Memory leak in asp.net application - W3WP and gen 2 heap continues to grow until AppPool recycles

We have a large asp.net application that is leaking memory. Perfmon shows that this leak is in managed memory as W3WP private bytes grows at the same rate as bytes in all heaps. I can also see that Gen 2 garbage collections are running but the Gen 2 heap size continues to grow.

I took a memory dump and analysed in WinDbg and can see a very large number of objects of lots of types. Strings are the biggest type and 20% of the size of the strings is mad开发者_C百科e from 51 objects.

Dumping these large strings shows outputted html either from controls or entire pages. Running !gcroot on these shows the root objects being of type System.Text.RegularExpressions.Regex or System.Web.RegularExpressions.GTRegex.

Any ideas of what could be happening or how I can investigate further?

Thanks, Simon

How about using a memory profiler such as dotTrace Memory or ANTZ Memory Profiler? Both products are available as a time-limited trial version.

That strings are the most common type on your heap is not strange at all. If you for example have 10 HashSet's containing 1000 strings each, the dump will show that you have 10 HashSet's on your heap, but 100 000 strings. Many objects contain one or several strings. Thus, the number of strings shown in the dump is the sum of all strings from all objects on the heap, which tend to be a lot.

However, if you have alot of System.Text.RegularExpressions.Regex on your heap, that can very well be the root to your memory problemts. Regex in .NET tend to take a lot of resources. Hence, my advice is that you go through your code and try to find any excessive use of regex. Also, make sure that any references to Regex objects are being taken care of, that is to say, that the references to the Regex objects are not kept alive. That way, the Garbage Collector can make sure the Regex objects are deallocated properly.

Good luck!

In theory it should be quite dificult to cause a memory leak in asp.net without using unmanaged resources. If everything is single threaded then all references to managed resources should be free to be garbage collected when the page life cycle is complete. Are you firing off worker threads to do anything and are these threads continuing to live beyond the life of the page? Or do you have any long running processes exposed as web methods that can be fired off by asynchounsly and are just taking a long time to run and being called repeatedly until the memory is full?