lsst-ctrl-bps-htcondor 29.2025.2100__tar.gz → 29.2025.2200__tar.gz

{lsst_ctrl_bps_htcondor-29.2025.2100/python/lsst_ctrl_bps_htcondor.egg-info → lsst_ctrl_bps_htcondor-29.2025.2200}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lsst-ctrl-bps-htcondor
-Version: 29.2025.2100
+Version: 29.2025.2200
 Summary: HTCondor plugin for lsst-ctrl-bps.
 Author-email: Rubin Observatory Data Management <dm-admin@lists.lsst.org>
 License: BSD 3-Clause License

{lsst_ctrl_bps_htcondor-29.2025.2100 → lsst_ctrl_bps_htcondor-29.2025.2200}/doc/lsst.ctrl.bps.htcondor/userguide.rst

@@ -148,11 +148,28 @@ See `bps submit`_.
 
 .. Describe any plugin specific aspects of a submission below if any.
 
-.. _htc-plugin-report:
+
+.. _htc-plugin-status:
 
 Checking status
 ---------------
 
+See `bps status`_.
+
+The plugin can take either the HTCondor ID (as shown in ``bps report`` or
+``condor_q``) or the submit path.
+
+For not-completed workflows, the speed of using the ID can depend on
+whether on the same submit machine (i.e., local schedd) or not and how
+busy the schedd machines are.  For completed workflows, using the ID
+may not work if the HTCondor logs have rolled over between the time of
+completion and time of the status command.
+
+.. _htc-plugin-report:
+
+Printing a report
+-----------------
+
 See `bps report`_.
 
 .. Describe any plugin specific aspects of checking a submission status below
@@ -555,7 +572,8 @@ complete your run.
 .. _HTCondor: https://htcondor.readthedocs.io/en/latest/
 .. _Slurm: https://slurm.schedmd.com/overview.html
 .. _bps cancel: https://pipelines.lsst.io/v/weekly/modules/lsst.ctrl.bps/quickstart.html#canceling-submitted-jobs
-.. _bps report: https://pipelines.lsst.io/v/weekly/modules/lsst.ctrl.bps/quickstart.html#checking-status
+.. _bps status: https://pipelines.lsst.io/v/weekly/modules/lsst.ctrl.bps/quickstart.html#checking-status
+.. _bps report: https://pipelines.lsst.io/v/weekly/modules/lsst.ctrl.bps/quickstart.html#printing-a-report
 .. _bps restart: https://pipelines.lsst.io/v/weekly/modules/lsst.ctrl.bps/quickstart.html#restarting-a-failed-run
 .. _bps submit: https://pipelines.lsst.io/v/weekly/modules/lsst.ctrl.bps/quickstart.html#submitting-a-run
 .. _ctrl_bps: https://github.com/lsst/ctrl_bps

{lsst_ctrl_bps_htcondor-29.2025.2100 → lsst_ctrl_bps_htcondor-29.2025.2200}/python/lsst/ctrl/bps/htcondor/htcondor_service.py

@@ -401,6 +401,54 @@ class HTCondorService(BaseWmsService):
         _LOG.debug("job_ids = %s", job_ids)
         return job_ids
 
+    def get_status(
+        self,
+        wms_workflow_id: str,
+        hist: float = 1,
+        is_global: bool = False,
+    ) -> tuple[WmsStates, str]:
+        """Return status of run based upon given constraints.
+
+        Parameters
+        ----------
+        wms_workflow_id : `str`
+            Limit to specific run based on id (queue id or path).
+        hist : `float`, optional
+            Limit history search to this many days. Defaults to 1.
+        is_global : `bool`, optional
+            If set, all job queues (and their histories) will be queried for
+            job information. Defaults to False which means that only the local
+            job queue will be queried.
+
+        Returns
+        -------
+        state : `lsst.ctrl.bps.WmsStates`
+            Status of single run from given information.
+        message : `str`
+            Extra message for status command to print.  This could be pointers
+            to documentation or to WMS specific commands.
+        """
+        _LOG.debug("get_status: id=%s, hist=%s, is_global=%s", wms_workflow_id, hist, is_global)
+
+        id_type = _wms_id_type(wms_workflow_id)
+        _LOG.debug("id_type = %s", id_type.name)
+
+        if id_type == WmsIdType.LOCAL:
+            schedulers = _locate_schedds(locate_all=is_global)
+            _LOG.debug("schedulers = %s", schedulers)
+            state, message = _get_status_from_id(wms_workflow_id, hist, schedds=schedulers)
+        elif id_type == WmsIdType.GLOBAL:
+            schedulers = _locate_schedds(locate_all=True)
+            _LOG.debug("schedulers = %s", schedulers)
+            state, message = _get_status_from_id(wms_workflow_id, hist, schedds=schedulers)
+        elif id_type == WmsIdType.PATH:
+            state, message = _get_status_from_path(wms_workflow_id)
+        else:
+            state, message = WmsStates.UNKNOWN, "Invalid job id"
+        _LOG.debug("state: %s, %s", state, message)
+
+        return state, message
+
     def report(
         self,
         wms_workflow_id=None,
@@ -1014,6 +1062,77 @@ def _handle_job_inputs(generic_workflow: GenericWorkflow, job_name: str, use_sha
     return htc_commands
 
