Update pb

Makefile

@@ -61,4 +61,10 @@ help:
 	@uv run python -c "import re; \
 	[[print(f'\033[36m{m[0]:<20}\033[0m {m[1]}') for m in re.findall(r'^([a-zA-Z_-]+):.*?## (.*)$$', open(makefile).read(), re.M)] for makefile in ('$(MAKEFILE_LIST)').strip().split()]"
 
+.PHONY: deploy
+deploy: ## Deploy to site
+	@echo "🚀 Deploying ComposeAPI to Site"
+	@kubectl kustomize kustomize/overlays/compose-api-rke | kubectl apply -f -
+
+
 .DEFAULT_GOAL := help

compose_api/api/routers/templates/copasi.jinja

@@ -2,7 +2,7 @@
     "state": {
         "time_course": {
             "_type": "step",
-            "address": "local:pbest.registry.simulators.copasi_process.CopasiUTCStep",
+            "address": "local:pbsim_common.simulators.copasi_process.CopasiUTCStep",
             "config": {
                 "model_source": "interesting.sbml",
                 "sim_start_time": {{ start_time }},

compose_api/api/routers/templates/tellurium.jinja

@@ -2,7 +2,7 @@
     "state": {
         "time_course": {
             "_type": "step",
-            "address": "local:pbest.registry.simulators.tellurium_process.TelluriumUTCStep",
+            "address": "local:pbsim_common.simulators.tellurium_process.TelluriumUTCStep",
             "config": {
                 "model_source": "interesting.sbml",
                 "sim_start_time": {{ start_time }},

compose_api/common/hpc/models.py

@@ -49,13 +49,16 @@ def get_sacct_format_string() -> str:
     def from_sacct_formatted_output(cls, line: str) -> "SlurmJob":
         # Split the line by delimiter
         fields = line.strip().split("|")
+        job_state = fields[4]
+        if "cancelled" in job_state.lower():  # Has 'cancelled by <User-ID>' which trips up mappings from string to Enum
+            job_state = "CANCELLED"  # so just set it to 'canceled'
         # Map fields to model attributes
         return cls(
             job_id=int(fields[0]),
             name=fields[1],
             account=fields[2],
             user_name=fields[3],
-            job_state=fields[4],
+            job_state=job_state,
             start_time=fields[5],
             end_time=fields[6],
             elapsed=fields[7],

compose_api/common/hpc/slurm_service.py

@@ -32,9 +32,7 @@ async def get_job_status_squeue(self, job_ids: list[int] | None = None) -> list[
         return slurm_jobs
 
     async def get_job_status_sacct(self, job_ids: list[int] | None = None) -> list[SlurmJob]:
-        command = (
-            f'sacct -u $USER --parsable --delimiter="|" --noheader --format="{SlurmJob.get_sacct_format_string()}"'
-        )
+        command = f'sacct -u $USER --parsable --allocations --delimiter="|" --noheader --format="{SlurmJob.get_sacct_format_string()}"'  # noqa: E501
         if job_ids is not None:
             job_ids_str = ",".join(map(str, job_ids)) if len(job_ids) > 1 else str(job_ids[0])
             command = command + f" -j {job_ids_str}"

compose_api/db/tables/hpc_tables.py

@@ -25,6 +25,11 @@ class JobStatusDB(enum.Enum):
     COMPLETED = "completed"
     FAILED = "failed"
     PENDING = "pending"
+    CANCELLED = "cancelled"
+    OUT_OF_MEMORY = "out_of_memory"
+    SUSPENDED = "suspended"
+    TIMEOUT = "timeout"
+    UNKNOWN = "unknown"
 
     def to_job_status(self) -> JobStatus:
         return JobStatus(self.value)

compose_api/simulation/handlers.py

@@ -126,7 +126,7 @@ async def run_curated_pbif(
     # Create OMEX with all necessary files
     with tempfile.TemporaryDirectory(delete=False) as tmp_dir:
         with zipfile.ZipFile(tmp_dir + "/input.omex", "w") as omex:
-            omex.writestr(data=templated_pbif, zinfo_or_arcname=f"{simulator_name}.pbif")
+            omex.writestr(data=templated_pbif, zinfo_or_arcname=f"{simulator_name}.pbg")
             if use_interesting:
                 omex.write(loaded_sbml.absolute(), arcname="interesting.sbml")
             else:

compose_api/simulation/job_monitor.py

@@ -100,15 +100,26 @@ async def update_running_jobs(self) -> None:
             slurm_job = slurm_job_map.get(hpc_run.slurmjobid)
             if not slurm_job or not slurm_job.job_state:
                 continue
-            new_status = JobStatus(slurm_job.job_state.lower())
-            if new_status == hpc_run.status:
-                logger.debug(f"HpcRun {hpc_run.database_id} is still running with status {new_status}")
-                continue
-            if hpc_run.status != new_status:
+            try:
+                new_status = JobStatus(slurm_job.job_state.lower())
+                if new_status == hpc_run.status:
+                    logger.debug(f"HpcRun {hpc_run.database_id} is still running with status {new_status}")
+                    continue
+                if hpc_run.status != new_status:
+                    await self.database_service.get_hpc_db().update_hpcrun_status(
+                        hpcrun_id=hpc_run.database_id, new_slurm_job=slurm_job
+                    )
+                    logger.info(f"Updated HpcRun {hpc_run.database_id} status to {new_status}")
+            except ValueError as e:
+                logger.exception(
+                    f"Error updating HpcRun {hpc_run.database_id} to status {slurm_job.job_state.lower()}."
+                    f" Setting status to UNKNOWN.",
+                    exc_info=e,
+                )
+                slurm_job.job_state = JobStatus.UNKNOWN.upper()
                 await self.database_service.get_hpc_db().update_hpcrun_status(
                     hpcrun_id=hpc_run.database_id, new_slurm_job=slurm_job
                 )
-                logger.info(f"Updated HpcRun {hpc_run.database_id} status to {new_status}")
 
             if slurm_job.job_id in self.internal_listeners:
                 self.internal_listeners[slurm_job.job_id].put_nowait(hpc_run)

compose_api/simulation/models.py

@@ -64,6 +64,11 @@ class JobStatus(StrEnum):
     COMPLETED = "completed"
     FAILED = "failed"
     PENDING = "pending"
+    CANCELLED = "cancelled"
+    OUT_OF_MEMORY = "out_of_memory"
+    SUSPENDED = "suspended"
+    TIMEOUT = "timeout"
+    UNKNOWN = "unknown"
 
 
 class HpcRun(BaseModel):

compose_api/simulation/simulation_service.py

@@ -163,8 +163,8 @@ async def build_container(self, simulator_version: SimulatorVersion, random_str:
                     #SBATCH --cpus-per-task 1
                     #SBATCH --mem=4GB
                     #SBATCH --nodelist={settings.slurm_build_node}
-                    #SBATCH --partition=general
-                    #SBATCH --qos=general
+                    #SBATCH --partition={settings.slurm_partition}
+                    #SBATCH --qos={settings.slurm_qos}
                     #SBATCH --output={get_slurm_log_file(slurm_job_name=slurm_job_name)}
 
                     set -e

compose_api/version.py

@@ -1 +1 @@
-__version__ = "0.3.5"
+__version__ = "0.3.8"

kustomize/config/compose-api-rke/shared.env

@@ -2,10 +2,10 @@
 SLURM_SUBMIT_HOST=hamantis.cam.uchc.edu
 SLURM_SUBMIT_USER=crbmapi
 #SLURM_SUBMIT_KEY_PATH=/run/secrets/slurmsubmitkeyfile/ssh-privatekey
-SLURM_PARTITION=general
-SLURM_NODE_LIST=mantis-039
-SLURM_BUILD_NODE=mantis-007
-SLURM_QOS=general
+SLURM_PARTITION=vcell
+SLURM_NODE_LIST=mantis-042
+SLURM_BUILD_NODE=mantis-042
+SLURM_QOS=vcell
 SLURM_LOG_BASE_PATH=/home/FCAM/crbmapi/compose_api/prod/htclogs
 SLURM_SBATCH_BASE_PATH=/home/FCAM/crbmapi/compose_api/prod/slurm_sbatch
 HPC_IMAGE_BASE_PATH=/home/FCAM/crbmapi/compose_api/prod/images

kustomize/overlays/compose-api-rke/kustomization.yaml

@@ -5,7 +5,7 @@ namespace: compose-api-rke
 
 images:
   - name: ghcr.io/biosimulations/compose-api
-    newTag: 0.3.4
+    newTag: 0.3.8
   - name: docker.io/library/mongo
     newTag: 8.0.12
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "compose-api"
-version = "0.3.5"
+version = "0.3.8"
 description = "An API server for reproducible biological workflows and cosimulations."
 authors = [{ name = "Jim Schaff", email = "[email protected]" }]
 requires-python = ">=3.13.2,<4.0"
@@ -36,7 +36,7 @@ dependencies = [
     "python-libsedml>=2.0.33",
     "biosimulators-utils>=0.2.3",
     "python-libsbml>=5.20.5",
-    "pbest==0.2.8",
+    "pbest==0.4.0",
 ]
 
 [project.urls]