nodeocc 1.0.16__tar.gz → 1.0.18__tar.gz

{nodeocc-1.0.16 → nodeocc-1.0.18}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nodeocc
-Version: 1.0.16
+Version: 1.0.18
 Summary: TUI application for viewing the status of GPU allocations on a Slurm cluster
 Author: Matteo Boschini
 Maintainer: Lorenzo Bonicelli, Aniello Panariello

{nodeocc-1.0.16 → nodeocc-1.0.18}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "nodeocc"
-version = "1.0.16"
+version = "1.0.18"
 authors = [{ name = "Matteo Boschini" }]
 maintainers = [{ name = "Lorenzo Bonicelli" }, { name = "Aniello Panariello" }]
 description = "TUI application for viewing the status of GPU allocations on a Slurm cluster"

{nodeocc-1.0.16 → nodeocc-1.0.18}/src/controller/controller.py

@@ -70,7 +70,7 @@ JOB_LIMIT_PER_MSG = 60
 parser = argparse.ArgumentParser(description='Visualize slurm jobs')
 parser.add_argument('--debug', action='store_true', help='Enable logging')
 parser.add_argument('--master_only', action='store_true', help='Disable all prints - only run in background')
-parser.add_argument('--basepath', type=str, default='/nas/softechict-nas-2/mboschini/cool_scripts/new_nodeocc/', help='Base path for nodeocc. Must be readable by all users')
+parser.add_argument('--basepath', type=str, default='/tmp', help='Base path for nodeocc. Must be readable by all users')
 args = parser.parse_args()
 
 # check basepath is readable
@@ -136,7 +136,6 @@ def update_data_master(instance: Singleton):
     queue_jobs = sorted(queue_jobs, key=lambda x: x['priority'], reverse=True)
     maxlen = min(JOB_LIMIT_PER_MSG, len(running_jobs) + len(queue_jobs))
 
-
     queue_jobs, extra_queue_jobs = queue_jobs[:maxlen - len(running_jobs)], queue_jobs[maxlen - len(running_jobs):]
     cur_timestamp = time.time()
     msg = json.dumps({'inf': inf.to_nested_dict(), 'jobs': running_jobs + queue_jobs, 'ts': str(cur_timestamp)})
@@ -195,7 +194,7 @@ async def get_data_slave(instance):
                 # check if timestamp is the same
                 if float(new_ts['ts']) != orig_timestamp:
                     instance.timeme(f"Timestamp mismatch, skipping")
-                    return inf, jobs, avg_wait_time # return what we have
+                    return inf, jobs, avg_wait_time  # return what we have
 
                 msg_store[msg_idx] += new_jobs
 

{nodeocc-1.0.16 → nodeocc-1.0.18}/src/nodeocc.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nodeocc
-Version: 1.0.16
+Version: 1.0.18
 Summary: TUI application for viewing the status of GPU allocations on a Slurm cluster
 Author: Matteo Boschini
 Maintainer: Lorenzo Bonicelli, Aniello Panariello

{nodeocc-1.0.16 → nodeocc-1.0.18}/src/view/curses_multiwindow.py

@@ -92,16 +92,17 @@ class Singleton:
 
     def clean_port_files(self):
         nodename = socket.getfqdn().split('.')[0]
-        for f in os.listdir(self.basepath):
-            if nodename in f and 'master_' in f and f.endswith('.portfile') and is_file_writable_byall(os.path.join(self.basepath, f)):
+        bp = os.path.join(self.basepath, nodename)
+        for f in os.listdir(bp):
+            if 'master_' in f and f.endswith('.portfile') and is_file_writable_byall(os.path.join(bp, f)):
                 # read pid from file
-                pid = Path(os.path.join(self.basepath, f)).read_text()
+                pid = Path(os.path.join(bp, f)).read_text()
                 # assert no process is running
                 if psutil.pid_exists(int(pid)):
                     self.log(f"Process {pid} is still running, killing self")
                     return False
 
-                os.system(f"rm {os.path.join(self.basepath, f)}")
+                os.system(f"rm {os.path.join(bp, f)}")
         return True
 
     def __init__(self, args=None):
@@ -165,8 +166,9 @@ class Singleton:
         Check if a port file exists in the basepath and the file has 666 permissions
         """
         nodename = socket.getfqdn().split('.')[0]
-        portfiles = [os.path.join(self.basepath, f) for f in os.listdir(self.basepath)
-                     if f.endswith('.portfile') and 'master_' in f and nodename in f]
+        bp = os.path.join(self.basepath, nodename)
+        portfiles = [os.path.join(bp, f) for f in os.listdir(bp)
+                     if f.endswith('.portfile') and 'master_' in f]
         portfiles = [f for f in portfiles if is_file_readable_byall(f) and is_file_writable_byall(f)]
         return portfiles
 
@@ -197,7 +199,11 @@ class Singleton:
 
         try:
             # create file to store port with 666 permissions to file
-            filepath = Path(self.basepath, f'{nodename}_master_{self.port}.portfile')
+            # file is in folder basepath/nodename/master_{port}.portfile
+            bp = os.path.join(self.basepath, nodename)
+            if not os.path.exists(bp):
+                os.makedirs(bp, mode=0o777)
+            filepath = Path(os.path.join(bp, f"master_{self.port}.portfile"))
             filepath.write_text(str(self.pid))
             os.chmod(filepath, 0o666)