[
https://issues.apache.org/jira/browse/HIVE-19975?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16525659#comment-16525659
]
Steve Yeom commented on HIVE-19975:
-----------------------------------
[~sershe] has added test case for partition here:
"
Consider two scenarios:
1) Serially, with no parallel txns:
Write ID 1 inserts into partition k=1, partition stats table in metastore has
valid write ID list is (1).
Reader tries to read, table’s valid write ID list is (1), partition k=1 list in
stats table is (1), they are equivalent, all good.
Write ID 2 inserts into partition k=2, partition stats table in metastore has
valid write ID list is (1,2).
Reader tries to read, table’s valid write ID list is (1,2). For partition k=2,
the list is (1,2), equivalent returns true, all good. But for partition k=1,
the list is still (1), because writer 2 doesn’t touch it. Equivalent returns
false, stats cannot be used.
"
My answer to this test case scenarios is
For #1, I have already simulated all the possible cases regarding reader’s
starting point in time. It should work with the current patch.
> Checking writeIdList per table may not check the commit level of a partition
> on a partitioned table.
> ----------------------------------------------------------------------------------------------------
>
> Key: HIVE-19975
> URL: https://issues.apache.org/jira/browse/HIVE-19975
> Project: Hive
> Issue Type: Sub-task
> Components: Transactions
> Affects Versions: 4.0.0
> Reporter: Steve Yeom
> Assignee: Sergey Shelukhin
> Priority: Major
> Fix For: 4.0.0
>
>
> writeIdList is per table entity but stats for a partitioned table are per
> partition.
> I.e., each record in PARTITIONS has an independent stats.
> So if we check the validity of a partition's stats, we need to check in the
> context of
> a partiton.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
