Hi Tomas,
Thanks for the feedback. The screen below, I think, results from
attempting to kill a WAITING task. In windows case, the instance is net
yet started (contrarily to its RSessionWrapper related object) when
canceling, so killing it won't work (we even don't have its PID yet).
Presumably easy to fix.
One last thing I was wondering about: Have you noticed a 'too long'
starting lag, while running many task at once under Windows? If this is
the case, don't you think, one could - in Windows case only - warmup a
certain amount of asleep instances that could be reused (never closed
again) when R requests are thrown? That would speed up significantly
processing the request. The drawback would be having N Rserve
instances/processes (even if they are sleeping) remaining open all along
the life time of the MZmine instance itself. So the question is: Is this
acceptable to have those (lets say nbCores or nbCores/2) child processes
alive on the machine, all along? Or shall we preferably deal with the
possibly 'quite long' instantiation time, each time we ask for a new R
depending Task (since the current implementation closes the Rserve
instance related to a given Task, once the Task is done)?
Implementation would become:
* One 'RSessionWrapper' object per task, and dying with the Task.
* Several (pooling system) 'Rsession' objects / Rserve asleep process -
from 'Rsession' library, kept alive and registered for later call (that
is: re-linkable with a new 'RSessionWrapper').
Note: Personally, I'm not a big fan of spawning many system processes
that way (threads are OK, but processes...), but may be this is a common
acceptable practice (I don't know much about that).
Hope I made myself clear enough.
Thanks for your lights and advices.
Cheers
Gauthier
On 12/10/2014 02:40 AM, Tomas Pluskal wrote:
Hi Gauthier,
First, I'd like to have feedback about the behavior of this new
Rserve wrapper under Windows, since it is the part that bugged me
mostly. As mentioned in a previous message, on this system, forking
processes is not a native behavior, so 'emulating' such a feature
under Windows is a bit tricky, and, as a consequence, throwing a task
requiring Rserve (AKA Rserve) in combo (so several tasks, on several
files) needs some sort of synchronization and can be slow and (may
be) not so thread safe (normally I made it safe, but still need more
feedback, just to make sure). Same thing when canceling several tasks
at once.
So thanks very much for testing it (the baseline module actually),
having the above in mind.
I noticed some error messages when canceling tasks:
(note it says pid '-1', and also Failed loading package 'ptw',
although this package is installed and working fine.
I agree providing the R services to other modules via a specialized
module is the right way to do. I would suggest, even more, to create
an MZmine*R*Module, the modules requiring the R services would
inherit from. This way would provide an easy way to distinguish those
modules among the others (which would certainly ease the development
at some point). What do you think about this approach?
Sure, why not.
Testing 'R_HOME' detection at startup is good idea too. I would
suggest, in case MZmine couldn't find R_HOME right away by itself, to
make it open the preference window on the fly and ask the user to
give a correct setting for it, or preferably, may be, warn him that
R_HOME ain't OK and give him a chance to fix it (YES/NO warning
dialog which would open the preferences setup dialog if YES was
pressed). Would you be fine with the latter approach?
That is fine, but this dialog should be shown only when the user
attempts to run a method that requires R (not at MZmine startup).
As said in the very beginning of this text, Windows does not really
start Rserve just once and then fork the main instance when an R
request has to be evaluated. It starts a new fresh Rserve (master
instance) every time (well, once per Task, actually). That is to say,
starting Rserve at MZmine startup/warmup is the elegant and efficient
way of doing under *nix, but it will slow down the whole application
startup, especially under Windows. Does it sounds acceptable to you?
(I don't have a predefined personal opinion about that particular
point. But this can be a drawback for those who do not have a war
machine as a computer).
OK then, it is not necessary to start Rserve on MZmine startup, we can
start it separately for each task.
[Tomas]: I am not sure about my understanding of your latest
suggestion: "Eventually, we can change the existing modules that use
R to take advantage of this new module, and finally we can remove the
JRI interface".
Did you mean to do things step by step like:
1/ implement the said new module and startup checking
2/ make the targeted modules use it (keeping JRI as a fallback feature)
3/ later remove JRI
Or, did you mean:
1/ implement the said new module and startup checking
2/ make the targeted modules use it and remove JRI in the meantime
?
I meant the second option. I would like to have only one R backend in
MZmine.
Cheers,
Tomas
===============================================
Tomas Pluskal
G0 Cell Unit, Okinawa Institute of Science and Technology Graduate
University
1919-1 Tancha, Onna-son, Okinawa 904-0495, Japan
WWW: https://groups.oist.jp/g0
TEL: +81-98-966-8684
Fax: +81-98-966-2890
------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=164703151&iu=/4140/ostg.clktrk
_______________________________________________
Mzmine-devel mailing list
Mzmine-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mzmine-devel
--
Gauthier BOAGLIO
CEFE - UMR 5175
1919 route de Mende
F-34293 Montpellier cedex 5
Tel: +33/0 4 67 61 32 15
Fax: +33/0 4 67 61 33 36
email: gauthier.boag...@cefe.cnrs.fr
www:
http://www.cefe.cnrs.fr/en/evolutionary-ecology-and-epidemiology/gauthier-boaglio
http://www.evolepid.org/people.php?name=boaglio
------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=164703151&iu=/4140/ostg.clktrk
_______________________________________________
Mzmine-devel mailing list
Mzmine-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mzmine-devel
