Note that in-memory cache is not shared between processes and/or threads, so if we have 1000 PHP processes we will ask for WSDL files at least 1000 times and also after each process restart. Also memory usage will very inefficient in this case. I think it is better to provide a chose to user use both, memory only, disk only cahce or don't use cache at all, through configuration directive. I am not agree to delete disk cache, especially because in-memory cache wasn't tested enough. And we need to disable in-memory cache by default for this release because of the same reason. Thanks. Dmitry. > -----Original Message----- > From: Rasmus Lerdorf [mailto:rasmus@lerdorf.com] > Sent: Tuesday, April 18, 2006 8:58 PM > To: Andrei Zmievski > Cc: Dmitry Stogov; 'Uwe Schindler'; soap@lists.php.net > Subject: Re: Frequent crashes in multithreaded > webserver since new WSDLcaching code in ext/soap > > > Especially since the speedup of the disk-based cache vs. > reparsing the > wsdl is minimal. It really is quite slow. > > -Rasmus > > Andrei Zmievski wrote: > > I agree that TTL for in-memory cache is a good idea. I also > think that > > in-memory cache should be the only type of cache we support > (why have > > two types of caches really). > > > > -Andrei > > > > On Apr 17, 2006, at 10:48 PM, Dmitry Stogov wrote: > > > >> Hi Andrei, > >> > >>> -----Original Message----- > >>> From: Andrei Zmievski [mailto:andrei@gravitonic.com] > >>> Sent: Tuesday, April 18, 2006 1:36 AM > >>> To: Dmitry Stogov > >>> Cc: 'Uwe Schindler'; 'Rasmus Lerdorf'; soap@lists.php.net > >>> Subject: Re: Frequent crashes in multithreaded webserver > >>> since new WSDLcaching code in ext/soap > >>> > >>> > >>>> Andrei, > >>>> > >>>> I was agree with your patch only after 5.1.3 release. > >>>> And I disagree with persistent WSDL by default at all. > Especially > >>>> without protection against huge memory usage. > >>> > >>> How huge is it really? We can put TTL on in-memory cache as well, > >>> but I think that we should have only in-memory cache and > not bother > >>> with the > >>> disk one anymore -- it's just too slow. > >> > >> Yes, we need some mechanism for cache expiration and > invalidation. I > >> see two big problems: > >> > >> 1) Now it is necessury to change PHP code to disable WSDL in-memory > >> cache to > >> reload changed WSDL. How to develop WSDL? > >> > >> 2) WEB hoster may have 1000 users, and how many memory > will take 1000 > >> cached > >> WSDL files? > >> > >> 3) On slow lines on-disk WSDL cache makes great speedup over direct > >> request. > >> Of course in-memory cache should operate faster. > >> > >> I suggest add some configuration directive to enable/disable > >> in-memory cache. Or change existing ones to have disk/memory/off > >> values. > >> > >>>> I hadn't time to look into you patch careful, and seems > it wasn't > >>>> tested enough. I looked into it today, found and fixed > another bugs > >>>> in addition to > >>>> your fix. I cannot be sure that these are latest bugs > >>> related to this > >>>> patch. > >>> > >>> I know you fixed those bugs. There is at least one more > remaining, > >>> according to the bug #37083. I am fine with reverting the > in-memory > >>> cache (or at least disabling the cache_wsdl) parameter > until after > >>> 5.1.3. > >> > >> I think we can stay code in CVS but disable it by default. (Invert > >> the behavior of "cache_wsdl" parameter.) Or we can use > default value > >> frim configuration directive. What do you think? > >> > >>>> Also for what reason you disabled disk cache? I re-enabled it. > >>> > >>> Because it's too slow? > >> > >> I prefer not to break working code, before making better one. > >> > >>> Do you have any ideas about the problem described in the last few > >>> comments in the bug? > >> > >> I'll look. > >> > >> Thanks. Dmitry. > >> > >>> -Andrei > >>> > >>> -- > >>> PHP Soap Mailing List (http://www.php.net/) > >>> To unsubscribe, visit: http://www.php.net/unsub.php > > -- PHP Soap Mailing List (http://www.php.net/) To unsubscribe, visit: http://www.php.net/unsub.php
