Lucene QueryParser in multiple threads: synchronize or construct new each time?

I have a web application where users submit queries to a Lucene开发者_StackOverflow index. The queries are parsed by a Lucene QueryParser. I learned the hard way that QueryParser is not thread-safe.

Is it better to use a single QueryParser instance, and synchronize on calls to its parse() method? Or is it better to construct a new instance for each query? (Or would I be better served by a pool of QueryParsers?)

I know that in general questions like this depend on the particulars and require profiling, but maybe someone out there can say definitively "QueryParsers are extremely inexpensive/expensive to construct"?

Create a new one each time. These are lightweight objects and the JVM handles object creation and garbage collection very well. Definitely do not use an object pool.