[
https://issues.apache.org/jira/browse/HIVE-19416?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16510366#comment-16510366
]
Steve Yeom commented on HIVE-19416:
-----------------------------------
Thank you Sergey for your comments. I really appreciate those.
1. I will add an "API spec" shortly to be clear or to be specific on
reader/updater.
But in short,
1.1 updater with alter... API method checks and remove
COLUMN_STATS_ACCURATE(CSA) for the table or partition
from the metastore. This is a global operation or metastore-wise persistent
operation.
One case is concurrent insert case.
1.2 reader checks whether the currently existing stats are
snapshot-isolation-level-compliant with the calling query and sets
CSA for the return object. Note that this is query by query return object.
Since the reader side check is per-query.
2. I will clear TODOs.
3. Regarding taking snapshot info
3.1. I think the Hive code is in the process of being changed to acquire
locks at the beginning of query optimization of a query before
taking a database snapshot. So if we are depending on locks for logical
consistency of Metadata objects, then we may have an issue.
3.2 Just FYI.
The concerned table-write-id-list getting locations in StatsOptimizer is
only used by a reader (not writer of a table/partition) so we don't
have the case where the current query is to increment write id of its
target tables. I.e., a global database snapshot is already taken
before starting optimization and the code changes for this jira in
StatsOptimizer is getting the info from the Metastore which is supposed to
guarantee Committed Read isolation there in the Metastore DBMS access.
> Create single version transactional table metastore statistics for
> aggregation queries
> --------------------------------------------------------------------------------------
>
> Key: HIVE-19416
> URL: https://issues.apache.org/jira/browse/HIVE-19416
> Project: Hive
> Issue Type: Bug
> Components: Transactions
> Reporter: Steve Yeom
> Assignee: Steve Yeom
> Priority: Major
>
> The system should use only statistics for aggregation queries like count on
> transactional tables.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
