Subject Re[3]: [firebird-support] Find duplicate multi-row entries
Author Daniel Miller
Post date 2017-10-01T22:31:57Z
Maybe a little progress.

This procedure does exactly what I want...but not exactly HOW I want:

CREATE PROCEDURE CHECK_ROUTE_EXISTS (
    CHECK_NODE smallint )
RETURNS (
    NODE smallint,
    ROUTE_UPDATED timestamp )
AS
declare variable PLACEHOLDER timestamp;
begin
  /* Assignment for valid return value */
  NODE = CHECK_NODE;
  for select RH1.ROUTE_UPDATED
    from ROUTE_HISTORY RH1
    join ROUTES R1 using (NODE, ROUTE_INDEX, LINK_NODE, QUALITY)
    where RH1.NODE = :CHECK_NODE and RH1.ROUTE_INDEX = 1
    order by NODE, ROUTE_UPDATED, ROUTE_INDEX
    into PLACEHOLDER do
      for select RH2.ROUTE_UPDATED
      from ROUTE_HISTORY RH2
      join ROUTES R2 using (NODE, ROUTE_INDEX, LINK_NODE, QUALITY)
      where RH2.NODE = :CHECK_NODE and RH2.ROUTE_INDEX = 2
      order by NODE, ROUTE_UPDATED, ROUTE_INDEX
      into PLACEHOLDER do
        for select RH3.ROUTE_UPDATED
        from ROUTE_HISTORY RH3
        join ROUTES R3 using (NODE, ROUTE_INDEX, LINK_NODE, QUALITY)
        where RH3.NODE = :CHECK_NODE and RH3.ROUTE_INDEX = 3
        order by NODE, ROUTE_UPDATED, ROUTE_INDEX
        into PLACEHOLDER do
          for select RH4.ROUTE_UPDATED
          from ROUTE_HISTORY RH4
          join ROUTES R4 using (NODE, ROUTE_INDEX, LINK_NODE, QUALITY)
          where RH4.NODE = :CHECK_NODE and RH4.ROUTE_INDEX = 4
          order by NODE, ROUTE_UPDATED, ROUTE_INDEX
          into :ROUTE_UPDATED do suspend;
  /* If matched across 4 levels this is a duplicate.
      otherwise ROUTE_UPDATED will be NULL */
END

So...this gives me a list of all duplicates for a given NODE. Since I now have something somewhat functional - some optimization questions:

1.  I don't think I would gain anything by tweaking the above into using recursion - but if someone thinks so...please share.

2.  I would like to be able to run this against/with other queries.  Among other reasons - I want to look at doing a bulk delete in my history table.  At the moment, I can "manually" run something like:
    SELECT p.NODE, p.ROUTE_UPDATED FROM CHECK_ROUTE_EXISTS ('CHECK_NODE') p;

but I'm not sure how to apply this against my entire NODES table, nor how to run a delete with it.  I tried:
    delete from ROUTE_HISTORY H where H.NODE=5003 and H.ROUTE_UPDATED in (select ROUTE_UPDATED from CHECK_ROUTE_EXISTS(5003))

but that broke my poor server's brain - I had to stop the statement.

--
Daniel

On 9/30/2017 9:39:31 PM, "Daniel Miller" <dmiller@...> wrote:

Set,

Thank you. This definitely does something...to be honest I'm not certain exactly what... I'm definitely going to need to study this for a while. I sincerely appreciate the effort - though I think I am going to reconsider my table structure.

--
Daniel

On 9/28/2017 1:42:18 PM, "setysvar setysvar@... [firebird-support]" <firebird-support@yahoogroups.com> wrote:

Sorry again, Daniel, I had to get to a place where I could test things before replying. My problem was that what I'd done before was to try to find identical sets, not trying to find unique sets, and that made me mess up the logic. I hope this query will get you what you want (and this time I've tested before replying):

with tmp(NODE, ROUTE_UPDATED) as
(select distinct r.NODE, rh.ROUTE_UPDATED
 from routes r
 left join route_history rh on r.NODE = rh.NODE),
new_routes(NODE) as
(select distinct t.NODE
 from tmp t
 where exists(select *
              from routes r
              left join route_history rh on r.NODE = rh.NODE
                                        and r.ROUTE_INDEX = rh.ROUTE_INDEX
                                        and r.LINK_NODE = rh.LINK_NODE
                                        and r.QUALITY is not distinct from rh.QUALITY
                                        and t.ROUTE_UPDATED = rh.ROUTE_UPDATED
              where t.NODE = r.NODE
                and rh.NODE is null)
    or exists(select *
              from route_history rh
              left join routes r on r.NODE = rh.NODE
                                and r.ROUTE_INDEX = rh.ROUTE_INDEX
                                and r.LINK_NODE = rh.LINK_NODE
                                and r.QUALITY is not distinct from rh.QUALITY
              where t.NODE = rh.NODE
                and t.ROUTE_UPDATED = rh.ROUTE_UPDATED
                and r.NODE is null))
select r.NODE, r.ROUTE_INDEX, r.LINK_NODE, r.QUALITY
from new_routes nr
join routes r on nr.NODE = r.NODE

