torchx-nightly 2025.5.22__py3-none-any.whl → 2025.5.23__py3-none-any.whl

torchx/cli/cmd_list.py

@@ -21,6 +21,7 @@ logger: logging.Logger = logging.getLogger(__name__)
 
 HANDLE_HEADER = "APP HANDLE"
 STATUS_HEADER = "APP STATUS"
+NAME_HEADER = "APP NAME"
 
 
 class CmdList(SubCommand):
@@ -39,5 +40,7 @@ class CmdList(SubCommand):
     def run(self, args: argparse.Namespace) -> None:
         with get_runner() as runner:
             apps = runner.list(args.scheduler)
-            apps_data = [[app.app_handle, str(app.state)] for app in apps]
-            print(tabulate(apps_data, headers=[HANDLE_HEADER, STATUS_HEADER]))
+            apps_data = [[app.app_handle, app.name, str(app.state)] for app in apps]
+            print(
+                tabulate(apps_data, headers=[HANDLE_HEADER, NAME_HEADER, STATUS_HEADER])
+            )

torchx/schedulers/api.py

@@ -86,6 +86,7 @@ class ListAppResponse:
     app_id: str
     state: AppState
     app_handle: str = "<NOT_SET>"
+    name: str = ""
 
     # Implementing __hash__() makes ListAppResponse hashable which makes
     # it easier to check if a ListAppResponse object exists in a list of

torchx/schedulers/slurm_scheduler.py

@@ -482,6 +482,12 @@ class SlurmScheduler(
         subprocess.run(["scancel", app_id], check=True)
 
     def describe(self, app_id: str) -> Optional[DescribeAppResponse]:
+        try:
+            return self._describe_sacct(app_id)
+        except subprocess.CalledProcessError:
+            return self._describe_squeue(app_id)
+
+    def _describe_sacct(self, app_id: str) -> Optional[DescribeAppResponse]:
         p = subprocess.run(
             ["sacct", "--parsable2", "-j", app_id], stdout=subprocess.PIPE, check=True
         )
@@ -534,6 +540,48 @@ class SlurmScheduler(
             msg=msg,
         )
 
+    def _describe_squeue(self, app_id: str) -> Optional[DescribeAppResponse]:
+        p = subprocess.run(
+            ["squeue", "--json", "-j", app_id], stdout=subprocess.PIPE, check=True
+        )
+        output_json = json.loads(p.stdout.decode("utf-8"))
+
+        roles = {}
+        roles_statuses = {}
+        msg = ""
+        app_state = AppState.UNKNOWN
+        for job in output_json["jobs"]:
+            state = job["job_state"][0]
+            msg = state
+            state_enum = SLURM_STATES.get(state)
+            assert (
+                state_enum
+            ), f"failed to translate slurm state {state} to torchx state"
+            app_state = state_enum
+
+            role, _, replica_id = job["name"].rpartition("-")
+            if not replica_id or not role:
+                # name should always have at least 3 parts but sometimes sacct
+                # is slow to update
+                continue
+            if role not in roles:
+                roles[role] = Role(name=role, num_replicas=0, image="")
+                roles_statuses[role] = RoleStatus(role, [])
+            roles[role].num_replicas += 1
+            roles_statuses[role].replicas.append(
+                ReplicaStatus(
+                    id=int(replica_id), role=role, state=app_state, hostname=""
+                ),
+            )
+
+        return DescribeAppResponse(
+            app_id=app_id,
+            roles=list(roles.values()),
+            roles_statuses=list(roles_statuses.values()),
+            state=app_state,
+            msg=msg,
+        )
+
     def log_iter(
         self,
         app_id: str,
@@ -574,6 +622,12 @@ class SlurmScheduler(
         return iterator
 
     def list(self) -> List[ListAppResponse]:
+        try:
+            return self._list_sacct()
+        except subprocess.CalledProcessError:
+            return self._list_squeue()
+
+    def _list_sacct(self) -> List[ListAppResponse]:
         # By default sacct only returns accounting information of jobs launched on the current day
         # To return all jobs launched, set starttime to one second past unix epoch time
         # Starttime will be modified when listing jobs by timeframe is supported
@@ -590,6 +644,38 @@ class SlurmScheduler(
             for job in output_json["jobs"]
         ]
 
