compare primary/alias groups across two tables - sql

Question

Gday,
We have two tables that contain exactly the same structure. There are two columns "PrimaryAddress" and "AliasAddress". These are for email addresses and aliases. We want to find any records that need to be added to either side to keep the records in sync. The catch is that the primary name in one table might be listed as an alias in the other. The good news is that an address wont appear twice in the "AliasAddress" column.
TABLE A
PrimaryAddress~~~~~AliasAdress
chris#work~~~~~~~~~chris#home
chris#work~~~~~~~~~c#work
chris#work~~~~~~~~~theboss#work
chris#work~~~~~~~~~thatguy#aol
bob#test~~~~~~~~~~~test1#test
bob#test~~~~~~~~~~~charles#work
bob#test~~~~~~~~~~~chuck#aol
sally#mars~~~~~~~~~sally#nasa
sally#mars~~~~~~~~~sally#gmail
TABLE B
PrimaryAddress~~~~~AliasAdress
chris#home~~~~~~~~~chris#work
chris#home~~~~~~~~~c#work
chris#home~~~~~~~~~theboss#work
chris#home~~~~~~~~~thatguy#aol
bob#test~~~~~~~~~~~test1#test
bob#test~~~~~~~~~~~charles#work
sally#nasa~~~~~~~~~sally#mars
sally#nasa~~~~~~~~~sally#gmail
sally#nasa~~~~~~~~~ripley#nostromo
The expected result is to return the following missing records from both tables:
bob#test~~~~~~~~~~~chuck#aol
sally#nasa~~~~~~~~~ripley#nostromo
Note that the chris#* block is a total match because the sum of all the aliases (plus primary) is still the same regardless of which address is considered primary. It doesnt matter which address is primary as along as the sum of the entire primary group contains all entries in both tables.
I dont mind if this is run in two passes A->B and B->A but I just cant get my head around a solution.
Any help appreciated :)

Answer 1

drop TABLE #TABLEA
CREATE TABLE #TABLEA
([PrimaryAddress] varchar(10), [AliasAdress] varchar(12))
;
INSERT INTO #TABLEA
([PrimaryAddress], [AliasAdress])
VALUES
('chris#work', 'chris#home'),
('chris#work', 'c#work'),
('chris#work', 'theboss#work'),
('chris#work', 'thatguy#aol'),
('bob#test', 'test1#test'),
('bob#test', 'charles#work'),
('bob#test', 'chuck#aol'),
('sally#mars', 'sally#nasa'),
('sally#mars', 'sally#gmail')
;
drop TABLE #TABLEB
CREATE TABLE #TABLEB
([PrimaryAddress] varchar(10), [AliasAdress] varchar(15))
;
INSERT INTO #TABLEB
([PrimaryAddress], [AliasAdress])
VALUES
('chris#home', 'chris#work'),
('chris#home', 'c#work'),
('chris#home', 'theboss#work'),
('chris#home', 'thatguy#aol'),
('bob#test', 'test1#test'),
('bob#test', 'charles#work'),
('sally#nasa', 'sally#mars'),
('sally#nasa', 'sally#gmail'),
('sally#nasa', 'ripley#nostromo')
;
try the following
select a.PrimaryAddress,a.AliasAdress from #TABLEA a left join #TABLEB b on a.AliasAdress=b.AliasAdress or b.PrimaryAddress=a.AliasAdress
where b.PrimaryAddress is null
union all
select a.PrimaryAddress,a.AliasAdress from #TABLEB a left join #TABLEA b on a.AliasAdress=b.AliasAdress or b.PrimaryAddress=a.AliasAdress
where b.PrimaryAddress is null

Answer 2

So you want to compare table A and B, and find rows which are unqiue in either table. How about an outer join, followed by looking for NULL values:
SELECT ta.*, tb.*
FROM table_a ta
FULL OUTER JOIN table_b tb ON tb.PrimaryAddress = ta.PrimaryAddress
AND tb.AliasAddress = ta.AliasAddress
WHERE ta.PrimaryAddress IS NULL
OR tb.PrimaryAddress IS NULL
If I understand the question correctly, this should return what you ask for.

