Question

I got a table with 2 columns (both INT), and there are 400 000 records (a lot).
The first column is random numbers ordered ASC. The second column has a rule on it (which is not important right now)
In the table there are 1000 records, that are exceptions. So, instead of the “rule”, there is only “-1” – valued cells.

How can I delete ~399 000 records, so i want to have in my table left only the ones with -1 and their “neighbors” (the records before and after the ones with -1)

UPDATE
sql server 2k5
first column values – yes unique, but not ID-s (it’s not ++ :D)

example:

before:

 20022518   13
 20022882   364
 20022885   -1
 20022887   5
 20022905   18
 20023200   295
 20023412   212
 20023696   284
 20024112   416
 20025015   903
 20025400   385
 20025401   -1
 20025683   283
 20025981   298
 20025989   8
 20026752   763
 20027779   1027
 20028344   565
 20028350   6
 20028896   546
 20028921   25
 20028924   -1
 20028998   77
 20029031   33
 20029051   20
 20029492   441
 20029530   38
 20029890   360

after:

 20022882   364
 20022885   -1
 20022887   5
 20025400   385
 20025401   -1
 20025683   283
 20028921   25
 20028924   -1
 20028998   77

Answer 1

If I understand correctly you want to keep all records with col2 = -1 and the records with the closest col1 to the records with -1. Assuming no duplicates in col1 I would do something like this

delete from table where not col1 in 
(
    (select col1 from table where col2 = -1)
union
    (select (select max(t2.col1) from table t2 where t2.col1 < t1.col1) from table t1 where t1.col2 = -1)
union
    (select (select min(t4.col1) from table t4 where t4.col1 > t3.col1) from table t3 where t3.col2 = -1)
)

Edit:
t4.col1 < t3.col1 should be t4.col1 > t3.col1
I created a test-table with col1 and col2, both int, col1 is PK, but not autonumber

SELECT * from adjacent

Gives

col1    col2
1   5
3   4
4   2
7   -1
11  8
12  2

With the above subselects:

SELECT * from adjacent
where
col1 in 
(
    (select col1 from adjacent where col2 = -1)
union
    (select (select max(t2.col1) from adjacent t2 where t2.col1 < t1.col1) from adjacent t1 where t1.col2 = -1)
union
    (select (select min(t4.col1) from adjacent t4 where t4.col1 > t3.col1) from adjacent t3 where t3.col2 = -1)
)

gives

col1    col2
4   2
7   -1
11  8

With the not also

SELECT * from adjacent
where
col1 not in 
(
    (select col1 from adjacent where col2 = -1)
union
    (select (select max(t2.col1) from adjacent t2 where t2.col1 < t1.col1) from adjacent t1 where t1.col2 = -1)
union
    (select (select min(t4.col1) from adjacent t4 where t4.col1 > t3.col1) from adjacent t3 where t3.col2 = -1)
)

gives

col1    col2
1   5
3   4
12  2

Finally a delete and select

delete from adjacent
where
col1 not in 
(
    (select col1 from adjacent where col2 = -1)
union
    (select (select max(t2.col1) from adjacent t2 where t2.col1 < t1.col1) from adjacent t1 where t1.col2 = -1)
union
    (select (select min(t4.col1) from adjacent t4 where t4.col1 > t3.col1) from adjacent t3 where t3.col2 = -1)
)

select * from adjacent

gives

col1    col2
4   2
7   -1
11  8