Re: [GENERAL] Inconsistent performance with LIKE and bind variable onlong-lived connection

I walked through the Java code in a debugger just now and have an explanation for the 5 vs. 9 discrepancy. The JDBC driver keeps a cache of queries that have been passed to a connection's prepareStatement() method, and inlines the bind values the first 4 times it sees a query in the hopes of reducing overhead on one-off queries. So I believe the sequence ends up being:

1-4: JDBC driver inlines the values, server sees no bind variables

5: JDBC driver prepares the statement, server sees bind variables and tries generic plan

6+: JDBC driver reuses the existing prepared statement from iteration 5

10: Server has seen the query 5 times before and switches to the custom plan

As for the broader problem, at the risk of being hopelessly naive about how all this works internally: Could the discrepancy between the estimated and actual row counts be tracked and fed back into the planner somehow?

-Steve