[HCoop-Discuss] Planning

[Adam Chlipala]

Derrick Brashear wrote:
> Adam Chlipala <adamc at hcoop.net> writes:
>   
>> It seems likely that most of our performance and availability
>> problems from the last few years would not have come up or would
>> have been less severe if we weren't using AFS.
>>     
>
>
> Can you quantify that?
>   

Keep in mind that I'm talking about the real world, where we have no one 
familiar with AFS willing to make any particular time commitment to 
HCoop.  That means that likely newbie mistakes in controlling AFS 
configuration and processes have to be counted as costs of using AFS.

We have a repeating problem where Apache gets into a state where 
processes don't have the right tokens to run dynamic content properly.  
I'm sure there's a way to fix it, but it doesn't seem like that's going 
to happen, given the realities of our admin situation.  This definitely 
would not be a problem with local filesystems.

We have had several incidents of monotonically increasing sets of zombie 
processes (mail daemon and otherwise) that are frozen in the 'D' state 
trying to access files in AFS.  This can effectively stop all IMAP/POP 
service.  The system wasn't otherwise overloaded in any detectable way.  
I've never seen this happen with local filesystems.  This was the root 
problem behind our recent downtime of a few hours.

I'm sure other things have come up that I'm not remembering now.  I 
stand by my characterization of the downsides of AFS for us, though 
maybe I should make clear that the lack of experienced dedicated staff 
is a key cause.