epoc

Handling model failues

Diagnosing errors

Check the job.err file for the cause of the error. 

Sometimes the issue is due to the machine, and the job can just be re-triggered. Some examples of this are:

An MPI_Abort in the file means the model failed. This is usually due to an instability which occurs occasionally in this version of the model at high resolution.

To see which component has caused the error:

See below for instructions on getting past these instabilities. These should be documented in the suite page.

If these solutions do not work, you will need to go back and apply the remedies to the previous cycle. See the instructions for re-running the previous cycle

NEMO instability

Instabilities in the ocean will appear as:

 stpctl: the zonal velocity is larger than 20 m/s
  ====== 
 kt= 87159 max abs(U):  6.0407E+07, i j k:  2914 3389    8

...

  ===>>> : E R R O R
         ===========

 MPPSTOP
 NEMO abort from dia_wri_state

Re-run the cycle with a reduced timestep in the ocean. 

Once the failed task completes successfully, revert back to the original settings:

UM instability

Atmosphere instabilities usually appear as:

Error message: North/South halos too small for advection

Perturb the atmos dump:

The script leaves a copy of the original and perturbed dumps as .orig and .perturb It also sets PYTHONHASHSEED=0 to create a deterministic perturbation. 

Once the script has completed and generated a new dump, re-trigger the failed task.

CICE instability

CICE instabilities appear as:

ABORT: Global i and j: 3347,  3604
ABORT:  Lat, Lon: 83.826494669668421,  73.000001621592844
ABORT: aice: 0.9873568057277784
...
About to call abort ice with: Vertical thermo error
ice: Vertical thermo error

In this case, reduce the ocean timestep (see above).