[
https://issues.apache.org/jira/browse/HIVE-17257?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16117565#comment-16117565
]
Chao Sun commented on HIVE-17257:
---------------------------------
[~kellyzly]: the empty files maybe generated if the result set is empty and if
you have multiple mapper/reducers with file sink. Example:
{code}
set hive.execution.engine=spark;
set hive.auto.convert.join=false;
set mapreduce.job.reduces=1000;
create table dummy (a string);
insert overwrite directory '/tmp/test' select src.key from src join dummy on
src.key = dummy.a;
{code}
The above will generate 1000 empty files in /tmp/test.
[~xuefuz]: I need to revise the patch. There's an issue where HoS won't launch
task for the final merge job since the input data is empty.
> Hive should merge empty files
> -----------------------------
>
> Key: HIVE-17257
> URL: https://issues.apache.org/jira/browse/HIVE-17257
> Project: Hive
> Issue Type: Bug
> Reporter: Chao Sun
> Assignee: Chao Sun
> Attachments: HIVE-17257.0.patch
>
>
> Currently if merging file option is turned on and the dest dir contains large
> number of empty files, Hive will not trigger merge task:
> {code}
> private long getMergeSize(FileSystem inpFs, Path dirPath, long avgSize) {
> AverageSize averageSize = getAverageSize(inpFs, dirPath);
> if (averageSize.getTotalSize() <= 0) {
> return -1;
> }
> if (averageSize.getNumFiles() <= 1) {
> return -1;
> }
> if (averageSize.getTotalSize()/averageSize.getNumFiles() < avgSize) {
> return averageSize.getTotalSize();
> }
> return -1;
> }
> {code}
> This logic doesn't seem right as the it seems better to combine these empty
> files into one.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
