[
https://issues.apache.org/jira/browse/PHOENIX-5116?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16756898#comment-16756898
]
Lars Hofhansl edited comment on PHOENIX-5116 at 1/31/19 5:49 AM:
-----------------------------------------------------------------
Instead of the table, I had to check the plan whether it has salted key-ranges,
and in that case remove those from consideration in choosing a plan. I think
it's fine to leak a bit of internals to the plan chooser due to the nature of
heuristic plan selection.
The change as such should be very uncontroversial. I can LocalIndexIT and
DistinctPrefixFilterIT, both pass now.
was (Author: lhofhansl):
Instead of the table, I had to check the plan whether it has salted key-ranges,
and in that case remove those from consideration in choosing a plan.
The change as such should be very uncontroversial. I can LocalIndexIT and
DistinctPrefixFilterIT, both pass now.
> DistinctPrefixFilterIT fails.
> -----------------------------
>
> Key: PHOENIX-5116
> URL: https://issues.apache.org/jira/browse/PHOENIX-5116
> Project: Phoenix
> Issue Type: Task
> Reporter: Lars Hofhansl
> Priority: Major
> Attachments: 5116.txt
>
>
> Caused by this scenario outlined on PHOENIX-5109:
> {quote}
> This just does not end. Now there's a special case for the following:
> * compound key of VARCHAR
> * SELECT DISTINCT <prefix of the key>
> * Salting is enabled
> * There is a local index on an unrelated column
> Now Phoenix selects a full local index scan, which is heuristically less
> efficient... I think.
> The problem is that there are cases where a plan for a salted table
> nonetheless gets a key range that is not salted - does not have the salt
> range in it. That throws the logic off in this case.
> {quote}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