+    def _list_squeue(self) -> List[ListAppResponse]:
+        # if sacct isn't configured on the cluster, fallback to squeue which
+        # only has currently running jobs
+        p = subprocess.run(
+            ["squeue", "--json"],
+            stdout=subprocess.PIPE,
+            check=True,
+        )
+        output_json = json.loads(p.stdout.decode("utf-8"))
+
+        out = []
+        for job in output_json["jobs"]:
+            job_id = job["job_id"]
+
+            het_job_id = job.get("het_job_id")
+            if (
+                het_job_id
+                and het_job_id["set"]
+                and het_job_id["number"] != job_id
+                and het_job_id["number"] > 0
+            ):
+                continue
+
+            out.append(
+                ListAppResponse(
+                    app_id=str(job["job_id"]),
+                    state=SLURM_STATES[job["job_state"][0]],
+                    name=job["name"],
+                )
+            )
+        return out
+
 
 def create_scheduler(session_name: str, **kwargs: Any) -> SlurmScheduler:
     return SlurmScheduler(

{torchx_nightly-2025.5.22.dist-info → torchx_nightly-2025.5.23.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: torchx-nightly
-Version: 2025.5.22
+Version: 2025.5.23
 Summary: TorchX SDK and Components
 Home-page: https://github.com/pytorch/torchx
 Author: TorchX Devs

