Re: Usable translators: nsmux details

[Sergiu Ivanov]

Hello,

On Mon, Jul 12, 2010 at 08:45:47AM +0200, Arne Babenhauserheide wrote:
> On Friday 09 July 2010 14:32:52 Sergiu Ivanov wrote:
> > Yes, you can kill them manually, but nsmux does nothing about dead
> > control ports; this memory leak you cannot avoid easily :-(
> > Nevertheless, I'd expect that killing translators won't have other bad
> > effect than leaving nsmux with invalid control ports: nsmux only uses
> > control ports for closing the translators when it itself closes.
> 
> But here the size of the memory leak is on the scale of one control port per 
> started translator, isn’t it? So it would be constantly growing, but not by 
> that much (as long as nsmux isn’t used for constant querying or similar). 

Indeed, in one step nsmux will not leak that much, but if one runs it
(and uses as intended) for a day, one will have quite a lot of space
wasted.
 
> How much is missing to make nsmux close translators again? 

Sorry, I cannot tell for sure -- I haven't touched nsmux for ages.  A
solution would be to remove everything which has to do with control
ports and rely on the translators to go away on themselves, since
every translator shuts down after a certain timeout if there are no
incoming RPCs.  However, in this way nsmux will have no absolutely no
control over the (anonymous) translators it starts, which is bad for
how it was designed.

Frankly speaking, my memory holds no remembering of discussions about
how to shut down anonymous translators.  In the current implementation
an nsmux client obtains a port to an nsmux node which redirects all
requests to the corresponding anonymous translator.  Perhaps, nsmux
could use a node cache and discard those translators which are pointed
to by nodes which are about to be evicted from the cache.

I don't think making nsmux capable of shutting down translators is a
hard task in itself: the central problem is that I am not sure we have
agreed on how to implement it best.

Best regards,
Sergiu