On Jul 27, 2010, at 3:09 PM, Chris Murphy wrote:

> 
> 
> On Jul 27, 2010, at 12:52 PM, Dan Shoop wrote:
> 
>> 
>> On Jul 27, 2010, at 2:49 PM, Chris Murphy wrote:
>> 
>>> On Jul 27, 2010, at 12:26 PM, Dan Shoop wrote:
>>> 
>>>> 
>>>> On Jul 27, 2010, at 12:13 PM, Chris Murphy wrote:
>>>> 
>>>>> 
>>>>> On Jul 27, 2010, at 9:32 AM, Dan Shoop wrote:
>>>>>> 
>>>>>> The latter is not a supported operation. Yes, I know that sounds crazy 
>>>>>> but that's not a workflow Adobe approves of with their products Instead 
>>>>>> save locally, then copy. Directly saving files in Adobe products to 
>>>>>> fileshares has never been supported by Adobe.
>>>>> 
>>>>> Understood but it has worked flawlessly from InDesign for a year with 
>>>>> POSIX permissions. Upon changing to ACLs, we get a permissions error from 
>>>>> InDesign. Today we removed the ACLs, and went back to conventional 
>>>>> permissions and now InDesign can write directly into these drop folders.
>>>> 
>>>> Which only indicates that you did not properly capture the required set of 
>>>> ACEs. Anything you can do with ownerships and permissions you can achieve 
>>>> better with ACLs.
>>> 
>>> Seems like a reasonable explanation.
>>> 
>>> Admins on the other end are telling me that during the test phase the ACLs 
>>> were working correctly for all users and all drop folders. And then 
>>> inexplicably a week into deployment users started getting the permissions 
>>> error from InDesign.
>> 
>> So something may have changed.
> 
> *shrug* Not intentionally! i.e. at least no one is fessing up to having 
> modified the shares' ACLs between the time it was working and the time it 
> wasn't working; and also the ACLs as reported by ls -le are the same when it 
> was working and when it wasn't working. But heck with marginal drives it may 
> not be the ACLs that are corrupting, but something else related to either the 
> interpretation of them.

This is the problem with silent data corruption. It can occur anyplace. However 
data corruption of filesystem metadata is about the only corruption that is 
repairable. 


>>> And this approximately coincides with the XServe announcing a SMART failure 
>>> prediction for one of the mirrored drives. So for all I know, iffy drives 
>>> are corrupting themselves and it's just coincidentally making ACLs look 
>>> like the problem.
>> 
>> 
>> Entirely possible. Google for "Silent Data Corruption". 
> 
> Wow. That's kinda shocking. I mean, it shouldn't be, but it's almost like 
> we're f'd with a basic backup plan because the odds of having simultaneous 
> corruption of a critical file over the course of time on two separate drives 
> is actually not insignificant.

Which is why it's such a problem. It's why there's ZFS. 

>> 
>> And this isn't a RAID5, is it? If so, you're in for problems.
> 
> It is RAID 1. So, yes.

Then no, you have significantly more protection against corruption, but a RAID1 
/pair/ has no idea which member has the non-corrupt data. It can just protect 
against a block being unreadable. 

> It's a relatively low priority server, in that it's not accumulating critical 
> data. We can blow it away, rebuild it, and restore the print queues very 
> quickly and be back up and running - even if we had to change platforms since 
> it's a 3-platform RIP. However, the lack of error detection means the 
> software RAID 1 likely took corruption as valid changes and propagated that 
> to the other drive.

It doesn't work that way. 

-d

------------------------------------------------------------------------
Dan Shoop
Computer Scientist
[email protected]

GoogleVoice: 1-646-402-5293

aim: iWiring
twitter: @colonelmode

_______________________________________________
MacOSX-admin mailing list
[email protected]
http://www.omnigroup.com/mailman/listinfo/macosx-admin

Reply via email to