I write this having solved the problem I was having, but I
feel weird about my solution for it. This is an amd64 -current system compiled on Aug 8th, with packages from Aug 9th. An Optiplex 745 at 2.4GHz, 8G ram using the stock GENERIC kernel. A vanilla system for Wordpress 4.53 using PHP-5.6.23 and Maria 10.0.26v1 with apache 2.4.23. /etc/login.conf had limits raised to infinity. The system was updated just before the wx changes. Under a light load Wordpress worked as expected. But every once in a while, an ah00037 ( Symbolic link not allowed or link target not accessible) error popped up. The client would see a page not accessible message. Under a heavy load of wget scripts the error was just about constant. Going back in the browser would get things working after a page denial, at least for a bit. Pages that once worked came up with the error often. After a period of time pages would generally not work at all. The fix to get apache working again was to restart it, but lots of wget scripts would ramp the problem up again. My "fix" was to get rid of the symlink of /var/www/htdocs to /u, and making /var/www/htdocs the main code area. In a 4 hour test with multiple wget scripts, it served about 113,000 pages without error, about 8 per second. After that test I was convinced the "fix" worked. But why? The basic apache/system setup was correct I pretty sure, or wordpress would have never worked. The problem seems like it's load related. If anyone can say "idiot--you forgot N Q and Z" I'd like top hear it, but I think I have found a bug either in Apache or OpenBSD. Ideas on the best way to test symlinks? I haven't found any comments on a symlink problem in apache or wrodpress. All the ah00037 comments talk of stuff I already verified. I'm certainly willing to do more work on this--I'd appreciate any ideas on what to test. I've never seen an error like this before... Right now I feel uncomfortably dumb. Thanks for ideas... --STeve Andre'
