DIRAC 9.0.0a62__tar.gz → 9.0.0a63__tar.gz

{dirac-9.0.0a62/src/DIRAC.egg-info → dirac-9.0.0a63}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: DIRAC
-Version: 9.0.0a62
+Version: 9.0.0a63
 Summary: DIRAC is an interware, meaning a software framework for distributed computing.
 Home-page: https://github.com/DIRACGrid/DIRAC/
 License: GPL-3.0-only
@@ -45,6 +45,7 @@ Requires-Dist: typing_extensions>=4.3.0
 Requires-Dist: Authlib>=1.0.0.a2
 Requires-Dist: pyjwt
 Requires-Dist: dominate
+Requires-Dist: zstandard
 Provides-Extra: server
 Requires-Dist: CMRESHandler; extra == "server"
 Requires-Dist: opensearch-py; extra == "server"

{dirac-9.0.0a62 → dirac-9.0.0a63}/integration_tests.py

@@ -25,6 +25,9 @@ from typer import colors as c
 DEFAULT_HOST_OS = "el9"
 DEFAULT_MYSQL_VER = "mysql:8.4.4"
 DEFAULT_ES_VER = "opensearchproject/opensearch:2.18.0"
+# In MacOSX with Arm (MX), there's an issue with opensearch
+# You *must* set the ES_PLATFORM flag to `linux/arm64` to make it work.
+DEFAULT_ES_PLATFORM = "linux/amd64"
 DEFAULT_IAM_VER = "indigoiam/iam-login-service:v1.10.2"
 FEATURE_VARIABLES = {
     "DIRACOSVER": "master",
@@ -36,8 +39,12 @@ FEATURE_VARIABLES = {
     "INSTALLATION_BRANCH": "",
     "DEBUG": "Yes",
 }
-DIRACX_OPTIONS = ()
 DEFAULT_MODULES = {"DIRAC": Path(__file__).parent.absolute()}
+# All services that have a FutureClient, but we *explicitly* deactivate
+# (for example if we did not finish to develop it)
+DIRACX_DISABLED_SERVICES = [
+    "WorkloadManagement/JobMonitoring",
+]
 
 # Static configuration
 DB_USER = "Dirac"
@@ -71,6 +78,10 @@ LOG_LEVEL_MAP = {
 }
 LOG_PATTERN = re.compile(r"^[\d\-]{10} [\d:]{8} UTC [^\s]+ ([A-Z]+):")
 
+# In niche cases where we use MacOSX with Orbstack, some commands may not work with docker compose
+# If you're in that case, set in your environment `export DOCKER_COMPOSE_CMD="docker-compose"`
+DOCKER_COMPOSE_CMD = shlex.split(os.environ.get("DOCKER_COMPOSE_CMD", "docker compose"))
+
 
 class NaturalOrderGroup(typer.core.TyperGroup):
     """Group for showing subcommands in the correct order"""
@@ -140,6 +151,19 @@ Command completion of typer based scripts can be enabled by running:
 After restarting your terminal you command completion is available using:
 
   typer ./integration_tests.py run ...
+
+## DiracX
+
+If you want to activate DiracX, you have to set the flag TEST_DIRACX to "Yes".
+It will search for legacy adapted services (services with a future client activated)
+and do the necessary to make DIRAC work alongside DiracX.
+
+To deactivate a legacy adapted service (to pass CI for example), you have to add it in
+the `DIRACX_DISABLED_SERVICES` list. If you don't, the program will set this service to be used
+with DiracX, and if it is badly adapted, errors will be raised.
+
+> Note that you can provide a DiracX project (repository, branch) by building it and providing
+the dist folder to the prepare-environment command.
 """,
 )
 
@@ -193,8 +217,8 @@ def destroy():
     typer.secho("Shutting down and removing containers", err=True, fg=c.GREEN)
     with _gen_docker_compose(DEFAULT_MODULES) as docker_compose_fn:
         os.execvpe(
-            "docker",
-            ["docker", "compose", "-f", docker_compose_fn, "down", "--remove-orphans", "-t", "0", "--volumes"],
+            DOCKER_COMPOSE_CMD[0],
+            [*DOCKER_COMPOSE_CMD, "-f", docker_compose_fn, "down", "--remove-orphans", "-t", "0", "--volumes"],
             _make_env({}),
         )
 