HTH,
Set

Den 28.09.2017 11:00, skrev 'Daniel Miller' dmiller@... [firebird-support]:
This appears no different than a simple:

    select * from routes

So...not there yet.  I think what I'm fighting is the lack of pivot or crosstab functionality - and I may have no choice (though it offends me deeply) but to setup my archive table for this comparison.  Either via the hated structure of column names ROUTE1, ROUTE2, ROUTE8...or the only slightly less offensive VARCHAR concatenation.

A concatenated LIST seems like it would be a elegant solution - and while operating over the whole table is time-consuming a single node isn't too bad.  And if I can get the history filtered and old duplicate records deleted it would be fine.  But the inconsistency is killing me.

There MUST be a Firebird-friendly, normalized, compact storage structure that will work.  I'm just not seeing it.
--
Daniel

On 9/28/2017 1:19:18 AM, "Svein Erling Tysvær setysvar@... [firebird-support]" <firebird-support@yahoogroups.com> wrote:

Sorry again, Daniel, fourth query required (I forgot that things may vary depending on time, this is something I've never done before):

with tmp(NODE, ROUTE_UPDATED) as
(select distinct rh.NODE, rh.ROUTE_UPDATED 
 from route_history rh
 join routes r on rh.NODE = r.NODE), //*This join is only for performance reasons in case ROUTE_HISTORY contains a lot more records than ROUTES and may be removed*/
new_routes(NODE) as
(select distinct t.NODE
 from tmp t
 where not exists(select * 
                  from routes r
                  left join route_history rh on r.NODE = rh.NODE
                                            and r.ROUTE_INDEX = rh.ROUTE_INDEX
                                            and r.LINK_NODE = rh.LINK_NODE
                                            and r.QUALITY is not distinct from rh.QUALITY
                                            and t.ROUTE_UPDATED = rh.ROUTE_UPDATED
                  where t.NODE = r.NODE
                    and rh.NODE is null)
  and not exists(select * 
                  from route_history rh
                  left join routes r on r.NODE = rh.NODE
                                            and r.ROUTE_INDEX = rh.ROUTE_INDEX
                                            and r.LINK_NODE = rh.LINK_NODE
                                            and r.QUALITY is not distinct from rh.QUALITY
                  where t.NODE = rh.NODE
                    and t.ROUTE_UPDATED = rh.ROUTE_UPDATED
                    and r.NODE is null))
select r.NODE, r.ROUTE_INDEX, r.LINK_NODE, r.QUALITY
from new_routes nr
join routes r on nr.NODE = r.NODE

Hope I finally got it right,
Set

2017-09-28 10:01 GMT+02:00 Svein Erling Tysvær <setysvar@...>:
Sorry, I forgot that should probably use 'IS NOT DISTINCT' and not '=' for QUALITY.

Note that this checks only checks if there has ever been anything identical in ROUTE_HISTORY, it doesn't restrain itself to the latest entry only (i.e. the query I've written is slightly similar to the second query I wrote, it has to be extended if you want something slightly similar to the first query). Also, my third query doesn't consider duplicate rows (since the fields seem to be the primary key in ROUTE, it isn't theoretically possible in your example).

Set

2017-09-28 9:52 GMT+02:00 Svein Erling Tysvær <setysvar@...>:
OK, that's completely different, but I've done something similar once before. It is not possible to directly compare sets for equality, but it is possible to check that set A doesn't contain anything that isn't in set B and that set B doesn't contain anything that isn't in set A. I assume NODE is the one common denominator:

