[
https://issues.apache.org/jira/browse/HIVE-18661?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16672685#comment-16672685
]
Daniel Dai commented on HIVE-18661:
-----------------------------------
Note with current patch, it could happen:
1. Metastore A update table t1
2. Metastore B update table t1
3. Metastore B read notification about A's update, overwrite B's newer change
4. Metastore B will eventually read B's update from notification, bring back
B's newer change
I think it might be ok with this eventual consistent behavior. Otherwise, we
need some more complex mechanism to track which change is newer. cc [~thejas]
> CachedStore: Use metastore notification log events to update cache
> ------------------------------------------------------------------
>
> Key: HIVE-18661
> URL: https://issues.apache.org/jira/browse/HIVE-18661
> Project: Hive
> Issue Type: Sub-task
> Components: Metastore
> Reporter: Vaibhav Gumashta
> Assignee: mahesh kumar behera
> Priority: Major
> Labels: pull-request-available
> Attachments: HIVE-18661.02.patch
>
>
> Currently, a background thread updates the entire cache which is pretty
> inefficient. We capture the updates to metadata in NOTIFICATION_LOG table
> which is getting used in the Replication work. We should have the background
> thread apply these notifications to incrementally update the cache.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