@@ -253,7 +277,7 @@ def prepare_environment(
     typer.secho("Running docker compose to create containers", fg=c.GREEN)
     with _gen_docker_compose(modules, diracx_dist_dir=diracx_dist_dir) as docker_compose_fn:
         subprocess.run(
-            ["docker", "compose", "-f", docker_compose_fn, "up", "-d", "dirac-server", "dirac-client", "dirac-pilot"]
+            [*DOCKER_COMPOSE_CMD, "-f", docker_compose_fn, "up", "-d", "dirac-server", "dirac-client", "dirac-pilot"]
             + extra_services,
             check=True,
             env=docker_compose_env,
@@ -360,7 +384,7 @@ def prepare_environment(
         subStderr = open(docker_compose_fn_final / "stderr", "w")
 
         subprocess.Popen(
-            ["docker", "compose", "-f", docker_compose_fn_final / "docker-compose.yml", "up", "-d", "diracx"],
+            [*DOCKER_COMPOSE_CMD, "-f", docker_compose_fn_final / "docker-compose.yml", "up", "-d", "diracx"],
             env=docker_compose_env,
             stdin=None,
             stdout=subStdout,
@@ -569,7 +593,7 @@ def _gen_docker_compose(modules, *, diracx_dist_dir=None):
     # Load the docker compose configuration and mount the necessary volumes
     input_fn = Path(__file__).parent / "tests/CI/docker-compose.yml"
     docker_compose = yaml.safe_load(input_fn.read_text())
-    # diracx-wait-for-db needs the volume to be able to run the witing script
+    # diracx-wait-for-db needs the volume to be able to run the waiting script
     for ctn in ("dirac-server", "dirac-client", "dirac-pilot", "diracx-wait-for-db"):
         if "volumes" not in docker_compose["services"][ctn]:
             docker_compose["services"][ctn]["volumes"] = []
@@ -619,7 +643,7 @@ def _gen_docker_compose(modules, *, diracx_dist_dir=None):
 def _check_containers_running(*, is_up=True):
     with _gen_docker_compose(DEFAULT_MODULES) as docker_compose_fn:
         running_containers = subprocess.run(
-            ["docker", "compose", "-f", docker_compose_fn, "ps", "-q", "-a"],
+            [*DOCKER_COMPOSE_CMD, "-f", docker_compose_fn, "ps", "-q", "-a"],
             stdout=subprocess.PIPE,
             env=_make_env({}),
             # docker compose ps has a non-zero exit code when no containers are running
@@ -701,6 +725,7 @@ def _make_env(flags):
     else:
         env["MYSQL_ADMIN_COMMAND"] = "mysqladmin"
     env["ES_VER"] = flags.pop("ES_VER", DEFAULT_ES_VER)
+    env["ES_PLATFORM"] = flags.pop("ES_PLATFORM", DEFAULT_ES_PLATFORM)
     env["IAM_VER"] = flags.pop("IAM_VER", DEFAULT_IAM_VER)
     if "CVMFS_DIR" not in env or not Path(env["CVMFS_DIR"]).is_dir():
         typer.secho(f"CVMFS_DIR environment value: {env.get('CVMFS_DIR', 'NOT SET')}", fg=c.YELLOW)
@@ -1163,10 +1188,16 @@ def _make_config(modules, flags, release_var, editable):
             typer.secho(f"Required feature variable {key!r} is missing", err=True, fg=c.RED)
             raise typer.Exit(code=1)
 
