[
https://issues.apache.org/jira/browse/HIVE-9251?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14264135#comment-14264135
]
Rui Li commented on HIVE-9251:
------------------------------
Hi [~xuefuz], I think {{hive.exec.reducers.bytes.per.reducer}} is not for
reducer memory. It's for how much data user wants a reducer to handle. Here's
the comment:
{noformat}
size per reducer.The default is 256Mb, i.e if the input size is 1G, it will use
4 reducers.
{noformat}
And here's how hive estimates # of reducers with it (the {{bytesPerReducer}}):
{code}
double bytes = Math.max(totalInputFileSize, bytesPerReducer);
int reducers = (int) Math.ceil(bytes / bytesPerReducer);
reducers = Math.max(1, reducers);
reducers = Math.min(maxReducers, reducers);
{code}
In MR mode, to configure reducer memory, I think some related properties are
{{mapreduce.reduce.memory.mb}}, {{mapreduce.map.java.opts}} etc.
Please let me know if I misunderstand something.
> SetSparkReducerParallelism is likely to set too small number of reducers
> [Spark Branch]
> ---------------------------------------------------------------------------------------
>
> Key: HIVE-9251
> URL: https://issues.apache.org/jira/browse/HIVE-9251
> Project: Hive
> Issue Type: Sub-task
> Components: Spark
> Reporter: Rui Li
> Assignee: Rui Li
> Attachments: HIVE-9251.1-spark.patch
>
>
> This may hurt performance or even lead to task failures. For example, spark's
> netty-based shuffle limits the max frame size to be 2G.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
