Re: generalising oxenstored for unix nodes too

[Anil Madhavapeddy <anil@xxxxxxxxxx>]

On 15 Feb 2012, at 17:08, Thomas Gazagnaire wrote:

>> Thomas, Dave, any thoughts on this? I was thinking of hanging a
>> first-class module from a node, which could be unpacked and decide
>> how to deal with that part of the namespace.  This way, we could
>> mix-and-match Xen portions of the namespace, as well as other nodes
>> that have completely different side-effects.
> 
> If I remind correctly, the only important thing in the xenbus protocol is to 
> be able to give an identifier to transactions ie, every query has to be 
> tagged with an ident which will identify the current transaction. If the 
> ident is 0, then there is no active transaction, and there is an API call to 
> start a new transaction which return a fresh ident. There are also some 
> specific code to handle connections coming from  dom0 (using a socket) and 
> from other domains (using ring buffers): we were needing both in xenserver so 
> that should be quite easy to reuse.
> 
> If needed, I have some ideas to improve the transaction aborting mechanisms, 
> but the current scheme is already quite good (at least for typical Xen uses).

Ok, just going with making it Lwt+functor for now then, as is
the quickest. I had a vague idea that we might be able to hang
arbitrary continuations as the value for an entry, rather than just
strings, so they could be woken up by the scheduler.

However, thats step 2 :)  Just having it working in a stub domain
for now would be most useful.

-a