-    # If we test DiracX, enable all the options
+    # If we test DiracX, add specific config
     if config["TEST_DIRACX"].lower() in ("yes", "true"):
-        for key in DIRACX_OPTIONS:
-            config[key] = "Yes"
+        if DIRACX_DISABLED_SERVICES:
+            # We link all disabled services
+            # config["DIRACX_DISABLED_SERVICES"] = "Service1 Service2 Service3 ..."
+            diracx_disabled_services = " ".join(DIRACX_DISABLED_SERVICES)
+
+            typer.secho(f"The following services won't be legacy adapted: {diracx_disabled_services}", fg="yellow")
+
+            config["DIRACX_DISABLED_SERVICES"] = diracx_disabled_services
 
     config["TESTREPO"] = [f"/home/dirac/LocalRepo/TestCode/{name}" for name in modules]
     config["ALTERNATIVE_MODULES"] = [f"/home/dirac/LocalRepo/ALTERNATIVE_MODULES/{name}" for name in modules]

{dirac-9.0.0a62 → dirac-9.0.0a63}/release.notes

@@ -1,3 +1,14 @@
+[v9.0.0a63]
+
+*WorkloadManagementSystem
+
+NEW: (#8276) Add zstandard compression support for sandbox files with backward compatibility
+CHANGE: (#8244) move the logic for jobs deleting and killing in an utility
+
+*Integration Tests
+
+FIX: (#8248) Add in CI and in `integration_test.py` legacy adaptors.
+
 [v9.0.0a62]
 
 *ConfigurationSystem

{dirac-9.0.0a62 → dirac-9.0.0a63}/setup.cfg

@@ -52,6 +52,7 @@ install_requires =
 	Authlib >=1.0.0.a2
 	pyjwt
 	dominate
+	zstandard
 zip_safe = False
 include_package_data = True
 
@@ -270,13 +271,10 @@ console_scripts =
 	dirac-stager-stage-files = DIRAC.StorageManagementSystem.scripts.dirac_stager_stage_files:main
 	dirac-production-runjoblocal = DIRAC.TransformationSystem.scripts.dirac_production_runjoblocal:main
 	dirac-transformation-add-files = DIRAC.TransformationSystem.scripts.dirac_transformation_add_files:main [admin]
-	dirac-transformation-archive = DIRAC.TransformationSystem.scripts.dirac_transformation_archive:main [admin]
-	dirac-transformation-clean = DIRAC.TransformationSystem.scripts.dirac_transformation_clean:main [admin]
 	dirac-transformation-cli = DIRAC.TransformationSystem.scripts.dirac_transformation_cli:main [admin]
 	dirac-transformation-get-files = DIRAC.TransformationSystem.scripts.dirac_transformation_get_files:main [admin]
 	dirac-transformation-information = DIRAC.TransformationSystem.scripts.dirac_transformation_information:main [admin]
 	dirac-transformation-recover-data = DIRAC.TransformationSystem.scripts.dirac_transformation_recover_data:main [admin]
-	dirac-transformation-remove-output = DIRAC.TransformationSystem.scripts.dirac_transformation_remove_output:main [admin]
 	dirac-transformation-replication = DIRAC.TransformationSystem.scripts.dirac_transformation_replication:main [admin]
 	dirac-transformation-verify-outputdata = DIRAC.TransformationSystem.scripts.dirac_transformation_verify_outputdata:main [admin]
 	dirac-transformation-update-derived = DIRAC.TransformationSystem.scripts.dirac_transformation_update_derived:main [admin]

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/ConfigurationSystem/Client/Helpers/Registry.py

@@ -457,7 +457,7 @@ def getVOForGroup(group):
 
     :return: str
     """
-    return getVO() or gConfig.getValue(f"{gBaseRegistrySection}/Groups/{group}/VO", "")
+    return gConfig.getValue(f"{gBaseRegistrySection}/Groups/{group}/VO", "") or getVO()
 
 
 def getIdPForGroup(group):

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/Core/Security/DiracX.py

@@ -47,7 +47,7 @@ RE_DIRACX_PEM = re.compile(rf"{PEM_BEGIN}\n(.*)\n{PEM_END}", re.MULTILINE | re.D
 def addTokenToPEM(pemPath, group):
     from DIRAC.Core.Base.Client import Client
 
-    vo = Registry.getVOMSVOForGroup(group)
+    vo = Registry.getVOForGroup(group)
     if not vo:
         gLogger.error(f"ERROR: Could not find VO for group {group}, DiracX will not work!")
     disabledVOs = gConfig.getValue("/DiracX/DisabledVOs", [])

