Hereby the PR that adds support for multiprocessing with MPI by using an MPIPool instead of the regular pool from dynasty. In this way, it can run on a cluster with multiple nodes. I added the mpi_pool, n_live, and resume parameters to fit.
I noticed that querying the Gaia catalog did not work in combination with MPI (or perhaps some other issue related to the cluster I was using), so I added the possibility to read this from a FITS file, which is created when running backtrack. To get this to work, I had to restructure some of the code a bit such that the class attributes are correctly set.
Let me know if you have any questions and/or feedback!
Thanks so much Tomas! This is very excellent. I am going to create a version (0.1) for the tool as it stands right now, then pull and create the next version (0.2) with your changes.
Hi @wbalmer,
Hereby the PR that adds support for multiprocessing with MPI by using an
MPIPool
instead of the regular pool fromdynasty
. In this way, it can run on a cluster with multiple nodes. I added thempi_pool
,n_live
, andresume
parameters tofit
.I noticed that querying the Gaia catalog did not work in combination with MPI (or perhaps some other issue related to the cluster I was using), so I added the possibility to read this from a FITS file, which is created when running
backtrack
. To get this to work, I had to restructure some of the code a bit such that the class attributes are correctly set.Let me know if you have any questions and/or feedback!