[
https://issues.apache.org/jira/browse/HIVE-10535?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14626637#comment-14626637
]
Siddharth Seth commented on HIVE-10535:
---------------------------------------
- {code} ObjectCacheFactory.removeLlapQueryCache(queryId == null ? savedQueryId
: queryId); {code}
This could end up being null if the query completes before a fragment starts
executing. (possible when fragments are queued and the query is killed for
example). Should either not be invoked or removeLlapQueryCache would need to
check this (ConcurrentHashMap doesn't handle nulls).
- {code} return getLlapObjectCache(queryId);{code}
Does this need to be wrapped in an ObjectCacheWrapper, to prefix the queryId
correctly to the key ?
- removeLlapQueryCache - Minor: if (isDebugEnabled) around the debug statement
- makeKey in ObjectCacheWrapper. Does this end up changing the key from
queryId__HASH_MAP to queryId_HASH_MAP. Not sure if this is significant
Changing ExecutionContext is a little strange, but I see why it's required -
since the queryId will only be available after parsing the payload. We'll
likely be able to avoid this later once LLAP execution is a little more
specialized, and parses the payloads up front.
Rest looks good to me. Would prefer if [~vikram.dixit] or [~hagleitn] took a
quick look at the changes to the cache and keys though.
> LLAP: Cleanup map join cache when a query completes
> ---------------------------------------------------
>
> Key: HIVE-10535
> URL: https://issues.apache.org/jira/browse/HIVE-10535
> Project: Hive
> Issue Type: Sub-task
> Reporter: Siddharth Seth
> Assignee: Sergey Shelukhin
> Fix For: llap
>
> Attachments: HIVE-10533.01.patch, HIVE-10533.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
