Re: [EMAIL PROTECTED] Apache 2.0: Recommended Mutex types for Solaris 8,9,10
On Wed, Oct 05, 2005 at 03:36:45PM +0200, Axel-Stéphane SMORGRAV wrote: Joe, Thanks for replying to my post. Isn't the AcceptMutex only used inside the parent process to serialise the access to the accept(2) system call ? It's used in *children* to do that, yes. Is your comment about cross-process mutex relevant for SSLMutex rather than AcceptMutex , and would that mean that in your opinion although it might be OK to use pthread for AcceptMutex it should be avoided in the case of SSLMutex ? The same concerns apply to both equally. If the process crashes holding the SSLMutex then all processes which try to server SSL are liable to deadlock. Would you have any idea of the performance implications of using fcntl or flock over sysvsem or pthread ? No idea on Solaris, sorry, you'd have to benchmark this. Regards, joe - The official User-To-User support forum of the Apache HTTP Server Project. See URL:http://httpd.apache.org/userslist.html for more info. To unsubscribe, e-mail: [EMAIL PROTECTED] from the digest: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [EMAIL PROTECTED] Apache 2.0: Recommended Mutex types for Solaris 8,9,10
On Mon, Oct 03, 2005 at 11:17:38AM +0200, Axel-Stéphane SMORGRAV wrote: As far as I can tell, in Apache versions 2.0.49 and prior the default Mutex type (at least for AcceptMutex) was pthread. It then changed to fcntl around Apache 2.0.50 leading me to post the following http://issues.eu.apache.org/bugzilla/show_bug.cgi?id=32325 issue, and explicitly set both mutexes to pthread which appeared to solve the problem. Several people have since posted similar problems and solved the issue by explicitly setting the mutex types to something other than AcceptMutex default and SSLMutex default. Does anyone have any idea of the pros and cons of the different mutex types, provided they are available on any given platform? Is there any reason to use a different mutex implementation for AcceptMutex than you would for SSLMutex? What are the performance and operational issues with each mutex type? A pthread cross-process mutex is not released if the process holding it segfaults; this can lead to the entire server deadlocking particularly if using a threaded MPM. I don't think anybody knows what conditions will lead to the fcntl EDEADLK issue on Solaris, that is really a problem which needs more investigation. (historically, 1.3 always used fnctl on Solaris IIRC, so it is definitely the most tried and tested) joe - The official User-To-User support forum of the Apache HTTP Server Project. See URL:http://httpd.apache.org/userslist.html for more info. To unsubscribe, e-mail: [EMAIL PROTECTED] from the digest: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
RE: [EMAIL PROTECTED] Apache 2.0: Recommended Mutex types for Solaris 8,9,10
Joe, Thanks for replying to my post. Isn't the AcceptMutex only used inside the parent process to serialise the access to the accept(2) system call ? Is your comment about cross-process mutex relevant for SSLMutex rather than AcceptMutex , and would that mean that in your opinion although it might be OK to use pthread for AcceptMutex it should be avoided in the case of SSLMutex ? Would you have any idea of the performance implications of using fcntl or flock over sysvsem or pthread ? Thanks -ascs -Original Message- From: Joe Orton [mailto:[EMAIL PROTECTED] Sent: Wednesday, October 05, 2005 3:14 PM To: Axel-Stéphane SMORGRAV Cc: users@httpd.apache.org Subject: Re: [EMAIL PROTECTED] Apache 2.0: Recommended Mutex types for Solaris 8,9,10 On Mon, Oct 03, 2005 at 11:17:38AM +0200, Axel-Stéphane SMORGRAV wrote: As far as I can tell, in Apache versions 2.0.49 and prior the default Mutex type (at least for AcceptMutex) was pthread. It then changed to fcntl around Apache 2.0.50 leading me to post the following http://issues.eu.apache.org/bugzilla/show_bug.cgi?id=32325 issue, and explicitly set both mutexes to pthread which appeared to solve the problem. Several people have since posted similar problems and solved the issue by explicitly setting the mutex types to something other than AcceptMutex default and SSLMutex default. Does anyone have any idea of the pros and cons of the different mutex types, provided they are available on any given platform? Is there any reason to use a different mutex implementation for AcceptMutex than you would for SSLMutex? What are the performance and operational issues with each mutex type? A pthread cross-process mutex is not released if the process holding it segfaults; this can lead to the entire server deadlocking particularly if using a threaded MPM. I don't think anybody knows what conditions will lead to the fcntl EDEADLK issue on Solaris, that is really a problem which needs more investigation. (historically, 1.3 always used fnctl on Solaris IIRC, so it is definitely the most tried and tested) joe - The official User-To-User support forum of the Apache HTTP Server Project. See URL:http://httpd.apache.org/userslist.html for more info. To unsubscribe, e-mail: [EMAIL PROTECTED] from the digest: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
[EMAIL PROTECTED] Apache 2.0: Recommended Mutex types for Solaris 8,9,10
As far as I can tell, in Apache versions 2.0.49 and prior the default Mutex type (at least for AcceptMutex) was pthread. It then changed to fcntl around Apache 2.0.50 leading me to post the following http://issues.eu.apache.org/bugzilla/show_bug.cgi?id=32325 issue, and explicitly set both mutexes to pthread which appeared to solve the problem. Several people have since posted similar problems and solved the issue by explicitly setting the mutex types to something other than AcceptMutex default and SSLMutex default. Does anyone have any idea of the pros and cons of the different mutex types, provided they are available on any given platform? Is there any reason to use a different mutex implementation for AcceptMutex than you would for SSLMutex? What are the performance and operational issues with each mutex type? Personally on Solaris I am reluctant to use a file-based mutex (fcntl or flock) for performance reasons, and I am reluctant to use a semaphore because I already have enough problems with orphaned semaphores as it is. I would however like to hear from someone that has some hard evidence of one type being better than the other. -ascs - The official User-To-User support forum of the Apache HTTP Server Project. See URL:http://httpd.apache.org/userslist.html for more info. To unsubscribe, e-mail: [EMAIL PROTECTED] from the digest: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
