Thanks. I have done more heap analysis and think I have it tracked closer to the source.
-- I started looking at the heap a different way. The random values I looked at before (of the 80,000) may not have be as representative as I thought. Examining the retained sizes in the heap, I am finding: I have two instances of AbstractProtocol$ConnectionHandler. One of these AbstractProtocol$ConnectionHandler instances has a ConcurrentHashMap called "connections" This map as 32 elements in its "table". Most of these are null. Some of the ones that are not, are huge. The entirety of the map retains 112MG. Two of these ConcurrentHashMap$Node elements take up around 50MG a piece. Looking at the "val" variable of a node, there is an UpgradeProcessorInternal Inside this is a variable called internalHttpUpgradeHandler (of type Http2UpgradeHandler) The one of these I am looking at now retains 16MG of memory. (Oddly, once I get this far, the retained sizes of its internal objects don't really add up.) Any ideas on how to work around this? Or if this is already fixed in a later version of Tomcat? Thanks, Mark Claassen Senior Software Engineer Donnell Systems, Inc. 130 South Main Street Leighton Plaza Suite 375 South Bend, IN 46601 E-mail: mailto:mclaas...@ocie.net Voice: (574)232-3784 Fax: (574)232-4014 Disclaimer: The opinions provided herein do not necessarily state or reflect those of Donnell Systems, Inc.(DSI). DSI makes no warranty for and assumes no legal liability or responsibility for the posting. -----Original Message----- From: Rob Sargent <rsarg...@xmission.com> Sent: Thursday, July 8, 2021 6:50 PM To: users@tomcat.apache.org Subject: Re: [Possible Spam] Re: HTTP/2 Memory Leak Importance: Low On 7/8/21 3:17 PM, Mark A. Claassen wrote: > Ok. That didn’t seem to work. I will investigate further and try to find a > way to send that information. > > It is not that busy a server, but the memory use increases very quickly. > Doing a class_histogram shows MessageBytes growing by the thousands every 30 > minutes. > > (We have a temporary monitor script in place that does a GC and then prints a > class_histogram every half hour to help us pinpoint what is happening.) > > Thanks, > Mark > Perhaps you've done this already, but grep -R 'static HashMap' src might locate some potential culprits.