 
+def _get_status_from_id(
+    wms_workflow_id: str, hist: float, schedds: dict[str, htcondor.Schedd]
+) -> tuple[WmsStates, str]:
+    """Gather run information using workflow id.
+
+    Parameters
+    ----------
+    wms_workflow_id : `str`
+        Limit to specific run based on id.
+    hist : `float`
+        Limit history search to this many days.
+    schedds : `dict` [ `str`, `htcondor.Schedd` ]
+        HTCondor schedulers which to query for job information. If empty
+        dictionary, all queries will be run against the local scheduler only.
+
+    Returns
+    -------
+    state : `lsst.ctrl.bps.WmsStates`
+        Status for the corresponding run.
+    message : `str`
+        Message with extra error information.
+    """
+    _LOG.debug("_get_status_from_id: id=%s, hist=%s, schedds=%s", wms_workflow_id, hist, schedds)
+
+    message = ""
+
+    # Collect information about the job by querying HTCondor schedd and
+    # HTCondor history.
+    schedd_dag_info = _get_info_from_schedd(wms_workflow_id, hist, schedds)
+    if len(schedd_dag_info) == 1:
+        schedd_name = next(iter(schedd_dag_info))
+        dag_id = next(iter(schedd_dag_info[schedd_name]))
+        dag_ad = schedd_dag_info[schedd_name][dag_id]
+        state = _htc_status_to_wms_state(dag_ad)
+    else:
+        state = WmsStates.UNKNOWN
+        message = f"DAGMan job {wms_workflow_id} not found in queue or history.  Check id or try path."
+    return state, message
+
+
+def _get_status_from_path(wms_path: str | os.PathLike) -> tuple[WmsStates, str]:
+    """Gather run status from a given run directory.
+
+    Parameters
+    ----------
+    wms_path : `str` | `os.PathLike`
+        The directory containing the submit side files (e.g., HTCondor files).
+
+    Returns
+    -------
+    state : `lsst.ctrl.bps.WmsStates`
+        Status for the run.
+    message : `str`
+        Message to be printed.
+    """
+    wms_path = Path(wms_path).resolve()
+    message = ""
+    try:
+        wms_workflow_id, dag_ad = read_dag_log(wms_path)
+    except FileNotFoundError:
+        wms_workflow_id = MISSING_ID
+        message = f"DAGMan log not found in {wms_path}.  Check path."
+
+    if wms_workflow_id == MISSING_ID:
+        state = WmsStates.UNKNOWN
+    else:
+        state = _htc_status_to_wms_state(dag_ad[wms_workflow_id])
+
+    return state, message
+
+
 def _report_from_path(wms_path):
     """Gather run information from a given run directory.
 
@@ -1139,11 +1258,11 @@ def _get_info_from_schedd(
     ----------
     wms_workflow_id : `str`
         Limit to specific run based on id.
-    hist : `int`
+    hist : `float`
         Limit history search to this many days.
-    schedds : `dict` [ `str`, `htcondor.Schedd` ], optional
-        HTCondor schedulers which to query for job information. If None
-        (default), all queries will be run against the local scheduler only.
+    schedds : `dict` [ `str`, `htcondor.Schedd` ]
+        HTCondor schedulers which to query for job information. If empty
+        dictionary, all queries will be run against the local scheduler only.
 
