Sync with upstream

bin/sync_slurm_acct.py

@@ -149,8 +149,6 @@ def main():
             logging.info("Executing %d commands", len(sacctmgr_commands))
             execute_commands(sacctmgr_commands)
 
-        # reset to go on with the remainder of the commands
-        sacctmgr_commands = []
 
         # safety to avoid emptying the cluster due to some error upstream
         if not opts.options.force and len(job_cancel_commands) > MAX_USERS_JOB_CANCEL:
@@ -160,7 +158,10 @@ def main():
                 logging.debug("%s", jc)
             raise SyncSanityError("Would cancel jobs for %d users" % len(job_cancel_commands))
 
-        sacctmgr_commands += [c for cl in job_cancel_commands.values() for c in cl]
+        scancel_commands = [c for cl in job_cancel_commands.values() for c in cl]
+
+        # reset to go on with the remainder of the commands
+        sacctmgr_commands = []
 
         # removing users may fail, so should be done last
         sacctmgr_commands += association_remove_commands
@@ -170,6 +171,7 @@ def main():
             print("\n".join([" ".join(c) for c in sacctmgr_commands]))
         else:
             logging.info("Executing %d commands", len(sacctmgr_commands))
+            execute_commands(scancel_commands, allow_failure=True)
             execute_commands(sacctmgr_commands)
 
         if not opts.options.dry_run:

lib/vsc/administration/slurm/sacctmgr.py

@@ -244,10 +244,10 @@ def create_default_account_command(user, account, cluster):
     """
     command = [
         "user",
-        "Name={0}".format(user),
-        "Cluster={0}".format(cluster),
+        f"Name={user}",
+        f"Cluster={cluster}",
         "set",
-        "DefaultAccount={0}".format(account),
+        f"DefaultAccount={account}",
     ]
     logging.debug(
         "Creating command to set default account to %s for %s on cluster %s",
@@ -262,10 +262,10 @@ def create_default_account_command(user, account, cluster):
 def create_change_account_fairshare_command(account, cluster, fairshare):
     command = [
         "account",
-        "name={0}".format(account),
-        "cluster={0}".format(cluster),
+        f"name={account}",
+        f"cluster={cluster}",
         "set",
-        "fairshare={0}".format(fairshare),
+        f"fairshare={fairshare}",
     ]
     logging.debug(
         "Adding command to change fairshare for account %s on cluster %s to %d",
@@ -278,7 +278,7 @@ def create_change_account_fairshare_command(account, cluster, fairshare):
 
 
 @mksacctmgr('add')
-def create_add_user_command(user, account, cluster, default_account=None):
+def create_add_user_command(user, account, cluster, partition=None, default_account=None):
     """
     Creates the command to add the given account.
 
@@ -292,12 +292,16 @@ def create_add_user_command(user, account, cluster, default_account=None):
     command = [
         "user",
         user,
-        "Account={0}".format(account),
-        "Cluster={0}".format(cluster)
+        f"Account={account}",
+        f"Cluster={cluster}",
     ]
+    if partition is not None:
+        command.append(
+            f"Partition={partition}"
+        )
     if default_account is not None:
         command.append(
-            "DefaultAccount={0}".format(account),
+            f"DefaultAccount={default_account}",
         )
     logging.debug(
         "Adding command to add user %s with Account=%s Cluster=%s",
@@ -340,16 +344,18 @@ def create_change_user_command(user, current_vo_id, new_vo_id, cluster):
 
 
 @mksacctmgr('remove')
-def create_remove_user_command(user, cluster):
+def create_remove_user_command(user, cluster=None):
     """Create the command to remove a user.
 
