[OpenAFS] Daemon logging to AFS-mounted logfile

[David Fulton]

In the past, we have just assigned each server a seperate log file and 
allowed them to log for a day. Then we would restart the daemons to 
flush the data to the file and then moved the file and a perl script put 
everything into a new file in proper sequence for stat purposes. This is 
the only way to avoid that unless you use syslog and have all the apache 
daemons log via syslog to one machine. This has the same effect and is 
prefferable as it does not require a restart of apache. If you plan on 
allowing clients access to the web server space (in particular CGI) 
restarting can be a bad thing if there is an htaccess file or something 
that is messed up it can keep the whole thing from starting durring the 
restart. Better to get a call from a customer whose htacces is not 
working than 1000 with down websites. I digress.

Regards,
    Dave

Sven wrote:

>In trying to create a distributed file system to emulate clustering (this
>will cluster a series of web servers together), I have come across a problem
>in logging to a file that exists in the AFS space. Using Solaris 9 and AFS
>compiled from the 2002-07-05 snapshot, I have mounted the apache data, conf,
>and logging directories into a RW-mounted directory. A total of 4 servers
>are currently being used in our test setup. The problem is that when Server
>2, for example, is started, it opens a pipe to the log file; any writes to
>the logfile are not "propogated" or released until the daemon is stopped
>(effectively closing the data pipe). At that point, the log file is visible
>in its entirety on the other servers.
>
>Is there a workaround for this, or am I limited to only having webpage
>updates propogate throughout the system? These will be frontpage sites so
>the file writing will be done by "nobody" or the apache daemon user which I
>assume will not function in the same manner as logging. I would assume that
>by setting the particular directory (/afs/cellname/www) acl to
>system:anyuser read,write, etc (essentially "all") that the daemon will be
>able to write the webpages without having to undergo authentication .....
>(which I guess is a second question afterall) ...
>
>Thanks,
>
>Sven
>
>_______________________________________________
>OpenAFS-info mailing list
>OpenAFS-info@openafs.org
>https://lists.openafs.org/mailman/listinfo/openafs-info
>