OK, finally getting time to have another look at this... https://issues.apache.org/bugzilla/show_bug.cgi?id=52191 summarises the situation - basically the SHM filename used now is not unique enough when there are multiple ISAPI Redirectors on a single website. My current proposal is to use the extension_uri to distinguish between the configurations (on the assumption that this will be the same on every site in a web server farm).
cheers tim On Thu, Aug 18, 2011 at 8:08 PM, Tim Whittington <t...@apache.org> wrote: > Yeah - not sure what I was smoking on that one. > Have reverted for another think. > > The basic problem we have is that the shared memory code assumes a > single worker configuration, and when you have multiple ISAPI > Redirectors on a single IIS with different configs that goes a bit > wonky. For a start, the SHM size is calculated by the first redirector > to start (and so may be too small for the other one), and then there's > the issue of the workers in the SHM not aligning. > > A possible solution might be a shm_config_name property that > distinguishes one redirector config from another (that should be > consistent across all processes). > The redirector DLL path could be used instead if that was reliable - I > don't know enough about IIS clusters to know if they span multiple > machines. > > If anyone's got any other ideas I'd be interested. > > cheers > tim > > On Thu, Aug 18, 2011 at 4:40 PM, Mladen Turk <mt...@apache.org> wrote: >> On 08/18/2011 03:54 AM, t...@apache.org wrote: >>> >>> Author: timw >>> Date: Thu Aug 18 01:54:31 2011 >>> New Revision: 1158991 >>> >>> URL: http://svn.apache.org/viewvc?rev=1158991&view=rev >>> Log: >>> Use the DLL handle to make the shared memory file name used by the ISAPI >>> Redirector unique for each DLL - the redirector supports multiple instances >>> per website, and without this multiple redirectors could access the same >>> shared memory file, corrupting the contents (evident when LB workers are >>> used on IIS 7). >>> >> >> >> Hmm, isn't that actually disabling the shared memory purpose? >> There is no point to have shared memory if it can't be shared >> across the processes. >> >> Your change makes no sense to me, cause now the shared memory >> is unique to each worker process which can be easily solved by >> not using the shared memory at all. >> >> IMHO if IIS7 corrupts the shared memory, we should find a way >> to better synchronize the access to it from multiple >> processes instead just making it unique to each process. >> >> >> Regards >> -- >> ^TM >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org >> For additional commands, e-mail: dev-h...@tomcat.apache.org >> >> > --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
