Merge pull request #276 from argonne-lcf/polaris

Update filesystems options for ALCF systems
argonne-lcf · Oct 26, 2022 · 2b63d83 · 2b63d83
2 parents 1f8405e + 93ff90b
commit 2b63d83
Show file tree

Hide file tree

Showing 7 changed files with 10 additions and 48 deletions.
diff --git a/balsam/__init__.py b/balsam/__init__.py
@@ -1,4 +1,4 @@
 from balsam.util import config_root_logger
 
-__version__ = "0.7.0.a17"
+__version__ = "0.7.0.a18"
 config_root_logger()
diff --git a/balsam/_api/models.py b/balsam/_api/models.py
@@ -1,5 +1,5 @@
-# This file was auto-generated via /Users/misha/workflow/balsam/.venv/bin/python balsam/schemas/api_generator.py
-# [git rev 2e64e31]
+# This file was auto-generated via /Users/turam/opt/miniconda3/bin/python balsam/schemas/api_generator.py
+# [git rev 8578c92]
 # Do *not* make changes to the API by changing this file!
 
 import datetime
@@ -765,7 +765,7 @@ class BatchJob(balsam._api.bases.BatchJobBase):
     job_mode = Field[balsam.schemas.batchjob.JobMode]()
     optional_params = Field[typing.Dict[str, str]]()
     filter_tags = Field[typing.Dict[str, str]]()
-    partitions = Field[typing.Optional[typing.List[balsam.schemas.batchjob.BatchJobPartition]]]()
+    partitions = Field[Optional[typing.Union[typing.List[balsam.schemas.batchjob.BatchJobPartition], None]]]()
     site_id = Field[int]()
     project = Field[str]()
     queue = Field[str]()
@@ -786,7 +786,7 @@ def __init__(
         queue: str,
         optional_params: Optional[typing.Dict[str, str]] = None,
         filter_tags: Optional[typing.Dict[str, str]] = None,
-        partitions: Optional[typing.Optional[typing.List[balsam.schemas.batchjob.BatchJobPartition]]] = None,
+        partitions: Optional[typing.Union[typing.List[balsam.schemas.batchjob.BatchJobPartition], None]] = None,
         **kwargs: Any,
     ) -> None:
         """
@@ -918,7 +918,7 @@ def create(
         queue: str,
         optional_params: Optional[typing.Dict[str, str]] = None,
         filter_tags: Optional[typing.Dict[str, str]] = None,
-        partitions: Optional[typing.Optional[typing.List[balsam.schemas.batchjob.BatchJobPartition]]] = None,
+        partitions: Optional[typing.Union[typing.List[balsam.schemas.batchjob.BatchJobPartition], None]] = None,
     ) -> BatchJob:
         """
         Create a new BatchJob object and save it to the API in one step.

diff --git a/balsam/cmdline/job.py b/balsam/cmdline/job.py
@@ -216,6 +216,7 @@ def count_by_state(job_qs: "JobQuery", verbose: bool) -> None:
 def list_verbose(job_qs: "JobQuery") -> None:
     for job in job_qs:
         click.echo(yaml.dump(job.display_dict(), sort_keys=False, indent=4))
+        click.echo(f"deserialized parameters: {str(job.get_parameters())}")
         click.echo("---\n")
 
 

diff --git a/balsam/config/defaults/alcf_cooley/job-template.sh b/balsam/config/defaults/alcf_cooley/job-template.sh
@@ -3,7 +3,7 @@
 #COBALT -n {{ num_nodes }}
 #COBALT -q {{ queue }}
 #COBALT -t {{ wall_time_min }}
-#COBALT --attrs pubnet
+#COBALT --attrs pubnet:filesystems=home,grand,eagle,theta-fs0
 
 # Uncomment this if the server is on an external network
 # (Note that https_proxy is set to use an `http://` protocol!

diff --git a/balsam/config/defaults/alcf_polaris/job-template.sh b/balsam/config/defaults/alcf_polaris/job-template.sh
@@ -4,49 +4,10 @@
 #PBS -l filesystems=home:grand:eagle
 #PBS -A {{ project }}
 #PBS -q {{ queue }}
-# #COBALT --attrs pubnet=true:enable_ssh=1:{% if optional_params.get("mig_count") %}mig-mode=true{% endif %}
 
 export http_proxy=http://proxy:3128
 export https_proxy=http://proxy:3128
 
-{% if optional_params.mig_count == "2" %}
-    cgi=9
-    mig_count=2
-    echo "Creating 2x MIG 3g.20gb (ID 9)"
-{% elif optional_params.mig_count == "3" %}
-    cgi=14
-    mig_count=3
-    echo "Creating 3x MIG 2g.10gb (ID 14)"
-{% elif optional_params.mig_count == "7" %}
-    cgi=19
-    mig_count=7
-    echo "Creating 7x MIG 1g.5gb (ID 19)"
-{% else %}
-    mig_count=0
-    echo "Not using MIG"
-{% endif %}
-
-if [ "$mig_count" -gt "0" ]
-then
-    # Create MIG Compute Instances
-    for i in $(seq 1 $mig_count)
-    do
-        mpirun -hostfile $COBALT_NODEFILE \
-        -n {{ num_nodes }} -npernode 1 \
-        nvidia-smi_mig -cgi "$cgi" -C
-    done
-
-    # Record Instance IDs in local /var/tmp:
-    for host in $(cat $COBALT_NODEFILE)
-    do
-        gpu_file="/var/tmp/balsam-$host-gpulist.txt"
-        mpirun -hostfile $COBALT_NODEFILE \
-        --host $host -n 1 nvidia-smi -L > $gpu_file
-        echo "Recorded GPU list for $host in $gpu_file"
-        cat $gpu_file
-    done
-fi
-
 #remove export PMI_NO_FORK=1
 export BALSAM_SITE_PATH={{balsam_site_path}}
 cd $BALSAM_SITE_PATH

diff --git a/balsam/config/defaults/alcf_theta/job-template.sh b/balsam/config/defaults/alcf_theta/job-template.sh
@@ -3,7 +3,7 @@
 #COBALT -n {{ num_nodes }}
 #COBALT -q {{ queue }}
 #COBALT -t {{ wall_time_min }}
-#COBALT --attrs ssds=required:ssd_size=128
+#COBALT --attrs ssds=required:ssd_size=128:filesystems=home,grand,eagle,theta-fs0
 
 
 {% if optional_params.get("singularity_prime_cache") %}

diff --git a/balsam/config/defaults/alcf_thetagpu/job-template.sh b/balsam/config/defaults/alcf_thetagpu/job-template.sh
@@ -3,7 +3,7 @@
 #COBALT -n {{ num_nodes }}
 #COBALT -q {{ queue }}
 #COBALT -t {{ wall_time_min }}
-#COBALT --attrs pubnet=true:enable_ssh=1:{% if optional_params.get("mig_count") %}mig-mode=true{% endif %}
+#COBALT --attrs pubnet=true:enable_ssh=1:filesystems=home,grand,eagle,theta-fs0:{% if optional_params.get("mig_count") %}mig-mode=true{% endif %}
 
 export https_proxy=http://theta-proxy.tmi.alcf.anl.gov:3128