[nycbug-talk] Apache and FreeBSD

[Bjorn Nelson]

Matt Juszczak wrote:
>> Is yours any different and just to double check, your /var/log is 
>> definitely local, right?  I couldn't tell from your lsof, didn't see 
>> the underlying device.
>
> Yes.  It is.
>
> It's weird also, I'm getting this:
>
>  12943 httpd    CALL  fcntl(0xf,F_SETLK,0xffffffffffffcab0)
>  12943 httpd    RET   fcntl -1 errno 35 Resource temporarily unavailable
>
>
> in lsof, I'm getting this:
>
> httpd     12943       www   14u    VREG               0,94 0        17 
> /tmp (/dev/amrd0s1e)
> httpd     12943       www   15uW   VREG               0,94 0        18 
> /tmp (/dev/amrd0s1e)
> httpd     12943       www   16u    VREG               0,94 0        19 
> /tmp (/dev/amrd0s1e)
>
>
> If you notice, the 15 (which maps to 0xf right, 0 - 15?) has a full 
> out lock on /tmp?  Or it's trying to?  Why?  Specific files?
>
> Also, the only two /var/log are:
>
> httpd     12943       www   19w    VREG               0,93 0     23591 
> /var/log/accept.lock.12878
> httpd     12943       www   20w    VREG               0,93 0     23591 
> /var/log/accept.lock.12878
>
> and don't appear in the resource unavailable.
Can you try to recreate what httpd is doing?  As www, try to mktemp in 
/tmp or run a flock on a file in /var/log.  Alternatively, try to remove 
perms from either dir to www to see if you can see the behavior just 
before it errors.  If you see the behavior before the crash, then you 
know it's not that dir but the other.  Maybe you can find an httpd.conf 
directive or a module that lets you start a sub process, then have it 
start something that gives a shell (xterm, screen, vnc), then you play 
around from the perspective of the process and see what doesn't work.

-Bjorn