[OpenAFS-devel] cached files writes/updates on disconnected from server client

u-openafsdev-lhjd@aetey.se u-openafsdev-lhjd@aetey.se
Thu, 6 Apr 2017 09:18:27 +0200


On Wed, Apr 05, 2017 at 08:31:26PM -0500, Benjamin Kaduk wrote:
> On Wed, Apr 05, 2017 at 04:21:08PM +0200, Yury Tsaregorodtsev wrote:
> > can anyone tell me how to cache writes/updates operations if client lost connection (or connection is poor and slow) ?
> > I have 6 hosts which need to use same files across them, hosts connected by LTE internet, all reads/writes should be close to realtime operations.
> > There is no huge read/writes (around few MB/hour only).
> 
> Sadly, I think there is some substantial engineering effort needed

The described scenario can happen to be a case for the Coda file system.

Be warned that Coda has its specific rough edges, but it was
devised from the beginning to cope with intermittent connections.

Depending on your write patterns (may be you can even adjust them to
become more "Coda-friendly") and on the connectivity behaviour you can
face update conflicts or not. We let among other things several hundred
hosts report their status via Coda, though on a different kind of network
than you mention.

To be more specific: if each object to be updated has a single writer
party and all the other parties are only reading the object, it is fine.
Trying to reliably update an object by many parties, which implies
either corresponding locking or write conflict resolution, is a very
hard task given an unreliable network (Coda would not help with this,
unless applying its "application-specific resolution" which is probably
out of your scope but is also a possibility).

Regards,
Rune