Opened 9 years ago

Closed 4 years ago

Last modified 4 years ago

#22 closed defect (fixed)

Model does not stop correctly for case of no land points in parallel mode

Reported by: jgipsl Owned by: jgipsl
Priority: minor Milestone: ORCHIDEE 2.0
Component: Driver files Version:
Keywords: Cc:

Description

When running a domain without land points the model should stop with an error message. This is the cas when compiling and running in sequential mode.

But while compiling and running in parallel mode, the model does not understand that there are no land points and stops with a :

SIGSEGV, segmentation fault occurred

Correct error message should be (as for the sequential case) :

FATAL ERROR FROM ROUTINE dim2_driver
 --> number of land points error.
 -->  is zero !
 --> stop driver

Change History (5)

comment:1 Changed 8 years ago by peylin

  • Owner changed from somebody to jgipsl
  • Status changed from new to assigned

comment:2 Changed 8 years ago by jgipsl

  • Status changed from assigned to accepted

comment:3 Changed 6 years ago by jgipsl

see also ticket #176

comment:4 Changed 4 years ago by jgipsl

  • Resolution set to fixed
  • Status changed from accepted to closed

Added test for enough grid-cells per MPI process, done in the trunk [4268].
Tests at curie shows that 2 grid-cells per MPI process is the minimum. I don't know why but 1 is not working.

comment:5 Changed 4 years ago by jgipsl

The stop when having only 1 grid-cell per proc was problematic. In some cases in offline when the master proc is heavy charged (for example if lots of input files needs to be read), when using the Load_date_balance.dat file, the master proc will have less and less number of grid-cells for each time the distribution is redone (in the end of the execution before calculating and writing the Load_balance_orchidee.dat file). For example in offline in the end, the master proc has 1 grid-cell whereas the other proc can have about 200 grid-cells (depending on the number of proc and the resolution). Running on 1 grid-cell in this case is not a problem.

Similar problem has also been seen in coupled mode.

The stop has been removed without any side effects. I do not know which case caused problem earlier at curie (see comment above).

Commit in trunk rev [4642].

Note: See TracTickets for help on using tickets.