[7084] in testers
AFS cache coherency tests
daemon@ATHENA.MIT.EDU (Greg Hudson)
Wed Jun 8 16:03:09 2005
Date: Wed, 8 Jun 2005 16:02:49 -0400
Message-Id: <200506082002.j58K2nGI020490@equal-rites.mit.edu>
From: Greg Hudson <ghudson@MIT.EDU>
To: testers@MIT.EDU
The AFS cache coherency problem appears to be easily reproduced, so I
ran the following test:
equal-rites% echo 1 > ~/dev/foo
equal-rites% echo 1 > ~/foo
men-at-arms% cat ~/dev/foo
1
men-at-arms% cat ~/foo
1
snuggle% cat ~/dev/foo
1
snuggle% cat ~/foo
1
equal-rites% echo 2 > ~/dev/foo
equal-rites% echo 2 > ~/foo
men-at-arms% cat ~/dev/foo
1
men-at-arms% cat ~/foo
2
snuggle% cat ~/dev/foo
1
snuggle% cat ~/foo
2
equal-rites is 9.4.7 linux, men-at-arms is 9.4.7 linux, and snuggle is
9.3.12 linux. I get the same results if the reader is downy, a 9.2.28
Solaris machine; the dev cell file does not update (without an "fs
flushv" to help it along), while the athena cell file does.
So, I blame the dev cell servers for the cache coherency issue.
(Unless it can possibly be blamed on the writing client. But my
expectation is that the AFS protocol does not allow the writing client
to screw this up.)
