asterixdb-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Michael J. Carey (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ASTERIXDB-1109) Index entries are not deleted on an open-type secondary index
Date Thu, 24 Sep 2015 00:57:04 GMT

    [ https://issues.apache.org/jira/browse/ASTERIXDB-1109?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14905597#comment-14905597
] 

Michael J. Carey commented on ASTERIXDB-1109:
---------------------------------------------

My take is that there are two problems:
1. The delete pipeline needs to think more like the insert pipeline.  Instead of using a closed
record constructor, it should use a typecast like the one in the delete pipeline that creates
an extended closed type (one with the indexed open fields tacked on to the end).  The bug
right now is that the downstream operators are expecting this - but in the delete case, it's
not happening.
2. The way that the index maintenance pipeline is being managed is not very good.  It would
be a lot better to only flow the fields that are relevant to indexes - it makes no sense (to
me) to be creating and/or flowing whole records.  Instead, just the PK and indexed SK's should
retained, and they should flow as Hyracks tuple fields.  This would require reworking how
bookkeeping is done in order for the indexes to know where to find what they need - right
now they probably "cheat" by just knowing where to look in a big record type.

> Index entries are not deleted on an open-type secondary index
> -------------------------------------------------------------
>
>                 Key: ASTERIXDB-1109
>                 URL: https://issues.apache.org/jira/browse/ASTERIXDB-1109
>             Project: Apache AsterixDB
>          Issue Type: Bug
>            Reporter: Taewoo Kim
>            Assignee: Ildar Absalyamov
>            Priority: Critical
>
> Currently, index entries (SK, PK) are not deleted on an open-type secondary index. This
issue was not surfaced due to the fact that every search after a secondary index search needs
to go through the primary index lookup. Therefore, un-deleted entries were removed during
this Pidx lookup.
> Now, we have an index-only plan that can skip the Pidx lookup after a Sidx search. Therefore,
the query result is not correct anymore.
>   



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message