[OpenAFS] solaris 10 versions supporting inode fileservers

Kevin Sumner ksumner@email.unc.edu
Fri, 08 May 2009 10:46:30 -0400


I wonder if the opposite would be better -- check the filesystem the fileserver 
is going to start writing to against a valid set of filesystems (and check for 
some specific bad cases like Solaris UFS logging) and print HUGE, 
end-of-the-world warnings if the fs is one that might not work well (eat data, 
etc).

I think this is better than refusing to start as it still gives the admin the 
possibility of laying a fileserver over a filesystem (or filesystem 
configuration) that isn't really supported.

Kevin Sumner
ITS Enterprise Storage Management
University of North Carolina at Chapel Hill
CB# 1150, 440 W. Franklin Street, Office G409
Chapel Hill, NC 27599-1150

ksumner@unc.edu

919.962.1547 (office)
919.259.9734 (mobile)
919.445.9485 (fax)



Dan Pritts wrote:
> Perhaps someone could update the inode fileserver to check whether the
> filesystem is mounted with logging?  And refuse to start if so?
> 
> It looks like a quick call to getmntany() would tell you.
> 
> It wouldn't prevent someone from coming along and remounting the filesystem
> with logging enabled once the fileserver had started, but i imagine it will 
> save someone some trouble somewhere along the line.
> 
> On Tue, May 05, 2009 at 06:30:31PM -0400, Jason Edgecombe wrote:
>> Harald Barth wrote:
>>>> Does anyone have any information on which versions of solaris 10
>>>> are not safe?  Any direction to a bug report or a mail list thread
>>>> would be appreciated.
>>>>    
>>> I think the sneaky thing was that
>>>  a) UFS with logging may rot your files
>>>  b) Some update made logging the default mount option
>>> I don't think ZFS is safe either.
>>>
>>> Harald.
>> ufs without logging is the only safe option for the inode fileserver on 
>> solaris.
>>
>> The default logging option was changed in Solaris 9 9/04
>> See http://docs.sun.com/app/docs/doc/817-5770/6ml72d6kd?a=view
>>
>> It's sneaky because the default is logging is on for disks over 100G or 
>> some similar threshold.
>>
>> UFS with logging WILL eat data with the inode server, you will have to 
>> salvage on each service restart when the /vicepX is unmounted and 
>> remounted (i.e. every boot/reboot).
>>
>> The namei fileserver is safe for ufs (with or without logging) and ZFS 
>> filesystems on solaris.
>>
>> Use namei, it's safer and faster on ufs w/logging than inode without 
>> logging.
>>
>> BTW, where should I document tidbits like this?
>>
>> Sincerely,,
>> Jason
>> _______________________________________________
>> OpenAFS-info mailing list
>> OpenAFS-info@openafs.org
>> https://lists.openafs.org/mailman/listinfo/openafs-info
> 
> 
> danno
> --
> Dan Pritts, Sr. Systems Engineer
> Internet2
> office: +1-734-352-4953 | mobile: +1-734-834-7224
> 
> ESCC/Internet2 Joint Techs
> July 19-23, 2009 - Indianapolis, Indiana
> http://jointtechs.es.net/indiana2009/
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info