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/Core/Tornado/Client/ClientSelector.py

@@ -17,7 +17,6 @@ from DIRAC.Core.DISET.RPCClient import RPCClient
 from DIRAC.Core.DISET.TransferClient import TransferClient
 from DIRAC.Core.Tornado.Client.TornadoClient import TornadoClient
 
-
 sLog = gLogger.getSubLogger(__name__)
 
 
@@ -82,6 +81,10 @@ def ClientSelector(disetClient, *args, **kwargs):  # We use same interface as RP
             rpc = tornadoClient(*args, **kwargs)
         else:
             rpc = disetClient(*args, **kwargs)
+    except NotImplementedError as e:
+        # We catch explicitly NotImplementedError to avoid just printing "there's an error"
+        # If we mis-configured the CS for legacy adapted services, we MUST have an error.
+        raise e
     except Exception as e:  # pylint: disable=broad-except
         # If anything went wrong in the resolution, we return default RPCClient
         # So the behaviour is exactly the same as before implementation of Tornado

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/Core/Utilities/Extensions.py

@@ -73,6 +73,15 @@ def findServices(modules):
     return findModules(modules, "Service", "*Handler")
 
 
+def findFutureServices(modules):
+    """Find the legacy adapted services for one or more DIRAC extension(s)
+
+    :param list/str/module module: One or more Python modules or Python module names
+    :returns: list of tuples of the form (SystemName, ServiceName)
+    """
+    return findModules(modules, "FutureClient")
+
+
 @iterateThenSort
 def findDatabases(module):
     """Find the DB SQL schema defintions for one or more DIRAC extension(s)
@@ -182,7 +191,7 @@ def parseArgs():
     parser = argparse.ArgumentParser()
     subparsers = parser.add_subparsers(required=True, dest="function")
     defaultExtensions = extensionsByPriority()
-    for func in [findSystems, findAgents, findExecutors, findServices, findDatabases]:
+    for func in [findSystems, findAgents, findExecutors, findServices, findDatabases, findFutureServices]:
         subparser = subparsers.add_parser(func.__name__)
         subparser.add_argument("--extensions", nargs="+", default=defaultExtensions)
         subparser.set_defaults(func=func)

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/Interfaces/API/Dirac.py

@@ -46,6 +46,7 @@ from DIRAC.WorkloadManagementSystem.Client import JobStatus
 from DIRAC.WorkloadManagementSystem.Client.JobMonitoringClient import JobMonitoringClient
 from DIRAC.WorkloadManagementSystem.Client.SandboxStoreClient import SandboxStoreClient
 from DIRAC.WorkloadManagementSystem.Client.WMSClient import WMSClient
+from DIRAC.WorkloadManagementSystem.Utilities.jobAdministration import _filterJobStateTransition
 
 
 def parseArguments(args):
@@ -1450,10 +1451,13 @@ class Dirac(API):
         # Remove any job IDs that can't change to the Killed or Deleted states
         filteredJobs = set()
         for filterState in (JobStatus.KILLED, JobStatus.DELETED):
-            filterRes = JobStatus.filterJobStateTransition(jobIDs, filterState)
-            if not filterRes["OK"]:
-                return filterRes
-            filteredJobs.update(filterRes["Value"])
+            # get a dictionary of jobID:status
+            res = JobMonitoringClient().getJobsStatus(jobIDs)
+            if not res["OK"]:
+                return res
+            js = {k: v["Status"] for k, v in res["Value"].items()}
+            # then filter
+            filteredJobs.update(_filterJobStateTransition(js, filterState))
 
         return WMSClient(useCertificates=self.useCertificates).deleteJob(list(filteredJobs))
 
@@ -1480,11 +1484,13 @@ class Dirac(API):
             return ret
         jobIDs = ret["Value"]
 
