hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Siddharth Seth (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HIVE-9976) Possible race condition in DynamicPartitionPruner for <200ms tasks
Date Wed, 25 Mar 2015 20:20:53 GMT

     [ https://issues.apache.org/jira/browse/HIVE-9976?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Siddharth Seth updated HIVE-9976:
---------------------------------
    Attachment: HIVE-9976.2.patch

Thanks for the review. Updated patch with comments addressed, and some more changes.

bq. Not your fault - but there are 2 paths through HiveSplitGenerator.
Moved the methods into SplitGrouper. There's a static cache in there which seems a little
strange. Will create a follow up jira to investigate this. For now I've changed that to a
ConcurrentMap since split generation can run in parallel.

bq. i see you've fixed calling close consistently on the data input stream. maybe use try{}finally
there?
Fixed. There was a bug with some of the other conditions which I'd changed. Fixed that as
well.

bq. it seems you're setting numexpectedevents to 0 first and then turn around and call decrement.
Why not just set to -1? Also - why atomic integers? as far as i can tell all access to these
maps is synchronized.
numExpectedEvents is decremented for each column for which a source will send events. That's
used to track total number of expected events from that source. Added a comment for this.
Moved from AtomicIntegers to MutableInt - this was just to avoid re-inserting the Integer
into the map, and not for thread safety.

bq. does it make sense to make initialize in the pruner private now? (can't be used to init
anymore - only from the constr). Also, the parameters aren't used anymore, right?
Done, along with some other methods.


> Possible race condition in DynamicPartitionPruner for <200ms tasks
> ------------------------------------------------------------------
>
>                 Key: HIVE-9976
>                 URL: https://issues.apache.org/jira/browse/HIVE-9976
>             Project: Hive
>          Issue Type: Bug
>          Components: Tez
>    Affects Versions: 1.0.0
>            Reporter: Gopal V
>            Assignee: Siddharth Seth
>         Attachments: HIVE-9976.1.patch, HIVE-9976.2.patch, llap_vertex_200ms.png
>
>
> Race condition in the DynamicPartitionPruner between DynamicPartitionPruner::processVertex()
and DynamicPartitionpruner::addEvent() for tasks which respond with both the result and success
in a single heartbeat sequence.
> {code}
> 2015-03-16 07:05:01,589 ERROR [InputInitializer [Map 1] #0] tez.DynamicPartitionPruner:
Expecting: 1, received: 0
> 2015-03-16 07:05:01,590 ERROR [Dispatcher thread: Central] impl.VertexImpl: Vertex Input:
store_sales initializer failed, vertex=vertex_1424502260528_1113_4_04 [Map 1]
> org.apache.tez.dag.app.dag.impl.AMUserCodeException: org.apache.hadoop.hive.ql.metadata.HiveException:
Incorrect event count in dynamic parition pruning
> {code}
> !llap_vertex_200ms.png!
> All 4 upstream vertices of Map 1 need to finish within ~200ms to trigger this, which
seems to be consistently happening with LLAP.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message