Add nice docstrings, fix tests

src/slurm_interface.jl

@@ -1,4 +1,29 @@
 
+"""
+    slurm_calibration(;
+        experiment_dir = dirname(Base.active_project()),
+        model_interface_path = joinpath(experiment_dir, "..", "..", "model_interface.jl"),
+        time_limit = "0:45:00",
+        ntasks = 1,
+        cpus_per_task = 1,
+        gpus_per_task = 0,
+        verbose = false,
+    )
+
+Runs a full calibration, scheduling forward model runs on the slurm cluster using `srun_model`.
+This function makes heavy assumptions and requires some setup.
+ - The correct project must be selected, and the observation map has already been `include`d
+ - The session is not running in an existing slurm job, and is running on the Resnick central cluster.
+
+Input arguments:
+ - `experiment_dir`: The directory storing relevant experiment information. (Default: dirname(Base.active_project()))
+ - `model_interface_path`: Model interface file to be included during the model run. (Default: joinpath(experiment_dir, "..", "..", "model_interface.jl"))
+ - `time_limit`: Slurm time limit
+ - `ntasks`: Slurm ntasks
+ - `cpus_per_task`: Slurm CPUs per task
+ - `gpus_per_task`: Slurm GPUs per task
+ - `verbose`: Turn on verbose model logging
+"""
 function slurm_calibration(;
     experiment_dir = dirname(Base.active_project()),
     model_interface_path = abspath(
@@ -37,7 +62,8 @@ function slurm_calibration(;
                 verbose,
             )
         end
-        handle_ensemble_procs(procs, iter, output_dir)
+
+        handle_ensemble_procs(procs, iter, output_dir, verbose)
         @info "Completed iteration $iter, updating ensemble"
         G_ensemble = observation_map(Val(Symbol(config.id)), iter)
         save_G_ensemble(config, iter, G_ensemble)
@@ -46,7 +72,13 @@ function slurm_calibration(;
     return eki
 end
 
-function handle_ensemble_procs(procs, iteration, output_dir)
+"""
+    handle_ensemble_procs(procs, iteration, output_dir, verbose)
+
+Helper function for `slurm_calibration`.
+    Handles the ensemble of processes running the forward model via slurm.
+"""
+function handle_ensemble_procs(procs, iteration, output_dir, verbose)
     # Initial try handles InterruptException
     try
         asyncmap(enumerate(procs)) do (member, p)
@@ -57,18 +89,17 @@ function handle_ensemble_procs(procs, iteration, output_dir)
             try
                 wait(p)
                 if p.exitcode != 0
-                    warn_on_member_error(member, member_log)
+                    warn_on_member_error(member, member_log, verbose)
                 end
             catch e
-                warn_on_member_error(member, member_log)
+                warn_on_member_error(member, member_log, verbose)
             end
         end
     catch e
-        e isa InterruptException && map(p -> kill(p), procs)
+        e isa InterruptException && foreach(kill, procs)
     end
     # Wait for processes to be killed
     sleep(0.25)
-
     exit_codes = map(x -> getproperty(x, :exitcode), procs)
     if !any(x -> x == 0, exit_codes)
         error("Full ensemble for iteration $iteration has failed. See model logs in $(path_to_iteration(output_dir, iteration)) for details.")
@@ -78,19 +109,32 @@ function handle_ensemble_procs(procs, iteration, output_dir)
 end
 
 function warn_on_member_error(member, member_log, verbose = false)
-    warn_str = if verbose
-        """
-        Ensemble member $member raised an error. See model log at $member_log for stacktrace:
-        $(replace(readchomp(member_log), "\\n" => "\n"))
-        """
-    else
-        """
-        Ensemble member $member raised an error. See model log at $member_log for stacktrace.
-        """
+    warn_str = "Ensemble member $member raised an error. See model log at $member_log for stacktrace"
+    if verbose
+        stacktrace = replace(readchomp(member_log), "\\n" => "\n")
+        warn_str = warn_str * ": \n$stacktrace"
     end
     @warn warn_str
 end
 
+"""
+    srun_model(;
+        output_dir,
+        iter,
+        member,
+        time_limit,
+        ntasks,
+        partition,
+        cpus_per_task,
+        gpus_per_task,
+        experiment_dir,
+        model_interface_path,
+        verbose,
+    )
+
+Runs a single forward model ensemble member. Constructs the `srun` command, then 
+runs it in a separate process.
+"""
 function srun_model(;
     output_dir,
     iter,

test/e2e_test.jl

@@ -35,7 +35,6 @@ experiment_config = ExperimentConfig(
     prior,
     output_dir,
     false,
-    false,
 )
 
 # Model interface