Hello, Long time listener, first time caller here - thanks for an excellent framework. We've been profiling our Hibernate 3.6.10-based app and noticed a perf bottleneck in javassist.util.proxy.RuntimeSupport.find2methods. Unfortunately, this method, which has a synch. block, is being called on every invocation of every proxied entity method (see javassist.util.proxy.ProxyFactory.makeForwarder(), called indirectly by ProxyFactory.createClass()). In our testing, the result is that our service call's latency increases from 33 to 55, 260, 400ms as concurrency increases 1-10-20-30 users on a 4-core CPU. At 20 and 30 users 51% of CPU time is spent contending for a monitor in RuntimeSupport.find2methods. Since find2methods merely interrogates class metadata, seems like its return values should be cached (in a ConcurrentMap?). Since this is a big problem for us, I am happy to submit a patch to javassist, but would like to know which version should I be patching, given that I am primarily interesting in using javassist in conjunction with Hibernate? Currently, hibernate-core 4.1.1 uses javassist 3.15.0-GA, 3.10.Final uses 3.12. Latest GA is 3.16-1. I'd appreciate any other advice as well (perhaps this problem's been discussed?) Thank you, -nikita _______________________________________________ hibernate-dev mailing list hibernate-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev

Filed here: https://issues.jboss.org/browse/JASSIST-163 From: Shaozhuang Liu [mailto:shaozhuang.liu@gmail.com] On Behalf Of Strong Liu Sent: Thursday, March 29, 2012 7:37 PM To: Nikita Tovstoles Cc: hibernate-dev(a)lists.jboss.org Subject: Re: [hibernate-dev] patching javassist - which ver? Hi Nikita, thanks for reporting this, I'd interested in all performance improvement :D I didn't notice this, but not sure if javassist team already aware of this. wrt the javassist version, I think that's javassist team's call to decide which version should be patched, but I guess it mostly should be the latest one. we, hibernate, shouldn't have any problem to upgrade to the latest javassist. and please let me know the javassist jira when you open it, I will keep an eye on it, thanks ------------------------- Best Regards, Strong Liu<stliu at hibernate.org<http://hibernate.org/> > http://about.me/stliu/bio On Mar 30, 2012, at 10:19 AM, Nikita Tovstoles wrote: Hello, Long time listener, first time caller here - thanks for an excellent framework. We've been profiling our Hibernate 3.6.10-based app and noticed a perf bottleneck in javassist.util.proxy.RuntimeSupport.find2methods. Unfortunately, this method, which has a synch. block, is being called on every invocation of every proxied entity method (see javassist.util.proxy.ProxyFactory.makeForwarder(), called indirectly by ProxyFactory.createClass()). In our testing, the result is that our service call's latency increases from 33 to 55, 260, 400ms as concurrency increases 1-10-20-30 users on a 4-core CPU. At 20 and 30 users 51% of CPU time is spent contending for a monitor in RuntimeSupport.find2methods. Since find2methods merely interrogates class metadata, seems like its return values should be cached (in a ConcurrentMap?). Since this is a big problem for us, I am happy to submit a patch to javassist, but would like to know which version should I be patching, given that I am primarily interesting in using javassist in conjunction with Hibernate? Currently, hibernate-core 4.1.1 uses javassist 3.15.0-GA, 3.10.Final uses 3.12. Latest GA is 3.16-1. I'd appreciate any other advice as well (perhaps this problem's been discussed?) Thank you, -nikita _______________________________________________ hibernate-dev mailing list hibernate-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev _______________________________________________ hibernate-dev mailing list hibernate-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev

What are some of the alt libraries you are considering? From my POV, perhaps selfishly, this is a pretty substantial scalability issue. I realize that cglib wasn't being actively maintained, but we didn't see any significant problems with that lib. AFAIK there is still a way to force using cglib for hibernate proxy gen (at least in 3.6.x) - off top of your head do you know of show-stoppers in switching back to cglib, at least temporarily? Thanks -nikita -----Original Message----- From: Steve Ebersole [mailto:steven.ebersole@gmail.com] On Behalf Of Steve Ebersole Sent: Friday, March 30, 2012 7:00 AM To: Nikita Tovstoles Cc: hibernate-dev(a)lists.jboss.org; 'Strong Liu' Subject: Re: [hibernate-dev] patching javassist - which ver? I have not had time to dig into this, so I cannot speak to the specific concern brought up here. I will say, however, that in some cases we are not using Javassist "right" (we should not be using its ProxyFactory class). And there is question whether we should be using Javassist at all for proxy generation. Alot of JBoss projects actually ended up moving to some new libraries specifically for proxy building. There are already issues in our JIRA for both of those things I mentioned. On Fri 30 Mar 2012 08:47:02 AM CDT, Nikita Tovstoles wrote: > Filed here: > > https://issues.jboss.org/browse/JASSIST-163 > > > > > > > > From: Shaozhuang Liu [mailto:shaozhuang.liu@gmail.com] On Behalf Of Strong > Liu > Sent: Thursday, March 29, 2012 7:37 PM > To: Nikita Tovstoles > Cc: hibernate-dev(a)lists.jboss.org > Subject: Re: [hibernate-dev] patching javassist - which ver? > > > > Hi Nikita, > > > > thanks for reporting this, I'd interested in all performance improvement :D > > > > I didn't notice this, but not sure if javassist team already aware of this. > > > > wrt the javassist version, I think that's javassist team's call to decide > which version should be patched, but I guess it mostly should be the latest > one. > > we, hibernate, shouldn't have any problem to upgrade to the latest > javassist. > > > > and please let me know the javassist jira when you open it, I will keep an > eye on it, thanks > > ------------------------- > Best Regards, > > Strong Liu<stliu at hibernate.org<http://hibernate.org/> > > > http://about.me/stliu/bio > > > > On Mar 30, 2012, at 10:19 AM, Nikita Tovstoles wrote: > > > > > > Hello, > > > > Long time listener, first time caller here - thanks for an excellent > framework. > > > > We've been profiling our Hibernate 3.6.10-based app and noticed a perf > bottleneck in javassist.util.proxy.RuntimeSupport.find2methods. > Unfortunately, this method, which has a synch. block, is being called on > every invocation of every proxied entity method (see > javassist.util.proxy.ProxyFactory.makeForwarder(), called indirectly by > ProxyFactory.createClass()). In our testing, the result is that our service > call's latency increases from 33 to 55, 260, 400ms as concurrency increases > 1-10-20-30 users on a 4-core CPU. At 20 and 30 users 51% of CPU time is > spent contending for a monitor in RuntimeSupport.find2methods. > > > > Since find2methods merely interrogates class metadata, seems like its return > values should be cached (in a ConcurrentMap?). Since this is a big problem > for us, I am happy to submit a patch to javassist, but would like to know > which version should I be patching, given that I am primarily interesting in > using javassist in conjunction with Hibernate? Currently, hibernate-core > 4.1.1 uses javassist 3.15.0-GA, 3.10.Final uses 3.12. Latest GA is 3.16-1. > I'd appreciate any other advice as well (perhaps this problem's been > discussed?) > > > > Thank you, > > -nikita > > _______________________________________________ > hibernate-dev mailing list > hibernate-dev(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/hibernate-dev > > > > _______________________________________________ > hibernate-dev mailing list > hibernate-dev(a)lists.jboss.org > https://lists.jboss.org/mailman/listinfo/hibernate-dev -- steve(a)hibernate.org http://hibernate.org