SQL Server, JDBC and compiled query plans
I had a query like this
select ... from big_table where key in (select key from other_table where id between @p1 and @p2)
I had noticed it was performing badly, and the plan didn’t make any sense. The values for
typically return a small number (200-500) rows from the subquery, which in turn retrieves a small fraction of the overall
big_table. Yet the estimate was off by an order of magnitude, showing that the subquery was expected
to return about 1000x the number of rows it would in practice.
I noticed if I ran
DBCC FREEPROCCACHE or
sp_recompile, I could force the query to be recompiled with the plan
I expected. So I assumed the issue was parameter sniffing, and maybe somehow the query compiled originally with
values that triggered a different plan. But it turned out in this case, it was exactly the opposite–my problem
was that parameter sniffing was not happening.
I was able to show this with a test case to reproduce my problem within SSMS.
If I ran the query with
sp_executesql multiple times, with different parameter values, the parameters on the
first execution would be “sniffed” as I expected and subsequent executions would use the same plan with frozen estimates.
If I ran the query wuth
sp_preparesql instead, then the query plan is compiled with no known parameter values, similar
to what I get with
OPTIMIZE FOR UNKNOWN.
After I found this post, I
believe that the JDBC driver must be using
sp_preparesql under the hood, and there is likely a setting I could tweak to
resolve the issue and get the expected plan.