[CP2K-user] [CP2K:11011] ASE CP2K interface

[Geng Sun]

Hello Ole,

Yes, I am using MPI. I am running CP2K on cori, which is  a Cray system.
I am not very sure, but it seems that the MPI is mpich, because the 
cray-mpich module is loaded when I run CP2K.

I have 331 atoms in the calculations. I tried to reduce the position string
self._shell.send('%.18e %.18e %.18e' % tuple(pos))
to 
self._shell.send('%.8e %.8e %.8e' % tuple(pos))

I assumed this may avoid possible deadlock of the PIPE, but it does not 
work in the end.

Best

Geng


在 2018年12月4日星期二 UTC-8下午12:59:20，Ole Schütt写道：
>
> Hi Geng, 
>
> are you using MPI? Then this is probably where the buffering happens. 
> Depending on which MPI implementation you are using there might be a way 
> to tweak its stdin/out forwarding. 
>
> Out of curiosity, how many atoms does your system have? Maybe the 
> Fortran side simple tries to read too many values? 
>
> -Ole 
>
>
> On 2018-12-04 21:27, Geng Sun wrote: 
> > Hello Ole, 
> > 
> > Thank you very much for your reply: 
> > I changed the code as you suggested (below), but the problem is still 
> > present in the test. 
> > 
> > Best 
> > 
> > Geng 
> > 
> >     def send(self, line): 
> >         """Send a line to the cp2k_shell""" 
> >         assert self._child.poll() is None  # child process still 
> > alive? 
> >         if self._debug: 
> >             #print('Sending: ' + line) 
> >             sys.stderr.write("Sending: {}\n".format(line)) 
> >             sys.stderr.flush() 
> > 
> >         if self.version < 2.1 and len(line) >= 80: 
> >             raise Exception('Buffer overflow, upgrade CP2K to r16779 
> > or later') 
> >         assert(len(line) < 800)  # new input buffer size 
> >         self.isready = False 
> >         self._child.stdin.write(line + '\n') 
> >         self._child.stdin.flush() 
> > 
> > 在 2018年12月4日星期二 UTC-8上午11:05:43，Ole 
> > Schütt写道： 
> > 
> >> Hi Geng, 
> >> 
> >> this sounds indeed like a buffering issue. Could you once try to add 
> >> a 
> >> flush() on the python side. 
> >> 
> >> Basically add the following line after cp2k.py:498. 
> >> 
> >> self._child.stdin.write(line + '\n') 
> >> +       self._child.stdin.flush() 
> >> 
> >> This is probably quite inefficient. So, if it works I'll add some 
> >> logic 
> >> to flush only when recv() follows a send(). 
> >> 
> >> -Ole 
> >> 
> >> On 2018-12-04 18:08, Geng Sun wrote: 
> >>> Dear CP2K users, 
> >>> 
> >>> In the past several weeks, I frequently faced a problem when I 
> >> use 
> >>> the CP2K -ASE interface. 
> >>> 
> >>> The calculations frequently got stuck during the calculations. 
> >>> 
> >>> 1) Firstly I switched on the debug=True option in the ASE-CP2K 
> >>> calculator and I found that the calculation always gets stuck at a 
> >> 
> >>> line with *END after sending the positions to the subroutine 
> >>> cp2k_shell.popt  (I printed the information to the standard error, 
> >> so 
> >>> they are not buffered) 
> >>> 
> >>> 2) Then, I modified the cp2k_shell.F to print a lot of "labels". 
> >> Then 
> >>> I found that the code may get stuck at the line of "READ 
> >>> (*,*,iostat=iostat) pos" like below. I can always get "begin to 
> >> read 
> >>> pos" in the standard error, but I can not reach "LABEL-1". 
> >>> 
> >>> WRITE(0,*) "begin to read pos" 
> >>> CALL m_flush(0) 
> >>> READ (*,*,iostat=iostat) pos 
> >>> WRITE(0,*) "LABEL-1" 
> >>> CALL m_flush(0) 
> >>> 
> >>> I attached the modified cp2k_shell.F and the standard out/error 
> >>> generated by the slurm batch system with this post. You can see 
> >> that 
> >>> the program get stuck at the second optimization step just after 
> >>> sending the positions. 
> >>> 
> >>> I greatly appreciate your suggestions for fixing this confusing 
> >> bugs. 
> >>> 
> >>> Thanks in advance. 
> >>> 
> >>> Geng 
> >>> 
> >>> -- 
> >>> You received this message because you are subscribed to the Google 
> >> 
> >>> Groups "cp2k" group. 
> >>> To unsubscribe from this group and stop receiving emails from it, 
> >> send 
> >>> an email to cp2k+... at googlegroups.com. 
> >>> To post to this group, send email to cp... at googlegroups.com. 
> >>> Visit this group at https://groups.google.com/group/cp2k [1]. 
> >>> For more options, visit https://groups.google.com/d/optout [2]. 
> > 
> >  -- 
> > You received this message because you are subscribed to the Google 
> > Groups "cp2k" group. 
> > To unsubscribe from this group and stop receiving emails from it, send 
> > an email to cp2k+... at googlegroups.com <javascript:>. 
> > To post to this group, send email to cp... at googlegroups.com 
> <javascript:>. 
> > Visit this group at https://groups.google.com/group/cp2k. 
> > For more options, visit https://groups.google.com/d/optout. 
> > 
> > 
> > Links: 
> > ------ 
> > [1] https://groups.google.com/group/cp2k 
> > [2] https://groups.google.com/d/optout 
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20181204/f7180dca/attachment.htm>