[654] in NetBSD-Development
afs stuff
daemon@ATHENA.MIT.EDU (Ken Raeburn)
Mon Mar 13 08:24:36 1995
Resent-From: John Kohl <jtk@atria.com>
Resent-To: netbsd-afs@MIT.EDU
Date: Sun, 12 Mar 1995 17:54:44 -0500
From: Ken Raeburn <raeburn@kr-pc.cygnus.com>
To: jtkohl@MIT.EDU
I've updated to a recent kernel (-current as of a few days ago) and recent
afs code. I think I may still be losing memory somewhere, but I'm not
sure.
While trying to investigate, one thing I tried was "fs setca 1" many
times; eventually, "fs getca" reported 0 blocks in use. Now an afsd is
using lots of cpu time, and seems to be constantly in disk wait. (Of the
six afsd processes, this is the next-to-lowest numbered one.) Raising the
cache size doesn't seem to fix the problem. Actually, raising the cache
size doesn't seem to work. I raised it to 100, and 10 blocks were in use
after I listed /afs/athena, but it's gone back to 0 again. Maybe afsd is
stuck in some sort of loop trying to purge everything from the cache that
it can, despite the fact that it no longer needs to?
Frequently console input gets delayed for several seconds at a time while
this is happening, too. The keystrokes still register, I can switch
virtual consoles, and I can get results from ^T, but none of the programs
on any VC actually receive the characters or interrupt signals until this
pause is over, then they all seem to respond. According to ^T some of the
programs are running, but I'm not convinced. They aren't producing any
output (though they're programs that wouldn't produce much anyways) and a
stack trace from ddb always indicates a traceback through several "_end"
invocations to _syscall. (Do loaded modules get their symbol tables made
known to ddb?)
I don't know if this second problem is related to afs directly, but I
believe it only started happening after the "fs setca 1".