-        # Remove any job IDs that can't change to the rescheduled state
-        filterRes = JobStatus.filterJobStateTransition(jobIDs, JobStatus.RESCHEDULED)
-        if not filterRes["OK"]:
-            return filterRes
-        jobIDsToReschedule = filterRes["Value"]
+        # get a dictionary of jobID:status
+        res = JobMonitoringClient().getJobsStatus(jobIDs)
+        if not res["OK"]:
+            return res
+        js = {k: v["Status"] for k, v in res["Value"].items()}
+        # then filter
+        jobIDsToReschedule = _filterJobStateTransition(js, JobStatus.RESCHEDULED)
 
         return WMSClient(useCertificates=self.useCertificates).rescheduleJob(jobIDsToReschedule)
 
@@ -1510,10 +1516,13 @@ class Dirac(API):
         # Remove any job IDs that can't change to the Killed or Deleted states
         filteredJobs = set()
         for filterState in (JobStatus.KILLED, JobStatus.DELETED):
-            filterRes = JobStatus.filterJobStateTransition(jobIDs, filterState)
-            if not filterRes["OK"]:
-                return filterRes
-            filteredJobs.update(filterRes["Value"])
+            # get a dictionary of jobID:status
+            res = JobMonitoringClient().getJobsStatus(jobIDs)
+            if not res["OK"]:
+                return res
+            js = {k: v["Status"] for k, v in res["Value"].items()}
+            # then filter
+            filteredJobs.update(_filterJobStateTransition(js, filterState))
 
         return WMSClient(useCertificates=self.useCertificates).killJob(list(filteredJobs))
 

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/TransformationSystem/Agent/TransformationCleaningAgent.py

@@ -32,9 +32,12 @@ from DIRAC.Resources.Catalog.FileCatalogClient import FileCatalogClient
 from DIRAC.Resources.Storage.StorageElement import StorageElement
 from DIRAC.TransformationSystem.Client import TransformationStatus
 from DIRAC.TransformationSystem.Client.TransformationClient import TransformationClient
-from DIRAC.WorkloadManagementSystem.Client.JobMonitoringClient import JobMonitoringClient
-from DIRAC.WorkloadManagementSystem.Client.WMSClient import WMSClient
 from DIRAC.WorkloadManagementSystem.DB.JobDB import JobDB
+from DIRAC.WorkloadManagementSystem.Service.JobPolicy import (
+    RIGHT_DELETE,
+    RIGHT_KILL,
+)
+from DIRAC.WorkloadManagementSystem.Utilities.jobAdministration import kill_delete_jobs
 
 # # agent's name
 AGENT_NAME = "Transformation/TransformationCleaningAgent"
@@ -58,8 +61,6 @@ class TransformationCleaningAgent(AgentModule):
 
         # # transformation client
         self.transClient = None
-        # # wms client
-        self.wmsClient = None
         # # request client
         self.reqClient = None
         # # file catalog client
@@ -120,14 +121,10 @@ class TransformationCleaningAgent(AgentModule):
 
         # # transformation client
         self.transClient = TransformationClient()
-        # # wms client
-        self.wmsClient = WMSClient()
         # # request client
         self.reqClient = ReqClient()
         # # file catalog client
         self.metadataClient = FileCatalogClient()
-        # # job monitoring client
-        self.jobMonitoringClient = JobMonitoringClient()
         # # job DB
         self.jobDB = JobDB()
 
@@ -271,7 +268,7 @@ class TransformationCleaningAgent(AgentModule):
 
             # Remove JobIDs that were unknown to the TransformationSystem
             jobGroupsToCheck = [str(transDict["TransformationID"]).zfill(8) for transDict in toClean + toArchive]
-            res = self.jobMonitoringClient.getJobs({"JobGroup": jobGroupsToCheck})
+            res = self.jobDB.selectJobs({"JobGroup": jobGroupsToCheck})
             if not res["OK"]:
                 return res
             jobIDsToRemove = [int(jobID) for jobID in res["Value"]]
@@ -613,8 +610,8 @@ class TransformationCleaningAgent(AgentModule):
         # Prevent 0 job IDs
         jobIDs = [int(j) for j in transJobIDs if int(j)]
         allRemove = True
-        for jobList in breakListIntoChunks(jobIDs, 500):
-            res = self.wmsClient.killJob(jobList, force=True)
+        for jobList in breakListIntoChunks(jobIDs, 1000):
+            res = kill_delete_jobs(RIGHT_KILL, jobList, force=True)
             if res["OK"]:
                 self.log.info(f"Successfully killed {len(jobList)} jobs from WMS")
             elif ("InvalidJobIDs" in res) and ("NonauthorizedJobIDs" not in res) and ("FailedJobIDs" not in res):
@@ -626,7 +623,7 @@ class TransformationCleaningAgent(AgentModule):
                 self.log.error("Failed to kill jobs", f"(n={len(res['FailedJobIDs'])})")
                 allRemove = False
 
