[rescue] Potential SVM issue in Solaris 10; anybody got a spud?

[Ahmed Ewing]

> Well, I was about to get the Ultra 60 ready to install on the two 
> brand new 73 GB Cheetahs (hopefully they do get recognized; if not, 
> I'll have a rather nice pair of drives to sell off, or put into an x86 
> blackbox or something).  I did a Google search for /"solaris 10" svm/ 
> and stumbled upon this:
>
> |http://sunsolve.sun.com/search/document.do?assetkey=1-26-101686-1
> |
> |Synopsis: Booting Off of a Single Disk From a Mirrored Root Pair May 
> Fail With a Panic
<snip>
>  The patch is out, but the date on the file is February 06, so I don't 
> think it's in the latest release of Solaris 10 (1/06).  I guess that 
> will be among the first things I seek after getting a stable system.

Basically, yes that would be a very good idea. I admittedly have not 
checked this specific one, but if the Sun Alert doc is public, one can 
safely assume that any patch referenced is as well.

Also worth noting: if this U60 is your personal box and not a business 
server with production platform considerations, you might just want to 
install the latest Solaris 10 Recommended Patch Cluster (RPC) from 
SunSolve's Patch Portal and not worry about the individual fixes in a 
piecemeal manner. That patch would certainly be included. Sun does 
*not* test these patches together, though; the cluster is updated 
dynamically and just give you the latest versions of all patches in a 
single whack for convenience (contrast this against, say, a Windows 
Service Pack, or the old Solaris Maintenance Updates). So YMMV re: 
stability/full compatibility. The cluster README does speak to this, 
but really, how often are those read thoroughly? :)

> What are the best practice rules on logging with UFS, for root and 
> other partitions?  I booted up the U60 on the 18 GB disk install of 
> the initial Solaris 10 release, and it only has logging on the slice I 
> used for /export/home.  One of my other Solaris-savvy friends says 
> that ideally, logging should be enabled on all partitions, because 
> otherwise the filesystems aren't journaled in case of a power outage 
> or spontaneous reboot.

Best practice is definitely to utilize logging on all. This allows you 
to sidestep fsck after ungraceful shutdowns/crashes, saving you time as 
well on the reboot. Rolling the log is MUCH faster (and safer) than 
getting caught in fsck hell, particularly when you start scaling to 
enormous filesystems/datasets.

I was under the impression that at least as of Solaris 9, logging was 
enabled by default on all UFS filesystems... but if you want to be 100% 
sure, edit the /etc/vfstab entry for each filesystem in question 
(except swap) to definitively include a "logging" option. Redundant at 
worst, cheap insurance at best.

> On another note, does anybody in the Boston area have a spud to sell 
> off that would fit in the U60?

Yup, I can scare one up. Email me offlist.

Regards,

-A