Answer 3

Here's how I did it, with a bit of throwing-hands-up-in-the-air at the end.
Step one, identify the sets of items to be compared. This is:
For a “primary” value, all values found in Alias
Including the “primary” value as well (to cover that nasa/nostromo case)
A set in a table (A or B) is identified by its primary value. What really makes it hard is that the primary value is not shared across the two tables (sally#mars, sally#nasa). So we can compare sets, but we have to be able to “go back” to the primary on each table separately (e.g. the stand-out from table B may be sally#nasa / ripley#nostroomo, but we have to add sally#mars / ripley#nostromo to table A)
Major problems arise if, in a table, a primary value appears as an alias for a different primary value (e.g. in table A, chris#work appears as an alias for bob#test). For the sake of sanity, I am going to assume this will not happen… but if it does, the problem becomes even harder.
This query works to add missing items in B that are not in A, where the PrimaryAddress is the same for both A and B:
;WITH setA (SetId, FullSet)
as (-- Complete sets in A
select PrimaryAddress, AliasAdress
from A
union select PrimaryAddress, PrimaryAddress
from A
)
,setB (SetId, FullSet)
as (-- Complete sets in B
select PrimaryAddress, AliasAdress
from B
union select PrimaryAddress, PrimaryAddress
from B
)
,NotInB (Missing)
as (-- What's in A that's not in B
select FullSet
from setA
except select FullSet -- This is the secret sauce. Definitely worth your time to read up on how EXCEPT works.
from setB
)
-- Take the missing values plus their primaries from A and load them into B
INSERT B (PrimaryAddress, AliasAdress)
select A.PrimaryAddress, nB.Missing
from NotInB nB
inner join A
on A.AliasAdress = nb.Missing
Run it again with the tables reversed (from “NotInB” on) to do the same for A.
HOWEVER
Doing so with your sample data for "in B not in A" will add (sally#nasa, ripley#nostromo) to A, and as that’s a different primary, it’d create a new set, and so does not solve the problem. It gets ugly quickly. Talking it out from here:
Takes two passes, one for A not in B, one for B not in A
For each pass, have to do two checks
First check is what’s above: what’s in A not in B where primary addresses match, and add it
Second check is ugly: what’s in A not in B where the primary addresses from A is NOT a primary address in B and, thus, must be an alias. Here, find A’s primary address in B’s alias list, get the primary key used for this set in B, and create the row(s) in B using those values.

Answer 4

OK, This is how we did it... As it was becoming a pain, we ran a procedure that added the primary address of each entry as an alias: xx#xx -> xx#xx so that all addresses were listed as aliases for each user. This is similar to what #Phillip Kelly did above. Then we ran the following code: (its messy but it works; in one pass too)
SELECT 'Missing from B:' as Reason, TableA.[primary] as APrimary, TableA.[alias] as AAlias, TableB.[primary] as BPrimary,TableB.[alias] as BAlias into #A FROM dbo.TableA LEFT OUTER JOIN TableB ON TableB.alias = TableA.alias
SELECT 'Missing from A:' as Reason,TableA.[primary] as APrimary, TableA.[alias] as AAlias, TableB.[primary] as BPrimary,TableB.[alias] as BAlias into #B FROM dbo.TableB LEFT OUTER JOIN TableA ON TableA.alias = TableB.alias
select * from #A
select * from #B
UPDATE #A
SET #A.APrimary = #B.BPrimary
FROM #B INNER JOIN #A ON #A.APrimary = #B.BPrimary
WHERE #A.BPrimary IS NULL
UPDATE #B
SET #B.BPrimary = #A.APrimary
FROM #B INNER JOIN #A ON #B.BPrimary = #A.BPrimary
WHERE #B.APrimary IS NULL
select * from #A
select * from #B
select * into #result from (
select Reason, BPrimary as [primary], BAlias as [alias] from #B where APrimary IS NULL
union
select Reason, APrimary as [primary], AAlias as [alias] from #A where BPrimary IS NULL
) as tmp
select * from #result
drop table #A
drop table #B
drop table #result
GO