Splitting strings when they are CLOBs

Like many of you, I have enjoyed using SQL to make data processing go hundreds of times faster. My new record is: 8000 times faster :-)

In a recent ODC thread, the question was how to split out comma separated values from two CLOBs, one containing “item ids” and the other “item keys”. The ids and the keys had to be matched by position: the first id with the first key, the second id with the second key, and so on. Then each id/key pair was to be inserted as a row into a target table.

With 50,000 ids of 7 characters and 50,000 keys of 172 characters, the current processing takes 2 hours.

Test data

First I’ll create a table with the two CLOBs.

--drop table t purge;
create table t as
      xmlelement(E, ',' || to_char(level,'fm0000009'))
      order by level
    ) AS CLOB
  ), 2
) i_item_id,
      xmlelement(E, ',' || rpad(to_char(level,'fm0000009'),172,'*'))
      order by level
    ) AS CLOB
  ), 2
) i_item_key
from dual
connect by level <= 50000;

Splitting CLOB strings

Instead of seeking a “perfect” custom-made solution, I just put together two functions I have already written about:

  1. The PIPE_CLOB function: a pipelined table function that reads a CLOB and outputs up to 4000 bytes at a time, respecting a delimiter.
  2. The STRING_TOKENIZE function: a pipelined table function that splits a VARCHAR2 string according to a delimiter.

All I have to do is feed the output from the first function into the second, and voilà!

select b.column_value item_id
from t,
table(pipe_clob( i_item_id, 4000, ',')) a,
table(string_tokenize(a.column_value,',')) b;

Combining the two CLOBs

Remember, the requirement was to combine the values from both CLOBs by position, in order to insert them into a target table. Since the functions return the values in order, I just use ROWNUM to number the values and use that number to join them.

--drop table u purge;
create table u as
with item_ids as (
  select rownum rn,
    b.column_value item_id
    from t,
  table(pipe_clob( i_item_id, 4000, ',')) a,
  table(string_tokenize(a.column_value,',')) b
, item_keys as (
  select rownum rn,
    b.column_value item_key
    from t,
  table(pipe_clob( i_item_key, 4000, ',')) a,
  table(string_tokenize(a.column_value,',')) b
select item_id, item_key 
from item_ids join item_keys using(rn);

Table U created.

Elapsed: 00:00:00.879

That’s right, processing time went from two hours to less than one second.

Now, I suspect that the questioner’s code is doing more than just a simple insert, but until he or she provides more details about the requirement, I get to maintain my claim:

8000 times faster!

(which I hope applies only to my code, not to the speed at which time flies…)


3 thoughts on “Splitting strings when they are CLOBs

  1. That’s a really good result. I wonder what the resources used for this look like, and how much slower this would be with a bulk insert rather than a CTAS.

    Now interested to check out the details of the function implementations – so link following and more reading for me.

    Thanks for sharing.

  2. Do you need the string_tokenize function? After a quick glimpse at the pipe_clob function I was under the impression that that function pipes string records, which are delimited by the same separator you use for string_tokenize.

  3. Hi Anton,

    The PIPE_CLOB function returns VARCHAR2 chunks that are meant to be as large as possible. Each VARCHAR2 can contain the separator multiple times. Think of PIPE_CLOB as analogous to a “fetch size” or a BULK COLLECT with LIMIT, except that the limit is a number of bytes, not a number of values.

    STRING_TOKENIZE takes each large VARCHAR2 and breaks out the individual delimited values.

    PIPE_CLOB takes a bite out of the CLOB and STRING_TOKENIZE chews it.

    Best regards,

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s