     @returns: list comprising the command
     """
     command = [
         "user",
-        "Name={user}".format(user=user),
-        "Cluster={cluster}".format(cluster=cluster)
+        f"Name={user}",
     ]
+
+    if cluster is not None:
+        command.append(f"Cluster={cluster}")
     logging.debug(
         "Adding command to remove user %s from Cluster=%s",
         user,
@@ -367,8 +373,8 @@ def create_remove_account_command(account, cluster):
     """
     command = [
         "account",
-        "Name={account}".format(account=account),
-        "Cluster={cluster}".format(cluster=cluster),
+        f"Name={account}",
+        f"Cluster={cluster}",
     ]
 
     logging.debug(
@@ -381,18 +387,21 @@ def create_remove_account_command(account, cluster):
 
 
 @mksacctmgr('remove')
-def create_remove_user_account_command(user, account, cluster):
+def create_remove_user_account_command(user, account, cluster, partition=None):
     """Create the command to remove a user.
 
     @returns: list comprising the command
     """
     command = [
         "user",
-        "Name={user}".format(user=user),
-        "Account={account}".format(account=account),
-        "Cluster={cluster}".format(cluster=cluster)
+        f"Name={user}",
+        f"Account={account}",
+        f"Cluster={cluster}",
     ]
 
+    if partition is not None:
+        command.append(f"Partition={partition}")
+
     logging.debug(
         "Adding command to remove user %s with account %s from Cluster=%s",
         user,
@@ -411,7 +420,7 @@ def create_add_qos_command(name):
     """
     command = [
         "qos",
-        "Name={0}".format(name)
+        f"Name={name}",
     ]
 
     return command
@@ -428,7 +437,7 @@ def create_remove_qos_command(name):
     command = [
         "qos",
         "where",
-        "Name={0}".format(name),
+        f"Name={name}",
     ]
 
     return command
@@ -451,7 +460,7 @@ def create_modify_qos_command(name, settings):
     ]
 
     for k, v in settings.items():
-        command.append("{0}={1}".format(k, v))
+        command.append(f"{k}={v}")
 
     return command
 
@@ -465,11 +474,11 @@ def create_add_resource_license_command(name, server, stype, clusters, count):
     command = [
         "resource",
         "Type=license",
-        "Name={0}".format(name),
-        "Server={0}".format(server),
-        "ServerType={0}".format(stype),
-        "Cluster={0}".format(",".join(clusters)),
-        "Count={0}".format(count),
+        f"Name={name}",
+        f"Server={server}",
+        f"ServerType={stype}",
+        "Cluster={}".format(",".join(clusters)),
+        f"Count={count}",
         "PercentAllowed=100",
     ]
 
@@ -486,9 +495,9 @@ def create_remove_resource_license_command(name, server, stype):
         "resource",
         "where",
         "Type=license",
-        "Name={0}".format(name),
-        "Server={0}".format(server),
-        "ServerType={0}".format(stype),
+        f"Name={name}",
+        f"Server={server}",
+        f"ServerType={stype}",
     ]
 
     return command
@@ -503,11 +512,12 @@ def create_modify_resource_license_command(name, server, stype, count):
     command = [
         "resource",
         "where",
-        "Name={0}".format(name),
-        "Server={0}".format(server),
-        "ServerType={0}".format(stype),
+        f"Name={name}",
+        f"Server={server}",
+        f"ServerType={stype}",
         "set",
-        "Count={0}".format(count),
+        f"Count={count}",
+        "PercentAllowed=100",
     ]
 
     return command

lib/vsc/administration/slurm/scancel.py

@@ -18,17 +18,19 @@
 SLURM_SCANCEL = "/usr/bin/scancel"
 
 
-def create_remove_user_jobs_command(user, cluster, state=None, account=None):
+def create_remove_user_jobs_command(user, cluster=None, state=None, account=None):
     """Create the command to remove a user's jobs in the given state.
 
     @returns: a list comprising the command
     """
     remove_user_jobs_command = [
         SLURM_SCANCEL,
-        "--cluster={cluster}".format(cluster=cluster),
         "--user={user}".format(user=user),
     ]
 
+    if cluster is not None:
+        remove_user_jobs_command.append("--clusters={cluster}".format(cluster=cluster))
+
     if state is not None:
         remove_user_jobs_command.append("--state={state}".format(state=state))