-            res = self.wmsClient.deleteJob(jobList)
+            res = kill_delete_jobs(RIGHT_DELETE, jobList, force=True)
             if res["OK"]:
                 self.log.info("Successfully deleted jobs from WMS", f"(n={len(jobList)})")
             elif ("InvalidJobIDs" in res) and ("NonauthorizedJobIDs" not in res) and ("FailedJobIDs" not in res):

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/WorkloadManagementSystem/Agent/JobCleaningAgent.py

@@ -35,10 +35,12 @@ from DIRAC.RequestManagementSystem.Client.Operation import Operation
 from DIRAC.RequestManagementSystem.Client.ReqClient import ReqClient
 from DIRAC.RequestManagementSystem.Client.Request import Request
 from DIRAC.WorkloadManagementSystem.Client import JobStatus
-from DIRAC.WorkloadManagementSystem.Client.JobMonitoringClient import JobMonitoringClient
 from DIRAC.WorkloadManagementSystem.Client.WMSClient import WMSClient
 from DIRAC.WorkloadManagementSystem.DB.JobDB import JobDB
 from DIRAC.WorkloadManagementSystem.DB.SandboxMetadataDB import SandboxMetadataDB
+from DIRAC.WorkloadManagementSystem.Service.JobPolicy import RIGHT_DELETE
+from DIRAC.WorkloadManagementSystem.Utilities.jobAdministration import kill_delete_jobs
+from DIRAC.WorkloadManagementSystem.Utilities.JobParameters import getJobParameters
 
 
 class JobCleaningAgent(AgentModule):
@@ -230,11 +232,11 @@ class JobCleaningAgent(AgentModule):
             if not res["OK"]:
                 self.log.error("No DN found", f"for {user}")
                 return res
-            wmsClient = WMSClient(useCertificates=True, delegatedDN=res["Value"][0], delegatedGroup=ownerGroup)
             if remove:
+                wmsClient = WMSClient(useCertificates=True, delegatedDN=res["Value"][0], delegatedGroup=ownerGroup)
                 result = wmsClient.removeJob(jobsList)
             else:
-                result = wmsClient.deleteJob(jobsList)
+                result = kill_delete_jobs(RIGHT_DELETE, jobsList)
             if not result["OK"]:
                 self.log.error(
                     f"Could not {'remove' if remove else 'delete'} jobs",
@@ -294,7 +296,8 @@ class JobCleaningAgent(AgentModule):
         failed = {}
         successful = {}
 
-        result = JobMonitoringClient().getJobParameters(jobIDList, ["OutputSandboxLFN"])
+        jobIDs = [int(jobID) for jobID in jobIDList]
+        result = getJobParameters(jobIDs, "OutputSandboxLFN")
         if not result["OK"]:
             return result
         osLFNDict = result["Value"]

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/WorkloadManagementSystem/Agent/StalledJobAgent.py