     Returns
     -------
@@ -1152,6 +1271,8 @@ def _get_info_from_schedd(
         Scheduler, local HTCondor job ids are mapped to their respective
         classads.
     """
+    _LOG.debug("_get_info_from_schedd: id=%s, hist=%s, schedds=%s", wms_workflow_id, hist, schedds)
+
     dag_constraint = 'regexp("dagman$", Cmd)'
     try:
         cluster_id = int(float(wms_workflow_id))

{lsst_ctrl_bps_htcondor-29.2025.2100 → lsst_ctrl_bps_htcondor-29.2025.2200}/python/lsst/ctrl/bps/htcondor/lssthtc.py

@@ -1376,6 +1376,7 @@ def condor_search(constraint=None, hist=None, schedds=None):
 
     job_info = condor_q(constraint=constraint, schedds=schedds)
     if hist is not None:
+        _LOG.debug("Searching history going back %s days", hist)
         epoch = (datetime.now() - timedelta(days=hist)).timestamp()
         constraint += f" && (CompletionDate >= {epoch} || JobFinishedHookDone >= {epoch})"
         hist_info = condor_history(constraint, schedds=schedds)

lsst_ctrl_bps_htcondor-29.2025.2200/python/lsst/ctrl/bps/htcondor/version.py

@@ -0,0 +1,2 @@
+__all__ = ["__version__"]
+__version__ = "29.2025.2200"

{lsst_ctrl_bps_htcondor-29.2025.2100 → lsst_ctrl_bps_htcondor-29.2025.2200/python/lsst_ctrl_bps_htcondor.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lsst-ctrl-bps-htcondor
-Version: 29.2025.2100
+Version: 29.2025.2200
 Summary: HTCondor plugin for lsst-ctrl-bps.
 Author-email: Rubin Observatory Data Management <dm-admin@lists.lsst.org>
 License: BSD 3-Clause License

{lsst_ctrl_bps_htcondor-29.2025.2100 → lsst_ctrl_bps_htcondor-29.2025.2200}/tests/test_htcondor_service.py

@@ -94,8 +94,8 @@ class HTCondorServiceTestCase(unittest.TestCase):
         self.assertEqual(self.service.defaults_uri, HTC_DEFAULTS_URI)
         self.assertFalse(self.service.defaults_uri.isdir())
 
-    @unittest.mock.patch.object(htcondor.Collector, "locate", return_value=LOCATE_SUCCESS)
     @unittest.mock.patch.object(htcondor.SecMan, "ping", return_value=PING_SUCCESS)
+    @unittest.mock.patch.object(htcondor.Collector, "locate", return_value=LOCATE_SUCCESS)
     def testPingSuccess(self, mock_locate, mock_ping):
         status, message = self.service.ping(None)
         self.assertEqual(status, 0)
@@ -116,6 +116,71 @@ class HTCondorServiceTestCase(unittest.TestCase):
             self.assertEqual(status, 1)
             self.assertEqual(message, "Permission problem with Schedd service.")
 
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._get_status_from_id")
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._locate_schedds")
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._wms_id_type")
+    def testGetStatusLocal(self, mock_type, mock_locate, mock_status):
+        mock_type.return_value = htcondor_service.WmsIdType.LOCAL
+        mock_locate.return_value = {}
+        mock_status.return_value = (WmsStates.RUNNING, "")
+
+        fake_id = "100"
+        state, message = self.service.get_status(fake_id)
+
+        mock_type.assert_called_once_with(fake_id)
+        mock_locate.assert_called_once_with(locate_all=False)
+        mock_status.assert_called_once_with(fake_id, 1, schedds={})
+
+        self.assertEqual(state, WmsStates.RUNNING)
+        self.assertEqual(message, "")
+
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._get_status_from_id")
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._locate_schedds")
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._wms_id_type")
+    def testGetStatusGlobal(self, mock_type, mock_locate, mock_status):
+        mock_type.return_value = htcondor_service.WmsIdType.GLOBAL
+        mock_locate.return_value = {}
+        fake_message = ""
+        mock_status.return_value = (WmsStates.RUNNING, fake_message)
+
+        fake_id = "100"
+        state, message = self.service.get_status(fake_id, 2)
+
+        mock_type.assert_called_once_with(fake_id)
+        mock_locate.assert_called_once_with(locate_all=True)
+        mock_status.assert_called_once_with(fake_id, 2, schedds={})
+
+        self.assertEqual(state, WmsStates.RUNNING)
+        self.assertEqual(message, fake_message)
+
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._get_status_from_path")
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._wms_id_type")
+    def testGetStatusPath(self, mock_type, mock_status):
+        fake_message = "fake message"
+        mock_type.return_value = htcondor_service.WmsIdType.PATH
+        mock_status.return_value = (WmsStates.FAILED, fake_message)
+
+        fake_id = "/fake/path"
+        state, message = self.service.get_status(fake_id)
+
+        mock_type.assert_called_once_with(fake_id)
+        mock_status.assert_called_once_with(fake_id)
+
+        self.assertEqual(state, WmsStates.FAILED)
+        self.assertEqual(message, fake_message)
+
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._wms_id_type")
+    def testGetStatusUnknownType(self, mock_type):
+        mock_type.return_value = htcondor_service.WmsIdType.UNKNOWN
+
+        fake_id = "100.0"
+        state, message = self.service.get_status(fake_id)
+
+        mock_type.assert_called_once_with(fake_id)
+
+        self.assertEqual(state, WmsStates.UNKNOWN)
+        self.assertEqual(message, "Invalid job id")
+
 
