[
https://issues.apache.org/jira/browse/HIVE-8136?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14498498#comment-14498498
]
Chaoyu Tang commented on HIVE-8136:
-----------------------------------
[~Ferd] I have a question, when Hive processes a query, it needs fetch
Table/Partition/Serde metadata from HMS, is this metadata cached at HMS client
side (as Table/Partition) or to be retrieved from HMS each time it is needed,
no matter the metadata is cached at HMS (as JDO entity object)? If the first
one is the case, it is easer and I think the change should be fine.
Otherwise, I think we should be more conservative and might even need add
DDL_EXCLUSIVE on anything which may affect other query behaviors (e.g. ADDPROP
etc). For example, querying data for an Avro or HBase backed tables, the Avro
schema url/literal or HBase column mapping is specified as a property. Can we
guarantee that Hive only need access these data once or all the metadata has
been cached at its HMS client?
> Reduce table locking
> --------------------
>
> Key: HIVE-8136
> URL: https://issues.apache.org/jira/browse/HIVE-8136
> Project: Hive
> Issue Type: Sub-task
> Reporter: Brock Noland
> Assignee: Ferdinand Xu
> Attachments: HIVE-8136.1.patch, HIVE-8136.patch
>
>
> When using ZK for concurrency control, some statements require an exclusive
> table lock when they are atomic. Such as setting a tables location.
> This JIRA is to analyze the scope of statements like ALTER TABLE and see if
> we can reduce the locking required.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
