trafodion-codereview mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From zellerh <...@git.apache.org>
Subject [GitHub] incubator-trafodion pull request: [TRAFODION-1887] Upsert into ali...
Date Tue, 15 Mar 2016 21:06:08 GMT
Github user zellerh commented on a diff in the pull request:

    https://github.com/apache/incubator-trafodion/pull/380#discussion_r56240661
  
    --- Diff: core/sql/optimizer/BindRelExpr.cpp ---
    @@ -10171,17 +10164,26 @@ matched clause of merge). If the upsert caused a row to be updated
in the
     base table then the old version of the row will have to be deleted from 
     indexes, and a new version inserted. Upsert is being transformed to merge
     so that we can delete the old version of an updated row from the index.
    +
    +Upsert is also converted into merge when there are omitted cols with default values and

    +TRAF_UPSERT_WITH_INSERT_DEFAULT_SEMANTICS is set to  OFF in case of aligned format table
or 
    +omitted current timestamp cols in case of non-aligned row format
    --- End diff --
    
    Can I suggest two changes - not sure if they are feasible, but let me give it a try and
ask for them.
    
    First, if we have a column with default current: Couldn't we just add an assignment of
the column to current, without transforming the upsert to a merge?
    
    Second, I would find it more clear if we would define UPSERT as something that either
inserts a new row or replaces an entire existing row, but does not merge old and new rows.
If a user wants a merge then that's what the MERGE command is for.
    
    What would it mean to do that? As far as I can tell, several things:
    
    * Automatically generate an assignment for columns with DEFAULT CURRENT, no need to transform
to MERGE because of that. This applies to both aligned and non-aligned format.
    * No need to transform to MERGE just because we have omitted columns in aligned format.
    * No need for the CQD for aligned format.
    * For non-aligned format, do one of two things:
      * Always generate all (default) values, even those not specified in the upsert command.
      * Have a CQD (could re-use the one now used for aligned) to suppress generation of defaults
and take a chance to see values from previous rows. This trades space savings for unclear
semantics.
    
    It's a bit late for this kind of design discussion, just want to bring it up here. If
you feel we should go with what you coded that's ok with me, it's a step up from the current
behavior. I can also file a JIRA for want I suggest and we can do that later.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

Mime
View raw message