> On Fri, May 03, 2019 at 03:25:25PM -0400, Nico Kadel-Garcia wrote:
> 
> (I put all your comments into one block)
> 
> None of the problems you've described sound familiar to me at all,
> however I've only been using AFS since 1999.  Unsurprisingly, there
> have been improvements in the code, and the in-kernel AFS client is a
> completely different client than the Transarc client from your era and
> also different from a modern OpenAFS client.

The AFS kernel modules of the IBM era are known to have suffered from the 
following limitations and implementation flaws:

1. distributed lock hierarchy violations resulting in distributed dead locks

2. /afs was backed by the configured cell's "root.afs" volume.  If there were 
connectivity issues or the fileservers failed to serve the volume, then there 
could be extended timeouts

3. most afs servers were configured to restart every Sunday morning and 
required the equivalent of "fsck" on every afs volume before restarting which 
would result in volumes becoming unavailable for an extended period of time.

4. all location server address information was stored in local configuration 
files.   If IP address changes were not reflected in local configuration 
updates, cache managers would experience timeouts or loss of access.

These are just a few of the many issues that have been addressed over the 
decades.  kafs is a 100% clean room implementation specifically designed for 
and integrated with the Linux kernel.  kafs does not mount "root.afs" volumes 
instead it follows the "dynamic root" model whereby /afs/ directory entries are 
evaluated on demand as cell names using a combination of DNS and local 
configuration files.  OpenAFS and AuriStorFS servers are rarely configured for 
weekly restarts and when they do restart the startup time is in fractions of a 
second instead of hours.  kafs attempts to be as "zero config" as possible.  
I'm not sure if there is anything more than specifying the name of the 
top-level mount point.

In any case, these implementation defects from the 90s should have no bearing 
on the packaging of kafs-client for Fedora.  The AF_RXRPC and AFS kernel 
features have been shipping in Fedora kernels distributed with F28, F29, and 
F30.   The kafs-client package is the final piece required to permit end users 
to choose the native Linux implementation over third-party, out-of-tree, GPL2 
license incompatible implementations.  

Sincerely,

Jeffrey Altman
_______________________________________________
devel mailing list -- devel@lists.fedoraproject.org
To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: 
https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org

Reply via email to