When running legacyhalos-mpi, the unfinished coadds and custom-coadds are divided equally to all the available MPI ranks. The right thing to do is to use the number of CCDs as a metric for the "size" of each job and divide the work that way.
In #116 we use the mosaic diameter (a proxy for the memory / compute needs) to distribute the sample to the MPI tasks. This code is in legacyhalos.LSLGA but will be inherited by the other projects.
When running
legacyhalos-mpi
, the unfinishedcoadds
andcustom-coadds
are divided equally to all the available MPI ranks. The right thing to do is to use the number of CCDs as a metric for the "size" of each job and divide the work that way.