def shutdown(self, request, computeset_id, format=None):
"""Shutdown the nodes in the identified ComputeSet."""
cset = ComputeSet.objects.get(pk=computeset_id)
if not cset.cluster.project in request.user.groups.all():
raise PermissionDenied()
computes = []
for compute in cset.computes.all():
computes.append(compute.rocks_name)
if compute.cluster.name != request.data["cluster"]:
cset.delete()
return Response(
"The node %s does not belong to the cluster %s, belongs to %s"
% (node, request.data["cluster"], compute.cluster.name),
status=status.HTTP_400_BAD_REQUEST,
)
cset.computes.add(compute)
submit_computeset.delay(FullComputeSetSerializer(cset).data)
# We should only poweron computes after entering jobscript and
# finishing the PROLOG on all allocated nodes. At that point the
# nodelist will be returned and we can call poweron_nodeset()
# poweron_nodeset.delay(nodes, hosts)
location = "/nucleus/v1/computeset/%s" % (cset.id)
serializer = ComputeSetSerializer(cset)
response = Response(serializer.data, status=201, headers={"Location": request.build_absolute_uri(location)})
return response
def poweron(self, request, format=None):
""" Power on a set of computes creating a ComputeSet."""
clust = get_object_or_404(Cluster, name=request.data["cluster"])
if not clust.project in request.user.groups.all():
raise PermissionDenied()
walltime_mins = request.data.get("walltime_mins")
if not walltime_mins:
walltime_mins = 2880
nodes = []
hosts = []
if request.data["computes"] is list:
for obj in request.data["computes"]:
nodes.append(obj["name"])
hosts.append(obj["host"])
else:
nodes = hostlist.expand_hostlist("%s" % request.data["computes"])
if request.data.get("hosts"):
hosts = hostlist.expand_hostlist("%s" % request.data["hosts"])
if hosts and len(nodes) != len(hosts):
return Response("The length of hosts should be equal to length of nodes",
status=status.HTTP_400_BAD_REQUEST)
cset = ComputeSet()
cset.cluster = clust
cset.user = self.request.user.username
cset.account = clust.project
cset.walltime_mins = walltime_mins
cset.jobid = None
cset.name = None
cset.nodelist = ""
cset.state = ComputeSet.CSET_STATE_CREATED
cset.node_count = len(nodes)
cset.save()
for node in nodes:
compute = Compute.objects.get(name=node, cluster=clust)
other_cs_query = ComputeSet.objects.filter(computes__id__exact=compute.id).exclude(
state__exact=ComputeSet.CSET_STATE_COMPLETED)
if other_cs_query.exists():
cset.delete()
err_cs = other_cs_query.get()
return Response("The compute %s belongs to computeset %s which is in %s state" % (node, err_cs.id, err_cs.state), status=status.HTTP_400_BAD_REQUEST)
if compute.cluster.name != request.data["cluster"]:
cset.delete()
return Response("The node %s does not belong to the cluster %s, belongs to %s" % (node, request.data["cluster"], compute.cluster.name), status=status.HTTP_400_BAD_REQUEST)
cset.computes.add(compute)
submit_computeset.delay(FullComputeSetSerializer(cset).data)
# We should only poweron computes after entering jobscript and
# finishing the PROLOG on all allocated nodes. At that point the
# nodelist will be returned and we can call poweron_nodeset()
#poweron_nodeset.delay(nodes, hosts)
location = "/nucleus/v1/computeset/%s" % (cset.id)
serializer = ComputeSetSerializer(cset)
response = Response(
serializer.data,
status=201,
headers={'Location': location})
return response
def poweron(self, request, format=None):
""" Power on a set of computes creating a ComputeSet."""
