shared blob storage is racy

[Bastian Blank]

The support for shared blob storage in ClientStorage assumes that it can read the blobs always. However if the blob directory is mounted via a network filesystem like NFS, it races against the cache on the client and server and can try to access it before the file creation was pushed through. At least for reading the ClientStorage should be prepared to retry to the load via the zodb connection if it can't read it directly.

[Jim Fulton (jim-zope)]

Have you seen this problem in practice?

[Bastian Blank (waldi)]

On Tue, Apr 12, 2011 at 02:27:28PM -0000, Jim Fulton wrote:
> Have you seen this problem in practice?

Yep. I was not able to finish a plone blob migration without errors.

Bastian

--
We fight only when there is no other choice. We prefer the ways of
peaceful contact.
  -- Kirk, "Spectre of the Gun", stardate 4385.3

[Jim Fulton (jim-zope)]

On Wed, Dec 1, 2010 at 12:26 PM, Bastian Blank
<email address hidden> wrote:
> The support for shared blob storage in ClientStorage assumes that it can
> read the blobs always. However if the blob directory is mounted via a
> network filesystem like NFS, it races against the cache on the client
> and server and can try to access it before the file creation was pushed
> through. At least for reading the ClientStorage should be prepared to
> retry to the load via the zodb connection if it can't read it directly.

Is this a case where one client was writing and a client on a different
machine was reading?

The notion of that you'd need to poll the blob data is disturbing.

Jim

--
Jim Fulton
http://www.linkedin.com/in/jimfulton