 class GetExitCodeSummaryTestCase(unittest.TestCase):
     """Test the function responsible for creating exit code summary."""
@@ -1290,5 +1355,68 @@ class CreatePeriodicRemoveExprTestCase(unittest.TestCase):
         self.assertEqual(results, truth)
 
 
+class GetStatusFromIdTestCase(unittest.TestCase):
+    """Test _get_status_from_id function."""
+
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._get_info_from_schedd")
+    def testNotFound(self, mock_get):
+        mock_get.return_value = {}
+
+        state, message = htcondor_service._get_status_from_id("100", 0, {})
+
+        mock_get.assert_called_once_with("100", 0, {})
+
+        self.assertEqual(state, WmsStates.UNKNOWN)
+        self.assertEqual(message, "DAGMan job 100 not found in queue or history.  Check id or try path.")
+
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._htc_status_to_wms_state")
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._get_info_from_schedd")
+    def testFound(self, mock_get, mock_conversion):
+        fake_id = "100.0"
+        dag_ads = {fake_id: {"JobStatus": lssthtc.JobStatus.RUNNING}}
+        mock_get.return_value = {"schedd1": dag_ads}
+        mock_conversion.return_value = WmsStates.RUNNING
+
+        state, message = htcondor_service._get_status_from_id(fake_id, 0, {})
+
+        mock_get.assert_called_once_with(fake_id, 0, {})
+        mock_conversion.assert_called_once_with(dag_ads[fake_id])
+
+        self.assertEqual(state, WmsStates.RUNNING)
+        self.assertEqual(message, "")
+
+
+class GetStatusFromPathTestCase(unittest.TestCase):
+    """Test _get_status_from_path function."""
+
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service.read_dag_log")
+    def testNoDagLog(self, mock_read):
+        mock_read.side_effect = FileNotFoundError
+
+        fake_path = "/fake/path"
+        state, message = htcondor_service._get_status_from_path(fake_path)
+
+        mock_read.assert_called_once_with(Path(fake_path))
+
+        self.assertEqual(state, WmsStates.UNKNOWN)
+        self.assertEqual(message, f"DAGMan log not found in {fake_path}.  Check path.")
+
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service._htc_status_to_wms_state")
+    @unittest.mock.patch("lsst.ctrl.bps.htcondor.htcondor_service.read_dag_log")
+    def testSuccess(self, mock_read, mock_conversion):
+        dag_ads = {"100.0": {"JobStatus": lssthtc.JobStatus.COMPLETED, "ExitBySignal": False, "ExitCode": 0}}
+        mock_read.return_value = ("100.0", dag_ads)
+        mock_conversion.return_value = WmsStates.SUCCEEDED
+
+        fake_path = "/fake/path"
+        state, message = htcondor_service._get_status_from_path(fake_path)
+
+        mock_read.assert_called_once_with(Path(fake_path))
+        mock_conversion.assert_called_once_with(dag_ads["100.0"])
+
+        self.assertEqual(state, WmsStates.SUCCEEDED)
+        self.assertEqual(message, "")
+
+
 if __name__ == "__main__":
     unittest.main()

lsst_ctrl_bps_htcondor-29.2025.2100/python/lsst/ctrl/bps/htcondor/version.py

@@ -1,2 +0,0 @@
-__all__ = ["__version__"]
-__version__ = "29.2025.2100"