Re: [tauruslib-devel] Taurus and GIL

[Carlos P. <cpa...@ce...>]

Hi, just in case someone stumbles into this who is not in the Tango List, see the very complete 
reply from Tiago in the Tango list:

On Mon 8 August 2016 09:41:27 coutinho wrote:
> Hi Giacomo,
> 
> On 07/28/2016 05:29 PM, Giacomo S. wrote:
> > Hello Carlos,
> > 
> > I have a technical question concerning Taurus and Python GIL.
> > How is multi threading managed inside Taurus?
> > Once I had a talk with Tiago about this topic and, as far as I can
> > remember, he said that, where required, the
> > GIL is somehow disabled by Taurus in order to achieve real multi
> > threaded behaviour (real POSIX/Windows threads).
> 
> The  GIL only needs to be managed by PyTango (not taurus) because this
> is where the blocking network calls occur.
> Basically, in every place PyTango calls a Tango C++ I/O blocking call,
> it releases the GIL so that other threads that may execute python
> code are not blocked.
> 
> > Could you please confirm this and, consequently, claim that Taurus
> > threads are really concurrent threads?
> 
> Taurus is based on the concept of pluggable Control Systems. By this I
> mean it can work with Tango, Epics or other third-party control
> systems. If the taurus plug-in that is written for a specific control
> system happens to block the thread there is nothing that taurus can
> do about it. It all depends on how the plug-in and the third-party
> libraries are written. For the case of Tango I can tell you that the
> PyTango library manages the GIL properly.
> 
> > Second question. In a situation where a third party library relies
> > on
> > the GIL (being enabled), is there any
> > concurrency issue potentially introduced when the same library
> > operates in a context where the GIL is disabled?
> 
> Since taurus does not handle the GIL by itself it does not generate
> any concurrency issue.
> 
> > Gracias, Giacomo.
> 
> Thanks Giacomo for your pertinent questions. It is important that in
> the context of a distributed (potentially blocking) control system,
> we understand what happens in python with the GIL, who has the
> responsibility to manage it and where blocking may occur.
> 
> In the case of Tango, it should be the library that handles Tango
> (PyTango) who abstracts the underlying complexities of multi-threaded,
> I/O blocking control system.
> 
> Taurus does not take care of the GIL because it trusts that the
> blocking library does it well. Having said that, taurus does create a
> pool of threads that poll the underlying attributes (tango or other)
> in an old style attempt to paralyze blocking I/O calls.
> We choose threads in taurus because at the time PyTango was not
> coroutine friendly. Probably if it was done today we would have used
> another simpler, lighter mechanism (like gevent or asyncio).
> 
> Hope it helps,
> 
> Cheers
> Tiago




-- 
+----------------------------------------------------+
 Carlos Pascual Izarra
 Scientific Software Coordinator
 Computing Division
 ALBA Synchrotron  [http://www.albasynchrotron.es]
 Carrer de la Llum 2-26
 E-08290 Cerdanyola del Valles (Barcelona), Spain
 E-mail: cpa...@ce...
 Phone: +34 93 592 4428
+----------------------------------------------------+