[OpenAFS] runtime-tunable AFS_RXDEADTIME

[Christopher Arnold]

I was browsing through some older threads regarding timeouts on the client 
side and I was just wondering if anything ever came out of the below thread 
since it was posted?  Have things changed to where I might be able to tune 
the RXDEADTIME value in RedHat linux without a re-compile or do I still have 
to go to the src?

It would probably come in really handy for our environment where our application 
servers are continuously grabbing data from AFS space.





------begin old thread----------


>/ 1. When one of my fileservers is down and I try to access a volume
/>/    on it ("ls -l"ing a directory with a mountpoint to that volume)
/>/    ls waits and shows something like "ls : 'volume': connection timed out".
/>/    Is it possible to reduce that timeout till that message is
/>/    shown? 5 seconds would be very good...
/
If you build AFS from source, you can change the timeout period;
it's AFS_RXDEADTIME in src/afs/afs.h, and defaults to 50 seconds.
Perhaps it would be clever to make this a runtime-tunable value,
using something like sysctl on Linux..

Another solution is to use fakestat: it's an experimental feature
in the development versions, which basically makes 'ls -ld dirname'
avoid talking to the fileserver which holds dirname, if dirname is
a mountpoint for some volume.  It works quite well on Solaris, and
probably works on Linux too.  You might encounter more than the
usual number of bugs in the development versions, though.

-- kolya

---------end old thread----------




-- 
Christopher Arnold
System Administrator
Pictage, Inc.