[
https://issues.apache.org/jira/browse/HIVE-20219?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-20219:
------------------------------------
Description:
-Discovered while adding tests in HIVE-20029
Step 1. Have a table with fully correct txn stats.
Step 2. Run some query that sets stats to invalid (e.g. parallel inserts).
Step 3. Run analyze without for columns...
Result is that while only basic stats are updated, the flag is set to true with
old column stats values.
I blame json; if stats state was stored in proper places this kind of bug would
be very unlikely.
Not sure if this would affect master, I'll add a test; ACID stats invalidation
definitely needs to invalidate all the stats, not just basic stats flag, when
invalidating-.
was:
Discovered while adding tests in HIVE-20029
Step 1. Have a table with fully correct txn stats.
Step 2. Run some query that sets stats to invalid (e.g. parallel inserts).
Step 3. Run analyze without for columns...
Result is that while only basic stats are updated, the flag is set to true with
old column stats values.
I blame json; if stats state was stored in proper places this kind of bug would
be very unlikely.
Not sure if this would affect master, I'll add a test; ACID stats invalidation
definitely needs to invalidate all the stats, not just basic stats flag, when
invalidating.
> verify that analyze and analyze for columns manage txn stats state correctly
> ----------------------------------------------------------------------------
>
> Key: HIVE-20219
> URL: https://issues.apache.org/jira/browse/HIVE-20219
> Project: Hive
> Issue Type: Sub-task
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Priority: Major
> Fix For: txnstats
>
>
> -Discovered while adding tests in HIVE-20029
> Step 1. Have a table with fully correct txn stats.
> Step 2. Run some query that sets stats to invalid (e.g. parallel inserts).
> Step 3. Run analyze without for columns...
> Result is that while only basic stats are updated, the flag is set to true
> with old column stats values.
> I blame json; if stats state was stored in proper places this kind of bug
> would be very unlikely.
> Not sure if this would affect master, I'll add a test; ACID stats
> invalidation definitely needs to invalidate all the stats, not just basic
> stats flag, when invalidating-.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
