00000
> -rw-r--r-- 3 hive hdfs 1 2019-05-28 13:13
> /apps/hive/warehouse/test_table/delta_0000001_0000001_0000/_orc_acid_version{noformat}
> Table created with
> {noformat}
> CREATE TABLE test_table (userid bigint)
> clustered by (userid)
> into 3 buckets
> stored as orc
> TBLPROPERTIES ('transactional'='true');
> INSERT INTO TABLE test_table SELECT * FROM source_table;
> {noformat}
>
> Is this expected behavior ?
>
> Change of behavior from earlier Hive versions seems to come from commit
>
> [https://github.com/apache/hive/commit/34b0e07a3bd5002b197b749e3e5a7992e196c237#diff-ede47c0c518f8577dd3895984f019330L1037]
> related to HIVE-17089.
> Earlier Hive versions return valid split data from
> OrcInputFormat#generateSplitsInfo.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
