Hmm, I am pretty suspicious of the config file
/etc/apache2/vhosts.d/30_subversion_ssl_vhost.conf
When you remove the '-D SSL' you will cause all parts of the configuration files that are enclosed in ... to be skipped. The default SSL vhost file on my Gentoo box is wrapped in that tag so I wonder if, by removing the '-D SSL', you are preventing the config in 30_subversion_ssl_vhost.conf from being run at all and if that is what is allowing Apache to start.
If you temporarily remove the file 30_subversion_ssl_vhost.conf from /etc/apache2/vhosts.d does Apache run? Are there any other SSL related vhost.conf files in vhosts.d? My reasonably fresh/unused Apache install's vhosts.d directory looks like this:
# pwd && ls
/etc/apache2/vhosts.d
00_default_ssl_vhost.conf 00_default_vhost.conf default_vhost.include
edit 1:
So much for that theory :) I am now wondering if the problem is with the Apache SSL setup itself. I apologize if I am covering ground you have already checked but I am wondering if you could do the following to help verify your Apache install.
On my Apache install with working SSL the use flags are as follows:
# emerge -av apache
These are the packages that would be merged, in order:
Calculating dependencies... done!
[ebuild U ] www-servers/apache-2.2.11-r2 [2.2.11] USE="ssl -debug -doc -ldap (-selinux) -sni -static -suexec -threads" APACHE2_MODULES="actions alias auth_basic authn_alias authn_anon authn_dbm authn_default authn_file authz_dbm authz_default authz_groupfile authz_host authz_owner authz_user autoindex cache dav dav_fs dav_lock deflate dir disk_cache env expires ext_filter file_cache filter headers include info log_config logio mem_cache mime mime_magic negotiation rewrite setenvif speling status unique_id userdir usertrack vhost_alias -asis -auth_digest -authn_dbd -cern_meta -charset_lite -dbd -dumpio -ident -imagemap -log_forensic -proxy -proxy_ajp -proxy_balancer -proxy_connect -proxy_ftp -proxy_http -substitute -version" APACHE2_MPMS="-event -itk -peruser -prefork -worker" 64 kB
In particular do you have the 'ssl' USE flag set?
Also, could you use equery to verify the integrity of your apache2 install? If you do not have the equery command you can install it by running 'emerge -av gentoolkit'. The following command should verify the integrity of your apache install:
equery check apache
On my server the above command gives the following output:
[ Checking www-servers/apache-2.2.11 ]
!!! /etc/apache2/vhosts.d/00_default_ssl_vhost.conf has wrong mtime (is 1256620928, should be 1246793824)
!!! /etc/apache2/modules.d/00_default_settings.conf has wrong mtime (is 1246796304, should be 1246793824)
!!! /etc/conf.d/apache2 has incorrect md5sum
* 429 out of 432 files good
edit 2:
Well the install looks good to me, so much for theory 2. I am wondering if we can coax Apache into giving some more information on startup. In /etc/conf.d/apache2 if you change your APACHE2_OPTS line from:
APACHE2_OPTS="-D DEFAULT_VHOST -D INFO -D SSL -D SSL_DEFAULT_VHOST -D LANGUAGE"
to
APACHE2_OPTS="-X -e debug -D DEFAULT_VHOST -D INFO -D SSL -D SSL_DEFAULT_VHOST -D LANGUAGE"
and then start Apache (/etc/init.d/apache2 start) the daemon should stay in the foreground (the -X flag) and output debuging messages as it starts (the -e debug option). Maybe this will give a clue as to why it is dying on startup.
ps aux | egrep '(apache|httpd)' typically will show what apache is running as.
Usually you do not need to change the default user, "nobody" or "apache" are typically fine users. As long as its not "root" ;)
edit: more accurate command for catching apache binaries too
Best Answer
Since you've already turned off
EnableSendfileandEnableMMAP, it looks like you've done your homework. The general cause of this error is that Apache got SIGPIPE, which happens when Apache tries to write to a socket that the other end has closed (described here, look at the last post). The post suggests that the user on the far end can trigger this by pressing their Stop or Reload button. However, if you're seeing flurries of these entries all the time, I think the following is more likely:TCP_KEEPALIVE, the kernel sees the socket is busted and returns SIGPIPE to Apache.If you have a busy server, I'd imagine this is actually the usual and expected behavior. In fact, even the Apache folks themselves don't seem to concerned about it, as they apparently set the
LogLevelfor apache.org towarnso they don't even see these messages (noted here). This message is apparently only seen at levelinfo, so changing it tonoticeor higher should prevent them from being logged.