From: J. S. E. <ev...@cp...> - 2003-07-16 15:24:08
|
Lars, At 10:56 AM 7/15/03 +0200, Lar...@pp... wrote: >Scott, > >is your Notification load test independent from the rest of your own code? >Would it be possible to contribute the stress test to the openorb source >base? No, the tests are part of my app server test framework and cannot easily be separated. At one point I did take an existing notify example and modify it to make sure that the leaks were not in my code but the mods were real hacks and I don't believe I have them anymore. It didn't take too much effort to use the existing structured event examples to create stress tests but the level of effort for a comprehensive test suite might be significant. >I'm asking because I am more and more convinced that the Notification >Service needs some serious modifications (or even a rewrite?), but I don't >dare to even start before we have some tests to verify that I didn't break >more than I fixed. Prudent idea, IMHO. Regression tests are life-saver (translation time-saver) when dealing with something complex like notify. FWIW, I'd be willing to contribute test cases to a test suite if someone is willing to establish a framework for straightforward insertion of test cases. Scott >Lars > > > -----Original Message----- > > From: J. Scott Evans > > Sent: Friday, May 02, 2003 6:40 PM > > To: ope...@so... > > Cc: John Haynes; smi...@fe... > > Subject: Re: [openorb-devel] Another notification leak resolved. > > > > > > > > Richard, > > > > At 04:20 PM 5/2/03 +0100, Richard Clark wrote: > > >Scott, > > > > > >I've updated org.openorb.notify.Logger so that child loggers > > created will use > > >prefix logging so they shouldn't hang around. You can also > > use --debug NONE to > > >use null logging. > > > > > >You'll also need the CVS HEAD of the tools modules. > > > > Excellent, thanks for your efforts. I ran an 8 hour load > > test last night > > (using my local patches) that effectively hammered the > > notification service > > (4 independent client threads each using 25 channels having > > 25 pushers & 24 > > consumers per channel where each pusher pushes a > > StructuredEvent to 24 > > consumers - repeat pusher/consumer connects, event push, and > > disconnects as > > fast as possible for 8 hours non stop). The test passed and > > the memory > > usage fluctuated, sometimes by +- 10 MB, but showed no > > indication of any > > measurable leaks. I'll be running 72-120 hour load tests > > over the next > > week or so to further verify no leaks (at least in the > > classes used in the > > test). Thanks for quick responses and patches for these problems! > > > > Scott > > |