[Gluster-devel] "namespace" bricks do what, exactly ?

[Amar S. Tumballi]

Hi Daniel,
 No need to pardon if you are asking question to learn something new. These
questions helps us a lot to refine our documentation, as its not yet ready.

I will try to answer your question inline with your mail.

>
> Pardon my (clearly newbie) question, but what are the namespace bricks
> for, exactly ?  I see that they are present in almost every example on
> the wiki, but having read a great deal of the documentation, i'm
> still fuzzy on exactly what this brick does, and how it incorporates
> into the storage definitions.


Hope this link gives a small idea about namespace.
http://www.gluster.org/docs/index.php/Understanding_Unify_Translator#Namespace_FAQ

As its a _must_ need part of unify translator, its present in all the
examples.

Also, in the "Install_and_run_GlusterFS_v1.3_in_10mins" wiki article,
> the "Example 2 [3 servers, single client]" server config example
> contains this rather ominous warning :
> # Note: Once exported, DO NOT WRITE DIRECTLY TO THIS DIRECTORY
>
> This, of course, relates to the "volume brick" definition, which
> exports "/tmp/export" on each server.  Why shouldn't we write to this
> directory after it has been exported ?  Does it have something to do
> with the namespace, perhaps ?
>

Its not related to 'namespace', but a general practice when using GlusterFS.
We _advice_ you not to write directly into exported directory because, it
may create problems when you use encryption translator, or stripe translator
etc.
While using unify, you can export a directory which already has data, or
write directly to backend, but this will cause a bit of performance overhead
for GlusterFS as it has to rebuild (or in our terms self-heal) the namespace
volume.
Hence, that note in the example.


>
> If the server generates some data which needs to be in the volume, and
> it can't (as per the warning) write directly to /tmp/export, how should
> it write the data to the volume instead ?
>

Server can also use glusterfs mountpoint, and write through that.


>
> Thank you all for your helpful responses thus far.
>

>
> --
> Daniel Maher <dma AT witbe.net>
>
>
-- 
Amar Tumballi
Gluster/GlusterFS Hacker
[bulde on #gluster/irc.gnu.org]
http://www.zresearch.com - Commoditizing Supercomputing and Superstorage!