On Tue, 16 Oct 2001, Stas Bekman wrote: > it all works now, but there is a new problem. I remember reading in one of > the docs that in 2.0 the interpreters queue is implemented as LIFO.
the docs don't say anything lifo or fifo. interpreters are pulled from and putback onto the head of the idle list. i don't think that falls under the definition of either 'lifo' or 'fifo'. > Therefore in my tests I've relied on the fact that two subsequent requests > will hit the same interpreter, assuming that normally during 'make test' > there is only one request at any given time. But that's not what I see, if > you apply this patch that sets MaxClients to 2 and run: > > ./t/TEST -v closure > > in ./ModPerl-Registry/ > > it fails, because the server does round robin and not LIFO rotation. I > need the MaxClient==2 setting for a different test (not committed yet). well yeah, with maxclients > 1 you have > 1 process, interpreters are not shared between processes. i don't think you can rely on what "order" apache will hand requests to children. i'm not aware of any difference between how this is managed from 1.x, but if it has changed i would expect it to be different depending on mpm/platform. > - is it a bug that I observe RR and not LIFO? no, you are not seeing a bug. > - if it's not a bug, any idea how to make sure that I work with the same > interpreter to be able to perform the tests? (without rerunning the > requests many times and waiting for a specific $$ I could catch on the > first request) it won't happen easily with maxclients > 1. i also tried to stay away from any tests that assume you're getting the same interpreter so things like t/TEST -times 2 and t/TEST -proxy will work. not that we shouldn't have tests that do, but they should be carefully written to also work with maxclients > 1 and t/TEST -times, etc. --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
