hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "jiraposter@reviews.apache.org (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-2218) speedup addInputPaths
Date Wed, 15 Jun 2011 19:00:50 GMT

    [ https://issues.apache.org/jira/browse/HIVE-2218?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13049963#comment-13049963
] 

jiraposter@reviews.apache.org commented on HIVE-2218:
-----------------------------------------------------


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/898/#review842
-----------------------------------------------------------



trunk/ql/src/java/org/apache/hadoop/hive/ql/exec/ExecDriver.java
<https://reviews.apache.org/r/898/#comment1826>

    This change will make the order of paths in pathProcessed non-deterministic. This means
mapred.input.dir will have not have the same order as before. Not sure if it is safe or not,
but if you change HashSet with LinkedHashSet, the order will be preserved.



trunk/ql/src/java/org/apache/hadoop/hive/ql/exec/ExecDriver.java
<https://reviews.apache.org/r/898/#comment1828>

    Here needs some comments why this case doesn't need to check empty paths. 
    
    In terms of efficiency, it seems to me that checking empty paths is not the most expensive
part (# of RPCs is large but each listStatus() should be fast). Also we should be able to
cache (needs to extend Utilities.isEmpty) the results of listStatus for each path, which are
anyway needed in other operations (compute splits etc). If 



trunk/ql/src/java/org/apache/hadoop/hive/ql/exec/ExecDriver.java
<https://reviews.apache.org/r/898/#comment1829>

    It seem that we are doing redundant work as FileInputFormat.setInputPaths(JobConf, CommaSeparatedString).
I think it would be safer and cleaner to first get an array of paths and call: 
    
    FileInputFormat.setInputPaths(StringUtils.stringToPath(String[] paths))



trunk/ql/src/java/org/apache/hadoop/hive/ql/exec/Utilities.java
<https://reviews.apache.org/r/898/#comment1830>

    indentation



trunk/ql/src/java/org/apache/hadoop/hive/ql/io/CombineHiveInputFormat.java
<https://reviews.apache.org/r/898/#comment1833>

    why do we need it here?



trunk/ql/src/java/org/apache/hadoop/hive/ql/io/CombineHiveInputFormat.java
<https://reviews.apache.org/r/898/#comment1834>

    ! mrwork.getPartDescToRework().isEmpty()



trunk/ql/src/java/org/apache/hadoop/hive/ql/io/CombineHiveInputFormat.java
<https://reviews.apache.org/r/898/#comment1835>

    The logic here is too complex and I think it's better to be refactored. Is the following
what you wanted? 
    
    if (all_partitions_are_rework()) {
      prepareNullCombineFilter(combine);
    } else {
      prepareNormalCombineFilter(combine);
    } 
    InputSplitShim[] iss = combine.getSplits()



trunk/ql/src/java/org/apache/hadoop/hive/ql/plan/MapredWork.java
<https://reviews.apache.org/r/898/#comment1832>

    Does here just need a HashSet<PartitionDesc> rather than a HashMap<PartitionDesc,
Boolean>?


- Ning


On 2011-06-14 21:09:13, Yongqiang He wrote:
bq.  
bq.  -----------------------------------------------------------
bq.  This is an automatically generated e-mail. To reply, visit:
bq.  https://reviews.apache.org/r/898/
bq.  -----------------------------------------------------------
bq.  
bq.  (Updated 2011-06-14 21:09:13)
bq.  
bq.  
bq.  Review request for hive.
bq.  
bq.  
bq.  Summary
bq.  -------
bq.  
bq.  speedup addInputPaths
bq.  
bq.  
bq.  This addresses bug HIVE-2218.
bq.      https://issues.apache.org/jira/browse/HIVE-2218
bq.  
bq.  
bq.  Diffs
bq.  -----
bq.  
bq.    trunk/ql/src/java/org/apache/hadoop/hive/ql/exec/ExecDriver.java 1135335 
bq.    trunk/ql/src/java/org/apache/hadoop/hive/ql/exec/Utilities.java 1135335 
bq.    trunk/ql/src/java/org/apache/hadoop/hive/ql/io/CombineHiveInputFormat.java 1135335

bq.    trunk/ql/src/java/org/apache/hadoop/hive/ql/plan/MapredWork.java 1135335 
bq.  
bq.  Diff: https://reviews.apache.org/r/898/diff
bq.  
bq.  
bq.  Testing
bq.  -------
bq.  
bq.  yes.
bq.  
bq.  
bq.  Thanks,
bq.  
bq.  Yongqiang
bq.  
bq.



> speedup addInputPaths
> ---------------------
>
>                 Key: HIVE-2218
>                 URL: https://issues.apache.org/jira/browse/HIVE-2218
>             Project: Hive
>          Issue Type: Improvement
>            Reporter: He Yongqiang
>            Assignee: He Yongqiang
>         Attachments: HIVE-2218.1.patch
>
>
> Speedup the addInputPaths for combined symlink inputformat, and added some other micro
optimizations which also work for normal cases.
> This can help reducing the start time of one query from 5 hours to less than 20 mins.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message