clust = get_object_or_404(Cluster, name=request.data["cluster"])
if not clust.project in request.user.groups.all():
raise PermissionDenied()
walltime_mins = request.data.get("walltime_mins")
if not walltime_mins:
walltime_mins = 2880
nodes = []
hosts = []
if request.data.get("computes"):
if request.data["computes"] is list:
for obj in request.data["computes"]:
nodes.append(obj["name"])
hosts.append(obj["host"])
else:
nodes = hostlist.expand_hostlist("%s" % request.data["computes"])
if request.data.get("hosts"):
hosts = hostlist.expand_hostlist("%s" % request.data["hosts"])
elif request.data.get("count"):
computes_selected = (
Compute.objects.filter(cluster=clust)
.exclude(
computeset__state__in=[
ComputeSet.CSET_STATE_CREATED,
ComputeSet.CSET_STATE_SUBMITTED,
ComputeSet.CSET_STATE_RUNNING,
ComputeSet.CSET_STATE_ENDING,
]
)
.exclude(state="active")
.filter(Q(image_state="unmapped") | Q(image_state__isnull=True))
.exclude(image_locked=True)[: int(request.data["count"])]
)
nodes.extend([comp.name for comp in computes_selected])
if len(nodes) < int(request.data["count"]) or int(request.data["count"]) == 0:
return Response(
"There are %i nodes available for starting. Requested number should be greater than zero."
% len(nodes),
status=status.HTTP_400_BAD_REQUEST,
)
if hosts and len(nodes) != len(hosts):
return Response(
"The length of hosts should be equal to length of nodes", status=status.HTTP_400_BAD_REQUEST
)
cset = ComputeSet()
cset.cluster = clust
cset.user = clust.username
if request.data.get("allocation"):
cset.account = request.data["allocation"]
elif clust.allocations.count() == 1:
cset.account = clust.allocations.get().allocation
else:
return Response("Please specify the allocation", status=status.HTTP_400_BAD_REQUEST)
if not clust.allocations.filter(allocation=cset.account).exists():
return Response(
"Allocation %s does not belong to the cluster." % cset.account, status=status.HTTP_400_BAD_REQUEST
)
cset.walltime_mins = walltime_mins
cset.jobid = None
cset.name = None
cset.nodelist = ""
cset.state = ComputeSet.CSET_STATE_CREATED
cset.node_count = len(nodes)
cset.save()
for node in nodes:
compute = Compute.objects.get(name=node, cluster=clust)
other_cs_query = ComputeSet.objects.filter(computes__id__exact=compute.id).exclude(
state__in=[
ComputeSet.CSET_STATE_COMPLETED,
ComputeSet.CSET_STATE_FAILED,
ComputeSet.CSET_STATE_CANCELLED,
]
)
if other_cs_query.exists():
cset.delete()
err_cs = other_cs_query.get()
return Response(
"The compute %s belongs to computeset %s which is in %s state" % (node, err_cs.id, err_cs.state),
status=status.HTTP_400_BAD_REQUEST,
)
if (compute.image_state not in ["unmapped", None]) or compute.image_locked:
cset.delete()
return Response(
"The node %s's image is in %s state and image locked status is %s. Please contact the user support if the VM is not running."
% (node, compute.image_state, compute.image_locked),
status=status.HTTP_400_BAD_REQUEST,
)
if compute.cluster.name != request.data["cluster"]:
cset.delete()
return Response(
"The node %s does not belong to the cluster %s, belongs to %s"
% (node, request.data["cluster"], compute.cluster.name),
status=status.HTTP_400_BAD_REQUEST,
)
cset.computes.add(compute)
submit_computeset.delay(FullComputeSetSerializer(cset).data)
# We should only poweron computes after entering jobscript and
# finishing the PROLOG on all allocated nodes. At that point the
# nodelist will be returned and we can call poweron_nodeset()
# poweron_nodeset.delay(nodes, hosts)
location = "/nucleus/v1/computeset/%s" % (cset.id)
serializer = ComputeSetSerializer(cset)
response = Response(serializer.data, status=201, headers={"Location": request.build_absolute_uri(location)})
return response
