[
https://issues.apache.org/jira/browse/HIVE-16503?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15982262#comment-15982262
]
Prasanth Jayachandran commented on HIVE-16503:
----------------------------------------------
bq. Is SESSIONS_PER_DEFAULT_QUEUE guaranteed to be >= 1?
The default value is 1 but it doesn't guard against values set by user. In .3
patch added guard against <=0 values and a unit test for the same.
bq. Does it make sense to add range validators for the new settings (0, 1.0)?
.1 patch had RangeValidator but then I removed it as >1.0 values are also
valid. Something like 120% oversubscription is a valid scenario. However under
subscription is not allowed and is guarded by the return value
{code}
return Math.max(maxSize, llapMaxSize);
{code}
where maxSize is the initial no conditional task size.
> LLAP: Oversubscribe memory for noconditional task size
> ------------------------------------------------------
>
> Key: HIVE-16503
> URL: https://issues.apache.org/jira/browse/HIVE-16503
> Project: Hive
> Issue Type: Improvement
> Components: llap
> Affects Versions: 3.0.0
> Reporter: Prasanth Jayachandran
> Assignee: Prasanth Jayachandran
> Attachments: HIVE-16503.1.patch, HIVE-16503.2.patch,
> HIVE-16503.3.patch
>
>
> When running map joins in llap, it can potentially use more memory for hash
> table loading (assuming other executors in the daemons have some memory to
> spare). This map join conversion decision has to be made during compilation
> that can provide some more room for LLAP.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)