Andy, Brian pointed out today that you aren't just leaking SQL-related objects, you're leaking requests... because requests hold a reference to every traversed object (via PARENTS), if you're leaking reuqests, its a very bad thing. It would also explain why you see so many leaked SQL-related objects if the site is heavily SQL-driven. It may be that your code is keeping references to requests (maybe keeping a global with tracebacks around?), or there is some other circular reference problem. As we haven't had any other corroborative reports of either the DA leaking or the SQLMethods stuff leaking on their own, I'm sort of at a loss to provide a solution. Is there any way to make this bug reproducible on a clean Zope install? Can we boil it down to the simplest case? - C
A new bug entry was added with the following information:
Title: Memory Leak with ZSQL Methods
At: http://classic.zope.org:8080/Collector/Collector/2154/sview
Submitter: andym
Email: andym@activestate.com
Description: When running a large number of SQL queries using ZSQLMethods I get a memory large leak. Following is the information from debug_info after 12 hours of running the server:
Shared.DC.ZRDB.RDB.SQLAlias 62627 62642 +15 Record.Record 10527 10531 +4 Acquisition.ImplicitAcquirerWrapper 32441 32443 +2 ExtensionClass 5522 5524 +2 ZPublisher.HTTPRequest.HTTPRequest 3749 3750 +1 ZServer.HTTPResponse.ZServerHTTPResponse 3742 3743 +1 ZPublisher.BaseRequest.RequestContainer 3739 3740 +1 Acquisition.Acquirer 2695 2696 +1 Shared.DC.ZRDB.RDB.NoBrains 2630 2631 +1
I can reproduce this easily. As a wild guess something is not being correctly destroyed, no matter how many sweeps clears I do.
A current workaround is to restart zope every 6 hours or so. This is obviously sub optimal.