Various other slurm/sockeye preset fixes

Julia-Tempering · Nov 25, 2023 · 7b615b8 · 7b615b8
1 parent 8c54c01
commit 7b615b8
Show file tree

Hide file tree

Showing 4 changed files with 13 additions and 19 deletions.
diff --git a/src/submission/MPI.jl b/src/submission/MPI.jl
@@ -109,21 +109,22 @@ function mpi_submission_script(exec_folder, mpi_submission::MPI, julia_cmd)
     #    MethodError(f=Core.Compiler.widenconst, args=(Symbol("#342"),), world=0x0000000000001342)
     export JULIA_PKG_PRECOMPILE_AUTO=0
 
-    mpiexec $(mpi_submission.mpiexec_args) --merge-stderr-to-stdout --output-filename $exec_folder $julia_cmd_str
+    mpiexec $(cmd_to_string(mpi_submission.mpiexec_args)) --merge-stderr-to-stdout --output-filename $exec_folder $julia_cmd_str
     """
     script_path = "$exec_folder/.submission_script.sh"
     write(script_path, code)
     return script_path
 end
 
+cmd_to_string(cmd::Cmd) = "$cmd"[2:(end-1)]
 
 # Internal: "rosetta stone" of submission commands
 const _rosetta = (;
     queue_concept = [:submit,   :del,     :directive, :job_name,    :output_file,   :error_file,    :submit_dir,            :job_status,    :job_status_all,    :ncpu_info],
 
     # tested:
-    pbs           = [`qsub`,    `qdel`,   "#PBS",     "-N ",        "-o ",          "-e ",          "\$PBS_O_WORKDIR",      `qstat -x`,     `qstat -u`,         `pbsnodes -aSj -F dsv`],
-    slurm         = [`sbatch`,  `scancel`,"#SBATCH",  "--job-name=","-o ",          "-e ",          "\$SLURM_SUBMIT_DIR",   `squeue --job`, `squeue -u`,        `sinfo`],
+    pbs           = [`qsub`,    `qdel`,   "#PBS",     "-N ",        "-o ",          "-e ",          "\$PBS_O_WORKDIR",      `qstat -x`,     `qstat -u`,         `pbsnodes`],
+    slurm         = [`sbatch`,  `scancel`,"#SBATCH",  "--job-name=","-o ",          "-e ",          "\$SLURM_SUBMIT_DIR",   `squeue --job`, `squeue -u`,        `sinfo -o%C`],
 
     # not yet tested:
     lsf           = [`bsub`,    `bkill`,  "#BSUB",    "-J ",        "-o ",          "-e ",          "\$LSB_SUBCWD",         `bjobs`,        `bjobs -u`,         `bhosts`],

diff --git a/src/submission/presets.jl b/src/submission/presets.jl
@@ -14,7 +14,10 @@ setup_mpi_sockeye(my_user_allocation_code) =
     setup_mpi(
         submission_system = :slurm,
         environment_modules = ["gcc", "openmpi", "git"],
-        add_to_submission = ["#PBS -A $my_user_allocation_code"], 
+        add_to_submission = [
+            "#SBATCH -A $my_user_allocation_code"
+            "#SBATCH --nodes=1-10000"  # required by cluster
+        ], 
         library_name = "/arc/software/spack-2023/opt/spack/linux-centos7-skylake_avx512/gcc-9.4.0/openmpi-4.1.1-d7o6cdvp67ngi5c5wdcw2qyjyseq3l3o/lib/libmpi"
     )
 

diff --git a/src/submission/submission_utils.jl b/src/submission/submission_utils.jl
@@ -35,20 +35,9 @@ function queue_status()
 end
 
 function queue_ncpus_free()
-    mpi_settings = load_mpi_settings()
-    @assert mpi_settings.submission_system == :pbs "Feature only supported on PBS at the moment"
     r = rosetta()
-    n = 0
-    for line in readlines(`$(r.ncpu_info)`)
-        for item in eachsplit(line, "|")
-            m = match(r"ncpus[(]f[/]t[)][=]([0-9]+)[/].*", item)
-            if m !== nothing
-                suffix = m.captures[1]
-                n += parse(Int, suffix)
-            end
-        end
-    end
-    return n
+    run(`$(r.ncpu_info)`)
+    return nothing
 end
 
 """ 
@@ -59,7 +48,7 @@ Instruct the scheduler to cancel or kill a job.
 function kill_job(result::Result) 
     r = rosetta()
     exec_folder = result.exec_folder 
-    submission_code = readline("$exec_folder/info/submission_output.txt")
+    submission_code = queue_code(result)
     run(`$(r.del) $submission_code`)
     return nothing
 end

diff --git a/test/test_stan.jl b/test/test_stan.jl
@@ -22,7 +22,8 @@ if !is_windows_in_CI()
         pigeons(target = Pigeons.stan_funnel(1), record = [online], n_chains = 1, n_rounds = 5, explorer = SliceSampler())
 
         # check we get reasonable accept on one real example 
-        pt = pigeons(target = pp.posterior, reference = pp.prior, n_rounds = 5, n_chains = 2)
+        post_prior = Pigeons.stan_mRNA_post_prior_pair()
+        pt = pigeons(target = post_prior.posterior, reference = post_prior.prior, n_rounds = 5, n_chains = 2)
         @test minimum(Pigeons.explorer_mh_prs(pt)) > 0.3
     end