[OpenAFS-devel] AFS and ufs/sds logging on solaris

Frank Batschulat Frank Batschulat <Frank.Batschulat@Sun.COM>
Wed, 26 Sep 2001 16:48:25 +0200 (MEST)


>To: Frank Batschulat <Frank.Batschulat@sun.com>
>cc: openafs-devel@openafs.org
>Subject: Re: [OpenAFS-devel] AFS and ufs/sds logging on solaris
>Date: Wed, 26 Sep 2001 10:38:43 -0400
>From: Nickolai Zeldovich <kolya@MIT.EDU>
>
>> 1.) has AFS been tested on UFS filesystems with
>>     logging turned on ? (since Solaris 7)
>
>For cache partitions, UFS logging leads to deadlock if there
>is anything else stored on the same partition as the cache.
>As far as I can tell, this happens when you copy a file from
>AFS onto the cache partition.  The cache manager tries to
>write something to the cache partition, and this operation
>gets placed in transaction A.  If the cache partition is too
>full, the truncate daemon is woken up and tries to truncate
>some files, but this operation gets placed into transaction
>B.  Now, transaction A can't close until the cache manager
>finishes the write, but it's waiting for the truncate daemon
>which can't finish either (B can't happen before A).
>
>(If anyone is interested, I could probably find my corefiles
>again and get more accurate data on this problem.)
>
>-- kolya

Exactly thats what happens in various different ways and hence
I'm asking myself:

Is there a statement "Dot use AFS on logging UFS" in the readme/notes ?


frankB