Red Hat Bugzilla – Bug 186670
Last modified: 2007-11-30 17:11:28 EST
eAccelerator 0.9.4 has been released about a month ago. This is mainly a bugfix
Do you know if any of the x86_64 specific locking issues are fixed? Or have
those been identified as being arch specific problems that have nothing to do
with eA itself?
I think most of these problems have been fixed. There were two things causing
1) spinlocks were never picked because they would only be enabled on x86 and not
x86_64. These will be selected now and they work fine.
2) After spinlocks sysvipc semaphores were selected but there was a error in the
implementation. The owner was root and they weren't publicly writeable, so the
apache children couldn't aquire a lock.
A lot of other bugs are fixed so upgrading would be a good thing.
OK, I'll be testing 0.9.4 on some of my busy x86_64 servers running 4.4.x then
update the packages if things work ok.
The default on x86_64 seems to be shm:sysvipc and sem:spinlock, and works fine
for me now... are these the optimal values for Fedora according to you?
I don't know if they are the fastest but they are used by 99.9% of all eA users
so they've gotten the most testing. Those are the ones I recommend using. That's
also the reason the configure script selects them ;)
I've rebuilt this package at last. I had been waiting to get the development
package of 0.9.5beta2 out first, but since I'm out of luck with a broken Rawhide
tree each time I try, I've decided to push this package anyway.
Thanks for all your advice!