@@ -14,16 +14,16 @@ import datetime
 from DIRAC import S_ERROR, S_OK, gConfig
 from DIRAC.AccountingSystem.Client.Types.Job import Job
 from DIRAC.ConfigurationSystem.Client.Helpers import cfgPath
-from DIRAC.ConfigurationSystem.Client.Helpers.Registry import getDNForUsername
 from DIRAC.Core.Base.AgentModule import AgentModule
 from DIRAC.Core.Utilities import DErrno
 from DIRAC.Core.Utilities.ClassAd.ClassAdLight import ClassAd
 from DIRAC.Core.Utilities.TimeUtilities import fromString, second, toEpoch
 from DIRAC.WorkloadManagementSystem.Client import JobMinorStatus, JobStatus
-from DIRAC.WorkloadManagementSystem.Client.WMSClient import WMSClient
 from DIRAC.WorkloadManagementSystem.DB.JobDB import JobDB
 from DIRAC.WorkloadManagementSystem.DB.JobLoggingDB import JobLoggingDB
 from DIRAC.WorkloadManagementSystem.DB.PilotAgentsDB import PilotAgentsDB
+from DIRAC.WorkloadManagementSystem.Service.JobPolicy import RIGHT_KILL
+from DIRAC.WorkloadManagementSystem.Utilities.jobAdministration import kill_delete_jobs
 from DIRAC.WorkloadManagementSystem.Utilities.JobParameters import getJobParameters
 from DIRAC.WorkloadManagementSystem.Utilities.Utils import rescheduleJobs
 
@@ -235,7 +235,7 @@ class StalledJobAgent(AgentModule):
         # Set the jobs Failed, send them a kill signal in case they are not really dead
         # and send accounting info
         if setFailed:
-            res = self._sendKillCommand(jobID)
+            res = kill_delete_jobs(RIGHT_KILL, [jobID], nonauthJobList=[], force=True)
             if not res["OK"]:
                 self.log.error("Failed to kill job", jobID)
 
@@ -574,26 +574,3 @@ class StalledJobAgent(AgentModule):
                 continue
 
         return S_OK()
-
-    def _sendKillCommand(self, job):
-        """Send a kill signal to the job such that it cannot continue running.
-
-        :param int job: ID of job to send kill command
-        """
-
-        res = self.jobDB.getJobAttribute(job, "Owner")
-        if not res["OK"]:
-            return res
-        owner = res["Value"]
-
-        res = self.jobDB.getJobAttribute(job, "OwnerGroup")
-        if not res["OK"]:
-            return res
-        ownerGroup = res["Value"]
-
-        wmsClient = WMSClient(
-            useCertificates=True,
-            delegatedDN=getDNForUsername(owner)["Value"][0] if owner else None,
-            delegatedGroup=ownerGroup,
-        )
-        return wmsClient.killJob(job)

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/WorkloadManagementSystem/Agent/StatesAccountingAgent.py

@@ -135,8 +135,15 @@ class StatesAccountingAgent(AgentModule):
             for backend in self.datastores:
                 if backend.lower() == "monitoring":
                     site_name = rD["Site"]
-                    rD["Tier"] = site_metadata[site_name]["Tier"]
-                    rD["Type"] = site_metadata[site_name]["Type"]
+                    if site_name not in site_metadata:
+                        self.log.warn(
+                            f"Site {site_name} not found in site metadata, using default values",
+                        )
+                        rD["Tier"] = "4"
+                        rD["Type"] = site_name.split(".")[0]
+                    else:
+                        rD["Tier"] = site_metadata[site_name]["Tier"]
+                        rD["Type"] = site_metadata[site_name]["Type"]
                     rD["timestamp"] = int(TimeUtilities.toEpochMilliSeconds(now))
                     self.datastores["Monitoring"].addRecord(rD)
 

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/WorkloadManagementSystem/Agent/test/Test_Agent_JobCleaningAgent.py

@@ -1,10 +1,11 @@
 """ Test class for Job Cleaning Agent
 """
-import pytest
 from unittest.mock import MagicMock
 
+import pytest
+
 # DIRAC Components
-from DIRAC import gLogger, S_OK
+from DIRAC import S_OK, gLogger
 from DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent import JobCleaningAgent
 
 gLogger.setLevel("DEBUG")
@@ -32,7 +33,6 @@ def jca(mocker):
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.JobDB.selectJobs", side_effect=mockReply)
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.JobDB.__init__", side_effect=mockNone)
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.ReqClient", return_value=mockNone)
-    mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.JobMonitoringClient", return_value=mockJMC)
 