with tmp (NODE) as
(select distinct NODE from routes),
new_routes(NODE) as
(select t.NODE
 from tmp t
 where not exists(select * 
                  from routes r
                  left join route_history rh on r.NODE = rh.NODE
                                            and r.ROUTE_INDEX = rh.ROUTE_INDEX
                                            and r.LINK_NODE = rh.LINK_NODE
                                            and r.QUALITY = rh.QUALITY
                  where t.NODE = r.NODE
                    and rh.NODE is null)
  and not exists(select * 
                  from route_history rh
                  left join routes r on r.NODE = rh.NODE
                                            and r.ROUTE_INDEX = rh.ROUTE_INDEX
                                            and r.LINK_NODE = rh.LINK_NODE
                                            and r.QUALITY = rh.QUALITY
                  where t.NODE = rh.NODE
                    and r.NODE is null))
select r.NODE, r.ROUTE_INDEX, r.LINK_NODE, r.QUALITY
from new_routes nr
join routes r on nr.NODE = r.NODE

Does this get you the result you want?
Set

2017-09-28 9:19 GMT+02:00 'Daniel Miller' dmiller@... [firebird-support] <firebird-support@yahoogroups. com >:


Thanks - but that still doesn't work.  I get a huge return set with a ton of duplicates.  I think something's being lost in translation.  A simple match of one row isn't sufficient - I need to match the "set".

With the following six records:
5557    1    1116    0
5557    2    1115    0
5557    3    5104    2
5557    4    5073    2
5557    5    5222    2
5557    6    5110    2

That is the list of routes a given node has at a specific point in time.  So it's perfectly possible at another time to have fewer, or more, or different routes.  And I need to consider ALL the routes in play for a given timestamp to be a single "set".  So if today I have the above list of routes, and yesterday route index 4 was looking at a different node - that would constitute a different and unique set even though the other 5 records match.

5557    1    1116    0
5557    2    1115    0
5557    3    5104    2
5557    4    9999    2
5557    5    5222    2
5557    6    5110    2

So it's quite possible I will have lots of duplicates for an individual row in ROUTES when compared in ROUTE_HISTORY - what I'm trying to filter is the pattern of rows.  Only if ALL the rows of a current entries in ROUTES exist in ROUTE_HISTORY should it be considered a duplicate condition.

--
Daniel

On 9/27/2017 11:53:48 PM, "Svein Erling Tysvær setysvar@... [firebird-support]" <firebird-support@yahoogroups. com > wrote:

Sure it is possible to write such a query:

select r.NODE, r.ROUTE_INDEX, r.LINK_NODE, r.QUALITY
from routes r
left join route_history rh1 on r.node = rh1.node
                          and r.route_index = rh1.route_index
                          and r.link_node = rh1.link_node
                          and r.quality is not distinct from rh1.quality
left join route_history rh2 on rh1.node = rh2.node
                          and rh1.route_index = rh2.route_index
                          and rh1.route_updated < rh2.route_updated
where rh1.node is null 
  and rh2.node is null

I use two LEFT JOINS because I assume you want things to be returned if things are changed back, e.g. if you have

5557    1    1116    0
5557    1    1116    0

then you only want one row in route_history, but if you have

5557    1    1116    0
5557    1    1116    1
5557    1    1116    0

then you don't want the last record to be considered a duplicate, but know when it changed back. If ROUTE_INDEX is just a running number, you consider

5557    1    1116    0
5557    2    1116    0

to be duplicates, and there are no gaps between the ROUTE_INDEX for each NODE, then the query can be simplified:

select r.NODE, r.ROUTE_INDEX, r.LINK_NODE, r.QUALITY
from routes r
left join route_history rh1 on r.node = rh1.node
                          and r.route_index+1 = rh1.route_index
                          and r.link_node = rh1.link_node
                          and r.quality is not distinct from rh1.quality
where rh1.node is null 

Note that I use IS NOT DISTINCT so that NULLs are considered equal. If you prefer, the left joins can be replaced by (nested) NOT EXISTS.

HTH,
Set

2017-09-28 7:22 GMT+02:00 'Daniel Miller' dmiller@... [firebird-support] <firebird-support@yahoogroups. com >:


On 9/27/2017 9:20:54 PM, "liviuslivius liviuslivius@... [firebird-support]" <firebird-support@yahoogroups. com > wrote:

group by + having
is your friend.
E.g.
Select field1, field2, count(*)
From tablex
Group by field1, field2
Having count(*)>1

But in your scenario i do not know if this is the solution. Question is if record was changed and in next update "restored" and once again changed is this duplicate or no?
If yes above query is ok if no then procedure is your friend.

And for the future change your audit trigger and check before if there was any change
Thank you, but I this isn't quite what I need.  I need to match against all the records of a set (up to 8 rows per set) - not just individual rows.

--
Daniel