[
https://issues.apache.org/jira/browse/IMPALA-8215?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16846941#comment-16846941
]
ASF subversion and git services commented on IMPALA-8215:
---------------------------------------------------------
Commit dfde59e587bd75d012c31ddf9c95ab315bf5460f in impala's branch
refs/heads/master from Todd Lipcon
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=dfde59e ]
IMPALA-8215, IMPALA-8458. Fix setting stats without setting NDVs in
local-catalog mode
This removes an old workaround in which columns with numDVs=-1 were
skipped when sending stats from catalogd to impalad. This workaround was
trying to fix some weird behavior with boolean stats but broke other
things, as noted in the JIRA. Additionally, it was used as a way to
avoid sending stats on HDFS clustering columns (which don't have
computed stats).
This change changes the workaround to be a bit more careful: we
explicitly ignore HDFS clustering columns, and for booleans, we use the
numNulls instead of numDVs to determine whether we have enough
information to infer numDVs.
NOTE: HBase table keys are also considered "clustering columns", but
those _do_ need computed stats, because HBase is range-partitioned, not
value-partitioned.
While I was in this section of the code, I stumbled upon IMPALA-8215, a
bug in which the 'numTrues' of boolean columns was getting set to '1'
for no apparent reason. I fixed that while I was in the area.
Change-Id: Ic0b95de22954c7ad6715143fc42a1506289c095f
Reviewed-on: http://gerrit.cloudera.org:8080/13382
Tested-by: Todd Lipcon <t...@apache.org>
Reviewed-by: Tim Armstrong <tarmstr...@cloudera.com>
> Don't set numTrue = 1
> ---------------------
>
> Key: IMPALA-8215
> URL: https://issues.apache.org/jira/browse/IMPALA-8215
> Project: IMPALA
> Issue Type: Sub-task
> Components: Frontend
> Reporter: Tim Armstrong
> Assignee: Todd Lipcon
> Priority: Major
> Labels: newbie, ramp-up
> Fix For: Impala 3.3.0
>
>
> See the parent task - there's an obvious bug where we set numTrues = 1 for no
> obvious reason. We should change the code and update any tests.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