{torchx_nightly-2025.5.22.dist-info → torchx_nightly-2025.5.23.dist-info}/RECORD

@@ -14,7 +14,7 @@ torchx/cli/cmd_base.py,sha256=SdqMtqi04CEqnzcgcS35DbDbsBeMxSgEhfynfpIkMGk,790
 torchx/cli/cmd_cancel.py,sha256=NKfOCu_44Lch9vliGSQ0Uv6BVqpUqj7Tob652TI-ua4,835
 torchx/cli/cmd_configure.py,sha256=1kTv0qbsbV44So74plAySwWu56pQrqjhfW_kbfdC3Rw,1722
 torchx/cli/cmd_describe.py,sha256=E5disbHoKTsqYKp2s3DaFW9GDLCCOgdOc3pQoHKoyCs,1283
-torchx/cli/cmd_list.py,sha256=BVqHEW2oTEJ3GqcFK7c1K-i2R-DUjaXQ-WBr0meeIGM,1429
+torchx/cli/cmd_list.py,sha256=4Y1ZOq-kqJbztoBt56hAW_InJEaJuDAjpKWgMhBw4II,1507
 torchx/cli/cmd_log.py,sha256=v-EZYUDOcG95rEgTnrsmPJMUyxM9Mk8YFAJtUxtgViE,5475
 torchx/cli/cmd_run.py,sha256=4M1JJc7YmEa5T_2OFakCwCwiP0Ibpy-3zcLp1arrj9w,12203
 torchx/cli/cmd_runopts.py,sha256=NWZiP8XpQjfTDJgays2c6MgL_8wxFoeDge6NstaZdKk,1302
@@ -65,7 +65,7 @@ torchx/runtime/__init__.py,sha256=Wxje2BryzeQneFu5r6P9JJiEKG-_C9W1CcZ_JNrKT6g,59
 torchx/runtime/tracking/__init__.py,sha256=dYnAPnrXYREfPXkpHhdOFkcYIODWEbA13PdD-wLQYBo,3055
 torchx/runtime/tracking/api.py,sha256=SmUQyUKZqG3KlAhT7CJOGqRz1O274E4m63wQeOVq3CU,5472
 torchx/schedulers/__init__.py,sha256=gwy1opmKOPzQ_Lqh2GY0chYycLmdissLfd4846mPEMY,2334
-torchx/schedulers/api.py,sha256=glgViY2QvhnYbj9J0XtzpjY5XrG_UCzPcHU9CwFR98I,14598
+torchx/schedulers/api.py,sha256=zUlVtZ8gE4QoNTbd_xCGKQCmGS47jjT-vV-E9mdvEUc,14617
 torchx/schedulers/aws_batch_scheduler.py,sha256=h95d3OBhxkB7QJlJaDY3s1H7EG0eLXnCXxAPU8Ume3w,28130
 torchx/schedulers/aws_sagemaker_scheduler.py,sha256=spmcTEZ_o05pdTzpXr5gmOA-a9W0xH-YX6AioqX78l8,20950
 torchx/schedulers/devices.py,sha256=RjVcu22ZRl_9OKtOtmA1A3vNXgu2qD6A9ST0L0Hsg4I,1734
@@ -77,7 +77,7 @@ torchx/schedulers/kubernetes_scheduler.py,sha256=7AR3ccfta0NXqahxz9LVrv-vkdZnYTA
 torchx/schedulers/local_scheduler.py,sha256=JMSGAO9RXeUiEz8BOTA_EnHDOd065oJ_tyV1E__m3OQ,41882
 torchx/schedulers/lsf_scheduler.py,sha256=e6BmJC6dNNNzzwATgJu5Sq4HxAPw_hI3EJFRojzAMlE,17690
 torchx/schedulers/ray_scheduler.py,sha256=9Sqesw3aOw_Z0gua2TY3aYE3OJ9MCi75hqVl_RUQwQY,15750
-torchx/schedulers/slurm_scheduler.py,sha256=RC1ze2w0oaoQDLgercW7yHz1rGv5FVB6em4HYbLmQRg,19434
+torchx/schedulers/slurm_scheduler.py,sha256=g-FrtdUxErdtBE_NbRzNL7yxwKZDuSSWXbaSXGADhZM,22376
 torchx/schedulers/streams.py,sha256=8_SLezgnWgfv_zXUsJCUM34-h2dtv25NmZuxEwkzmxw,2007
 torchx/schedulers/ray/__init__.py,sha256=fE0IHi1JJpxsNVBNzWNee2thrNXFFRhY94c80RxNSIE,231
 torchx/schedulers/ray/ray_common.py,sha256=pyNYFvTKVwdjDAeCBNbPwAWwVNmlLOJWExfn90XY8u8,610
@@ -115,9 +115,9 @@ torchx/workspace/__init__.py,sha256=FqN8AN4VhR1C_SBY10MggQvNZmyanbbuPuE-JCjkyUY,
 torchx/workspace/api.py,sha256=PtDkGTC5lX03pRoYpuMz2KCmM1ZOycRP1UknqvNb97Y,6341
 torchx/workspace/dir_workspace.py,sha256=npNW_IjUZm_yS5r-8hrRkH46ndDd9a_eApT64m1S1T4,2268
 torchx/workspace/docker_workspace.py,sha256=PFu2KQNVC-0p2aKJ-W_BKA9ZOmXdCY2ABEkCExp3udQ,10269
-torchx_nightly-2025.5.22.dist-info/LICENSE,sha256=WVHfXhFC0Ia8LTKt_nJVYobdqTJVg_4J3Crrfm2A8KQ,1721
-torchx_nightly-2025.5.22.dist-info/METADATA,sha256=aF4BSx1lats2dcSuICxIzlDgQOXeVx4S-MOd1cjfSb0,6120
-torchx_nightly-2025.5.22.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
-torchx_nightly-2025.5.22.dist-info/entry_points.txt,sha256=T328AMXeKI3JZnnxfkEew2ZcMN1oQDtkXjMz7lkV-P4,169
-torchx_nightly-2025.5.22.dist-info/top_level.txt,sha256=pxew3bc2gsiViS0zADs0jb6kC5v8o_Yy_85fhHj_J1A,7
-torchx_nightly-2025.5.22.dist-info/RECORD,,
+torchx_nightly-2025.5.23.dist-info/LICENSE,sha256=WVHfXhFC0Ia8LTKt_nJVYobdqTJVg_4J3Crrfm2A8KQ,1721
+torchx_nightly-2025.5.23.dist-info/METADATA,sha256=ZZ_WpF3Xlg9RsqCW0cub5N4MEZiIMIfid9quYURz-us,6120
+torchx_nightly-2025.5.23.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
+torchx_nightly-2025.5.23.dist-info/entry_points.txt,sha256=T328AMXeKI3JZnnxfkEew2ZcMN1oQDtkXjMz7lkV-P4,169
+torchx_nightly-2025.5.23.dist-info/top_level.txt,sha256=pxew3bc2gsiViS0zADs0jb6kC5v8o_Yy_85fhHj_J1A,7
+torchx_nightly-2025.5.23.dist-info/RECORD,,