🐛 FIX: max_memory_kb usage for direct scheduler (#4825)

In most schedulers the `metadata.options.max_memory_kb` input to a `CalcJob`
specifies the physical or "real" memory limit.
However, for the direct scheduler this imposed a virtual limit with `ulimit -v`.
Virtual memory is inequivalent to physical memory; it can often be much larger,
and one cannot always systematically map physical to virtual memory usage.
Therefore, in this commit: 

1. In all scheduler plugins `virtual_memory_kb` is correctly renamed to `physical_memory_kb`.
2. In the direct scheduler, the `max_memory_kb` limit is ignored,
   since no straightforward way exists to directly limit the physical memory usage.

Co-authored-by: Chris Sewell <chrisj_sewell@hotmail.com>

aiida/schedulers/plugins/direct.py

@@ -142,17 +142,8 @@ def _get_submit_script_header(self, job_tmpl):
                 lines.append('exec 2>&1')
 
         if job_tmpl.max_memory_kb:
-            try:
-                virtual_memory_kb = int(job_tmpl.max_memory_kb)
-                if virtual_memory_kb <= 0:
-                    raise ValueError
-            except ValueError:
-                raise ValueError(
-                    'max_memory_kb must be '
-                    "a positive integer (in kB)! It is instead '{}'"
-                    ''.format((job_tmpl.max_memory_kb))
-                )
-            lines.append(f'ulimit -v {virtual_memory_kb}')
+            self.logger.warning('Physical memory limiting is not supported by the direct scheduler.')
+
         if not job_tmpl.import_sys_environment:
             lines.append('env --ignore-environment \\')
 

aiida/schedulers/plugins/lsf.py

@@ -401,18 +401,18 @@ def _get_submit_script_header(self, job_tmpl):
         # TODO: check if this is the memory per node  # pylint: disable=fixme
         if job_tmpl.max_memory_kb:
             try:
-                virtual_memory_kb = int(job_tmpl.max_memory_kb)
-                if virtual_memory_kb <= 0:
+                physical_memory_kb = int(job_tmpl.max_memory_kb)
+                if physical_memory_kb <= 0:
                     raise ValueError
             except ValueError:
                 raise ValueError(
                     'max_memory_kb must be '
                     "a positive integer (in kB)! It is instead '{}'"
-                    ''.format((job_tmpl.MaxMemoryKb))
+                    ''.format((job_tmpl.max_memory_kb))
                 )
             # The -M option sets a per-process (soft) memory limit for all the
             # processes that belong to this job
-            lines.append(f'#BSUB -M {virtual_memory_kb}')
+            lines.append(f'#BSUB -M {physical_memory_kb}')
 
         if job_tmpl.custom_scheduler_commands:
             lines.append(job_tmpl.custom_scheduler_commands)

aiida/schedulers/plugins/pbspro.py

@@ -86,16 +86,16 @@ def _get_resource_lines(
 
         if max_memory_kb:
             try:
-                virtual_memory_kb = int(max_memory_kb)
-                if virtual_memory_kb <= 0:
+                physical_memory_kb = int(max_memory_kb)
+                if physical_memory_kb <= 0:
                     raise ValueError
             except ValueError:
                 raise ValueError(
                     'max_memory_kb must be '
                     "a positive integer (in kB)! It is instead '{}'"
                     ''.format((max_memory_kb))
                 )
-            select_string += f':mem={virtual_memory_kb}kb'
+            select_string += f':mem={physical_memory_kb}kb'
 
         return_lines.append(f'#PBS -l {select_string}')
         return return_lines

aiida/schedulers/plugins/slurm.py

@@ -382,18 +382,18 @@ def _get_submit_script_header(self, job_tmpl):
         # It is the memory per node, not per cpu!
         if job_tmpl.max_memory_kb:
             try:
-                virtual_memory_kb = int(job_tmpl.max_memory_kb)
-                if virtual_memory_kb <= 0:
+                physical_memory_kb = int(job_tmpl.max_memory_kb)
+                if physical_memory_kb <= 0:
                     raise ValueError
             except ValueError:
                 raise ValueError(
                     'max_memory_kb must be '
                     "a positive integer (in kB)! It is instead '{}'"
-                    ''.format((job_tmpl.MaxMemoryKb))
+                    ''.format((job_tmpl.max_memory_kb))
                 )
             # --mem: Specify the real memory required per node in MegaBytes.
             # --mem and  --mem-per-cpu  are  mutually exclusive.
-            lines.append(f'#SBATCH --mem={virtual_memory_kb // 1024}')
+            lines.append(f'#SBATCH --mem={physical_memory_kb // 1024}')
 
         if job_tmpl.custom_scheduler_commands:
             lines.append(job_tmpl.custom_scheduler_commands)

aiida/schedulers/plugins/torque.py

@@ -81,8 +81,8 @@ def _get_resource_lines(
 
         if max_memory_kb:
             try:
-                virtual_memory_kb = int(max_memory_kb)
-                if virtual_memory_kb <= 0:
+                physical_memory_kb = int(max_memory_kb)
+                if physical_memory_kb <= 0:
                     raise ValueError
             except ValueError:
                 raise ValueError(
@@ -92,7 +92,7 @@ def _get_resource_lines(
                 )
             # There is always something before, at least the total #
             # of nodes
-            select_string += f',mem={virtual_memory_kb}kb'
+            select_string += f',mem={physical_memory_kb}kb'
 
         return_lines.append(f'#PBS -l {select_string}')
         return return_lines