     jca = JobCleaningAgent()
     jca.log = gLogger
@@ -98,7 +98,7 @@ def test_deleteJobsByStatus(jca, conditions, mockReplyInput, expected):
     "inputs, params, expected",
     [
         ([], {"OK": True, "Value": {}}, {"OK": True, "Value": {"Failed": {}, "Successful": {}}}),
-        (["a", "b"], {"OK": True, "Value": {}}, {"OK": True, "Value": {"Failed": {}, "Successful": {}}}),
+        (["123", "456"], {"OK": True, "Value": {}}, {"OK": True, "Value": {"Failed": {}, "Successful": {}}}),
         (
             [],
             {"OK": True, "Value": {1: {"OutputSandboxLFN": "/some/lfn/1.txt"}}},
@@ -113,11 +113,11 @@ def test_deleteJobsByStatus(jca, conditions, mockReplyInput, expected):
             {"OK": True, "Value": {"Failed": {}, "Successful": {1: "/some/lfn/1.txt", 2: "/some/other/lfn/2.txt"}}},
         ),
         (
-            ["a", "b"],
+            ["123", "456"],
             {"OK": True, "Value": {1: {"OutputSandboxLFN": "/some/lfn/1.txt"}}},
             {"OK": True, "Value": {"Failed": {}, "Successful": {1: "/some/lfn/1.txt"}}},
         ),
-        (["a", "b"], {"OK": False}, {"OK": False}),
+        (["123", "456"], {"OK": False}, {"OK": False}),
     ],
 )
 def test_deleteJobOversizedSandbox(mocker, inputs, params, expected):
@@ -127,10 +127,10 @@ def test_deleteJobOversizedSandbox(mocker, inputs, params, expected):
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.AgentModule.am_getOption", return_value=mockAM)
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.JobDB", return_value=mockNone)
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.ReqClient", return_value=mockNone)
-    mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.JobMonitoringClient", return_value=mockJMC)
     mocker.patch(
         "DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.getDNForUsername", return_value=S_OK(["/bih/boh/DN"])
     )
+    mocker.patch("DIRAC.WorkloadManagementSystem.Agent.JobCleaningAgent.getJobParameters", return_value=params)
 
     jobCleaningAgent = JobCleaningAgent()
     jobCleaningAgent.log = gLogger
@@ -138,8 +138,6 @@ def test_deleteJobOversizedSandbox(mocker, inputs, params, expected):
     jobCleaningAgent._AgentModule__configDefaults = mockAM
     jobCleaningAgent.initialize()
 
-    mockJMC.getJobParameters.return_value = params
-
     result = jobCleaningAgent.deleteJobOversizedSandbox(inputs)
 
     assert result == expected

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/WorkloadManagementSystem/Agent/test/Test_Agent_StalledJobAgent.py

@@ -28,8 +28,7 @@ def sja(mocker):
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.StalledJobAgent.rescheduleJobs", return_value=MagicMock())
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.StalledJobAgent.PilotAgentsDB", return_value=MagicMock())
     mocker.patch("DIRAC.WorkloadManagementSystem.Agent.StalledJobAgent.getJobParameters", return_value=MagicMock())
-    mocker.patch("DIRAC.WorkloadManagementSystem.Agent.StalledJobAgent.WMSClient", return_value=MagicMock())
-    mocker.patch("DIRAC.WorkloadManagementSystem.Agent.StalledJobAgent.getDNForUsername", return_value=MagicMock())
+    mocker.patch("DIRAC.WorkloadManagementSystem.Agent.StalledJobAgent.kill_delete_jobs", return_value=MagicMock())
 
     stalledJobAgent = StalledJobAgent()
     stalledJobAgent._AgentModule__configDefaults = mockAM

{dirac-9.0.0a62 → dirac-9.0.0a63}/src/DIRAC/WorkloadManagementSystem/Client/JobMonitoringClient.py

@@ -14,12 +14,14 @@ except ImportError:
 
 @createClient("WorkloadManagement/JobMonitoring")
 class JobMonitoringClient(Client):
+    # Set to None to raise an error if this service is set as "legacy adapted"
+    # See ClientSelector
+    diracxClient = None
+
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
         self.setServer("WorkloadManagement/JobMonitoring")
 
-    diracxClient = futureJobMonitoringClient
-
     @ignoreEncodeWarning
     def getJobsStatus(self, jobIDs):
         res = self._getRPC().getJobsStatus(jobIDs)