[Gluster-users] Docs on setting up translators

[Jeff Darcy]

On 08/01/2013 12:25 PM, Joel Young wrote:
> So looking at "Translator performance/io-threads" section I see a
> snippet of configuration file (shown in violation of the philosophy
> statement below).  Adapting from that I guess I should type:
>
> "gluster volume set home performance.io-threads 32"

AFAICT (looking at current 3.4+ code) the option name is now client-io-threads
and it's boolean, so the correct command would be:

	gluster volume set home performance.client-io-threads on

> But a little lower in orange there is "thread-count Number of threads
> to keep in the thread pool."  So I figure I should type:
>
> "gluster volume set home performance.thread-count 32"

If you don't set io-thread-count (this name changed too) explicitly, then the
io-threads translator will try to auto-scale the number of threads. So enabling
io-threads without setting a count is reasonable, but setting a count without
enabling would be meaningless. I wonder if we check that.

> But then I check my self by looking at
> http://gluster.org/community/documentation/index.php/Gluster_3.2:_Setting_Volume_Options
> which I found by googling for "gluster volume set"
>
> which shows "performance.io-thread-count" as the option available.
> But is that current for 3.4.0?  Backing up the URL tree I find docs
> for 3.3.0.  Digging in to these 3.3.0 docs which are only published as
> a pdf or html tar ball at:
> http://www.gluster.org/wp-content/uploads/2012/05/Gluster_File_System-3.3.0-Administration_Guide-en-US.pdf
> it seems to still be correct then.  The admin guide doesn't seem to be
> published for 3.4 so maybe it will work, maybe it won't.
>
> What does it mean that "translators can be loaded"?  That statement
> implies that I have to do an action to load a translator and then I
> can set options.  I think that translators being loaded is something
> that the user doesn't need to know about?  It is happening
> automatically when an option is set or changed with gluster volume
> set"?

That looks like another holdover from the older style of configuration. Yes,
from a "would the code work" perspective, io-threads could be loaded at
multiple points on either the client or the server. However, from a "can I get
the CLI to do it" perspective it's loaded exactly one place on the client.

> I see on the devel list that there is a discussion about revamping the
> docs.  Am I right that some of this language has been carried forward
> from gluster 2.x days and isn't really quite accurate now?

Exactly.