[rules-users] Re: Help with MatchesEvaluatorsDefinition that caches compiled regular expression patterns
Edson Tirelli <tirelli <at> post.com> writes:
Jared, Your idea is actually interesting. Although, we would need to take
care
of a few edge scenarios:* The actual evaluator instance is supposed to be a
"fly weight" implementation, and as so, it might be used by multiple threads.
So, first thing, the cache map implementation must be thread safe. * We had
problems in the past when using static caches like that in JEE/Spring/OSGi
environments where you have multiple modules being loaded by different
classloaders etc. So, I think best would be to have a non-static map for each:
the StringMatchesEvaluator and the StringNotMatchesEvaluator inner classes.* We
need a way to clean up the cache for no-longer used expressions. Easiest way
would probably be to use an LRU map as cache. The size of the cache would
probably have to be a configuration on the KnowledgeBaseConfiguration, with a
given default value. This would avoid memory leaks. Wanna try providing a
patch for that? Thanks, Edson
2009/5/4 Jared Davis <sunray <at>
davisprogramming.com>I need some help placing a cache of compiled regular
expressions using the
MatchesEvaluatorsDefinition as a starting point.
Just for a test I placed a static map in the class:
static Map<String,Pattern> uglyHack = new HashMap<String,Pattern>();
Then in each of the evaluators I have different versions of:
Pattern p = uglyHack.get(value2);
if (p == null ) {
p = Pattern.compile(value2);
uglyHack.put(value2, p);
}
Matcher m = p.matcher(value1);
return m.matches();
instead of
return value1.matches( value2 );
The time elapsed to inserting objects into the workspace has been cut in half.
.4 seconds -> .2
Has someone already done this "optimization?" Where should the cache be
stored?
_______________________________________________
<snip>
Edson,
One persons memory leak is another's optimization.<g>
I'll give it a shot. Is there a drools preferred thread safe LRUCache
implementation? Seems like hours/days could be spent on the cache algorithm.
Jared