How can you force an function in a where clause to execute once in oracle?

问题

I am attempting to query a rather large IP/City (3,000,000+ rows) table based on an IP address. My source IP address is dotted notation such as 127.0.0.1 and the table has two fields stored as integers such as 2130706433. So I need to convert the dotted notation to the integer in my where clause like this;

select get_ip_integer('74.253.103.98') ,icb.*,icl.* 
from ip_city_block icb, ip_city_location icl 
where get_ip_integer('74.253.103.98') between  icb.startipnum and  icb.endipnum  and
      icl.locid = icb.locid;

This query takes over 4 ~(4.33) seconds on a reasonably fast DB. The following query takes .062 seconds;

select 1258121058,icb.*,icl.* 
from ip_city_block icb, ip_city_location icl 
where icb.startipnum <= 1258121058 and  icb.endipnum >= 1258121058 and
      icl.locid = icb.locid;

The only difference is I replace the function get_ip_integer with the value that the function returns. If all I were doing is one lookup I would execute the second query and be done with it, but I'm not.

I actually want to join to another table that contains many ip addresses in dotted format, and when I do it takes forever. For fun I also tried;

select ip_integer ,icb.*,icl.* 
from (select get_ip_integer('74.253.103.98') ip_integer from dual),ip_city_block icb, ip_city_location icl 
where icb.startipnum <= ip_integer and  icb.endipnum >= ip_integer and
      icl.locid = icb.locid;

and this took ~4.33 seconds as well.

So the question is how can I force the get_ip_integer function to execute only once and use the results for the comparison?

I updated my function to Deterministic, and it seemed to help on the original query, but the more complex query is still unusable, performance wise. Here it is;

SELECT COUNTRY, REGION,CITY, WEBLOG_USERID, WEBLOG_IP, WEBLOG_COUNT 
FROM (
  select WEBLOG_USERID,WEBLOG_IP,get_ip_integer(WEBLOG_IP) ip_integer,count(*) WEBLOG_COUNT 
  from weblog 
  where weblog_date > '20130217' 
  group by WEBLOG_USERID,weblog_ip
),ip_city_block icb, ip_city_location icl 
where ip_integer between icb.startipnum and icb.endipnum and icl.locid = icb.locid
ORDER BY 1,2,3;

Any thoughts on this?

After a little thought on my own I came up with this, while not blindingly fast it is acceptable;

SELECT COUNTRY, REGION,CITY, WEBLOG_USERID, WEBLOG_IP, WEBLOG_COUNT 
FROM (
  select WEBLOG_USERID,WEBLOG_IP, count(*) WEBLOG_COUNT 
  from weblog 
  where weblog_date > '20130000' 
  group by WEBLOG_USERID,weblog_ip
),ip_city_block icb, ip_city_location icl 
where get_ip_integer(WEBLOG_IP) between icb.startipnum and icb.endipnum and icl.locid = icb.locid
ORDER BY 1,2,3;

回答1:

Why are you using PL/SQL for this at all? From what you've said you're doing some maths, why not just do that in SQL? This'll also be possible with a combination of INSTR and SUBSTR but it's prettier to look at with REGEXP_SUBSTR.

select to_number(regexp_substr(ip, '[^.]+', 1, 1)) * power(2,24)
        + to_number(regexp_substr(ip, '[^.]+', 1, 2)) * power(2,16)
        + to_number(regexp_substr(ip, '[^.]+', 1, 3)) * power(2,8)
        + to_number(regexp_substr(ip, '[^.]+', 1, 4))
     , icb.*
     , icl.* 
  from ip_city_block icb
  join ip_city_location icl
    on icl.locid = icb.locid  
 where to_number(regexp_substr(ip, '[^.]+', 1, 1)) * power(2,24)
        + to_number(regexp_substr(ip, '[^.]+', 1, 2)) * power(2,16)
        + to_number(regexp_substr(ip, '[^.]+', 1, 3)) * power(2,8)
        + to_number(regexp_substr(ip, '[^.]+', 1, 4))
       between icb.startipnum and icb.endipnum

SQL Fiddle demonstration of REGEXP_SUBSTR output

If you have to do this in PL/SQL you should do two things:

See if you can declare your function as deterministic.
Try and take advantage of sub-query caching.

It appears as though you're already doing 2, but you could try and extend this by using a WITH clause:

with the_ip as ( select get_ip_integer('74.253.103.98') as ip from dual )
select the_ip.ip
     , icb.*
     , icl.* 
  from ip_city_block icb
  join ip_city_location icl
    on icl.locid = icb.locid
  join the_ip
    on the_ip.ip between icb.startipnum and icb.endipnum

回答2:

you were on the right track with dual, but for subquery caching, you do it in the select.

select (select get_ip_integer('74.253.103.98') from dual) ip,
       icb.*,icl.* 
from ip_city_block icb, ip_city_location icl 
where get_ip_integer('74.253.103.98') between  icb.startipnum and  icb.endipnum  and
      icl.locid = icb.locid;

also you should define your function with result_cache.

see here for more details: http://www.oracle.com/technetwork/issue-archive/2011/11-sep/o51asktom-453438.html

来源：https://stackoverflow.com/questions/14964562/how-can-you-force-an-function-in-a-where-clause-to-execute-once-in-oracle

标签

Oracle

oracle11g