Domain index query takes 12 hours to execute

Question

Hi Friends,
My query which uses domain index takes 12 hours to execute. Can you please help me tuning this query?
select /*+ NO_UNNEST ORDERED index_ffs(Term idx_recanon_term_ysm1) parallel_index(Term, idx_recanon_term_ysm1, 8) */ term.rowid
from cmpgn.recanon_search_terms,cmpgn.recanon_term_ysm Term
where cmpgn.recanon_search_terms.search_type=3 and
contains(Term.RAW_TERM_TEXT,cmpgn.recanon_search_terms.search_text) > 0 and
Term.pod_id=11
Thanks in advance.
Regards
Bala

Answer 1

First your driving table is recanon_search_terms to get the required search terms, then use them to query Oracle Text. What you are trying to do is to get a subset of table term first and than run every individual row of table term against the search terms. This approach will take a very long time.
Not sure what you want to do, but it looks something I have been working on previously. As far I can see there is a table with controlled terms which are matched against other raw terms in a text document using Oracle Text. The issue is that you do a join in the contains clause without knowing the number of query expressions formed. It can be that 7 thousand individual queries run at once, than use indexes of the other predicates, do some sorting, etc. which would explain the long time needed. It will probably run out of memory causing all sorts of issues.
As a quick fix try first following statement without hints.
Tell us how many rows you get, the distinct counts for pod_id and search_type and how long it takes.
CREATE TABLE test_table AS
select term.pod_id pod_id, cmpgn.recanon_search_terms.search_type search_type, term_primary_key, ....
from cmpgn.recanon_search_terms,cmpgn.recanon_term_ysm Term
where contains(Term.RAW_TERM_TEXT,cmpgn.recanon_search_terms.search_text) > 0
Create index test_idx on test_table(pod_id, search_type) and use test_table instead to get the results, just by providing pod_id and search_type without the join in the contains clause.
SELECT ..
FROM test_table
WHERE pod_id = X
AND search_type = Y
Maybe this approach is sufficient for your purpose. For sure, it will give you instant results. In that case a materialized view instead of the table could work for maintenance reasons; I had some issues with materialized views for above scenario.
However check very carefully the results. I would have some doubts that all rows in search_text form a valid query expression for Oracle Text. If search_text has just single tokens or phrases wrapping curly brackets around will probably resolve the issue.
Think about to form one query expression through a function call instead of a table join inside the contains clause. Sometimes to run a set of individual queries are faster than one big query.
select term.rowid --, form_query_exrpession(3) query_exrpession
from cmpgn.recanon_term_ysm Term
where contains(Term.RAW_TERM_TEXT, form_query_exrpession(3)) > 0
The above function will form one valid Oracle Text query expression by using the table recanon_search_terms inside the function. This approach normally helps, at least in debugging and fine-tuning. Avoid using bind variables first in order to identify highly skewed distribution of search_type.
The other performance issue is the additional predicate of pod_id = X, here the suggestion from radord works very well. If you want to get your hands dirty have a look at user_datastore in the Oracle Text documentation, this will give you all the freedom you want.