Why are subsequent queries so much slower? - jpa

Question

I'm trying to sort out a very weird behavior.
I'm working with:
JBoss AS 7.1.1
EJB 3.0
JPA
XA DataSource
Oracle 11g
In one of the systems fuctionalitites, the user can see the status of each Store. For each Store I fire a query, to sum up all the files that have been processed. The query is something like this:
SELECT
SUM(CASE file.type
WHEN 'TYPE_1' THEN 1
ELSE 0
END)
,
SUM(CASE file.type
WHEN 'TYPE_2' THEN 1
ELSE 0
END)
,
SUM(CASE file.type
WHEN 'TYPE_3' THEN 1
ELSE 0
END)
FROM
File file
WHERE
file.type IN ('TYPE_1', 'TYPE_2', 'TYPE_3')
AND file.status = 'RECEIVED'
AND file.store.id = :storeId
The thing is, the user can select which of the stores he wants to check, and that's where things get weird.
When I check the first store, the result comes blazing fast, but all subsequent queries take significantly more time. Let me exemplify:
User checks store 15 (Blazing fast result) - About 200 ms
User checks store 2 (Very slow result) - About 8000 ms
Now pay attention to this part, it's very important.
User logs out, and logs in again.
User checks store 2 (the one that took 8000ms), and now the result is blazing fast.
This is very odd, the same store that took a while before, is now loading pretty fast.
Whenever I try the queries on SQLDeveloper the results come pretty fast as well.
I annotated my EJB with #TransactionAttribute(TransactionAttributeType.NOT_SUPPORTED) but I didn't get any difference on the execution time.
I created a standalone project to run the queries using JDBC and the result was fast again, which leaves me thinking it may be some configuration on my DataSource, persistence.xml or anything like that.
Does anyone have any clues why this happens?

Answer 1

few things :
when the user checks store2 in the second time , the oracle optimizer is probably using it's "cache" , and therefore it is blazing fast.
how much File records does store2 has ? try to perform a group by sentence to see if this File table needs special statistics , if for example the store2 has dramaticly more File records than other stores then try to perform this method :
begin
dbms_stats.gather_table_stats(user,'file' , estimate_percent=>100);
end;
this will ensure the table's statistics are accurate.
you can optimize the query , you don't have to perform 3 times "sum" , you can do something like this :
select f.type , count(*)
from File f
where f.store.id = :storeId
and f.type IN ('TYPE_1', 'TYPE_2', 'TYPE_3')
group by f.type

Answer 2

you may run in a cardinality feedback problem; just look in this blog;
http://orcasoracle.squarespace.com/oracle-rdbms/2012/12/18/when-a-query-runs-slower-on-second-execution-a-possible-side.html
/KR