topic Re: Proc sql update in SAS Procedures

Proc sql update

Bjs09f — Tue, 23 Feb 2016 23:33:51 GMT

Hi All,

I am trying to update a table in proc sql, to set our columns = to a temporary work tables values.

The below code is not working and giving me this error:

ERROR: Unresolved reference to table/correlation name transaction.

proc sql;

update table1

set col1 = (select col1 from table2 where table1.txn = txn)

where txn in (select txn from table2);

quit;

I am trying to avoid a datastep update statement, as it is recreating our table and seems to be a long time for a job that should be very simple.

Thanks in advance,

Ben

Re: Proc sql update

mohamed_zaki — Tue, 23 Feb 2016 23:57:26 GMT

Are you sure that the column txn exists in table1 and table2?

Also, what you trying to do exactly?

Re: Proc sql update

RW9 — Wed, 24 Feb 2016 09:23:08 GMT

Hi,

Well, firstly, I am not sure that an update statement like that would be quicker than a datastep. Your datastep could be taking a long time for a number of reasons - where is it located network/locally, how many observations/columns etc.

For your update, have you tried putting the Alias in the step:

proc sql;
  update table1 table1  /* Second one here is the alias */
  set col1 = (select col1 from table2 where table1.txn = txn)
  where txn in (select txn from table2);
quit;

However I am not sure your logic is right at all. The where clause will select multiple rows if they exist. Post example test data, and what you want the output to look like and we can suggest code.

Re: Proc sql update

LinusH — Wed, 24 Feb 2016 12:03:29 GMT

May I that in many cases SQL is quite inefficient when it comes to updates.
On an indexed data set, a data step with modify by can be very fast - it does not recreate the data set, it updates it in place.

Re: Proc sql update

Bjs09f — Wed, 24 Feb 2016 14:01:12 GMT

Thanks for reply,

Our current code is in a datastep. The table is indexed, and it does not maintain indexes if we perform like we do below

data table1;

update table1

table2

UPDATEMODE=NOMISSINGCHECK;

by transaction_id;

run;

So we are updating 2 columns on table1, which are both on table2. This step is recreating the table each time and it seems like an inefficient way to update a table.

Re: Proc sql update

Bjs09f — Wed, 24 Feb 2016 14:05:39 GMT

Here's some sample data of the two columns that should be updating, also the column they are updating by:txn

Table1 (Table to update)

txn flag date

1021912 X 12/02/2015

Table2(updating table one data with flag and date) flag is empty

txn flag date

1021912 2/24/2016

Note: Txn is a unique index on both tables.

Re: Proc sql update

LinusH — Wed, 24 Feb 2016 18:11:21 GMT

As stated, modify by will update in place which includes maintaining indexes.