Fastest way of updating a row

[Asger Jensen]

In relation to my last post, I have a question for the SQL-gurus.

I need to update 70k records, and mark all those updated in a special
column for further processing by another system.

So, if the record was

Key1, foo, foo, ""

it needs to become

Key1, fap, fap, "U"

iff and only iff the datavalues are actually different (as above, foo
becomes fap),

otherwise it must become

Key1, foo,foo, ""
Is it quicker to :
1) get the row of the destination table, inspect all values
programatically, and determine IF an update query is needed

OR

2) just do a update on all rows, but adding
and (field1 <> value1 or field2<>value2) to the update query

that is
update myTable
set
field1 = "foo"
markField="u"
where key="mykey" and (field1 <> foo)

The first one will not generate new update queries if the record has
not changed, on account of doing a select, whereas the second version
always runs an update, but some of them will not affect any lines.

Will I need a full index on the second version?

Thanks in advance,
Asger Henriksen

[Erland Sommarskog]

[posted and mailed, vänligen svara i nys]

Asger Jensen (ak*@tmnet.dk) writes:

Is it quicker to :
1) get the row of the destination table, inspect all values
programatically, and determine IF an update query is needed

OR

2) just do a update on all rows, but adding
and (field1 <> value1 or field2<>value2) to the update query

that is
update myTable
set
field1 = "foo"
markField="u"
where key="mykey" and (field1 <> foo)

I'm not sure that I follow, but it sounds to me that the in first
approach you would retrieve rows one by one.

In any case, the second approach leaves all the jub to the computer,
and there is a reason why we have computers, isn't there? :-)

The only catch is that with too many rows in the table there can be
a strain on the transaction log. But with only 70000 rows, this is
not worth worrying about.

Obviously there query will run faster if there is a clustered index
on the column "key".

--
Erland Sommarskog, SQL Server MVP, so****@algonet.se

Books Online for SQL Server SP3 at
http://www.microsoft.com/sql/techinf...2000/books.asp

[Asger Jensen]

> I'm not sure that I follow, but it sounds to me that the in first

approach you would retrieve rows one by one.
Yes, thats what it was. Nasty.
In any case, the second approach leaves all the jub to the computer,
and there is a reason why we have computers, isn't there? :-)
:-), yes, only in my scenario it took soooo long.
The only catch is that with too many rows in the table there can be
a strain on the transaction log. But with only 70000 rows, this is
not worth worrying about.

ok,
Obviously there query will run faster if there is a clustered index
on the column "key".

There was, actually, but during the 70000 records it became slower and
slower. I solved it by adding a dynamically generated index on ALL
fields, not just "key", this sped things up considerably, so I went
from 1 hour to 10 minutes.

I guess it is due to the server needing to do a lookup/record on all
value fields to see if they have changed, and this can be done more
efficiently with a n index.

Regards
Asger