def GetUserPendingJobs(userName, vcName):
jobs = []
allJobs = DataManager.GetAllPendingJobs(vcName)
for job in allJobs:
if userName == "all" or userName == job["userName"]:
jobs.append(job)
return jobs
def GetIdentityInfoFromDB(name):
lst = DataManager.GetIdentityInfo(name)
if lst:
return lst[0]
else:
logger.warning("Identity name %s not found in DB", name)
return INVALID_INFO
def ListVCs(userName):
ret = []
vcList = DataManager.ListVCs()
for vc in vcList:
if AuthorizationManager.HasAccess(userName, ResourceType.VC, vc["vcName"], Permission.User):
ret.append(vc)
# web portal (client) can filter out Default VC
return ret
def GetVC(userName, vcName):
ret = None
clusterStatus, dummy = DataManager.GetClusterStatus()
clusterTotalRes = ResourceInfo(clusterStatus["gpu_capacity"])
clusterReservedRes = ResourceInfo(clusterStatus["gpu_unschedulable"])
user_status = {}
vcList = DataManager.ListVCs()
for vc in vcList:
if vc["vcName"] == vcName and AuthorizationManager.HasAccess(userName, ResourceType.VC, vcName, Permission.User):
vcTotalRes = ResourceInfo(json.loads(vc["quota"]))
vcConsumedRes = ResourceInfo()
jobs = DataManager.GetAllPendingJobs(vcName)
for job in jobs:
if job["jobStatus"] == "running":
username = job["userName"]
jobParam = json.loads(base64.b64decode(job["jobParams"]))
if "gpuType" in jobParam and not jobParam["preemptionAllowed"]:
vcConsumedRes.Add(ResourceInfo({jobParam["gpuType"] : GetJobTotalGpu(jobParam)}))
if username not in user_status:
user_status[username] = ResourceInfo()
user_status[username].Add(ResourceInfo({jobParam["gpuType"] : GetJobTotalGpu(jobParam)}))
vcReservedRes = clusterReservedRes.GetFraction(vcTotalRes, clusterTotalRes)
vcAvailableRes = ResourceInfo.Difference(ResourceInfo.Difference(vcTotalRes, vcConsumedRes), vcReservedRes)
vc["gpu_capacity"] = vcTotalRes.ToSerializable()
vc["gpu_used"] = vcConsumedRes.ToSerializable()
vc["gpu_unschedulable"] = vcReservedRes.ToSerializable()
vc["gpu_avaliable"] = vcAvailableRes.ToSerializable()
vc["AvaliableJobNum"] = len(jobs)
vc["node_status"] = clusterStatus["node_status"]
vc["user_status"] = []
for user_name, user_gpu in user_status.iteritems():
# TODO: job_manager.getAlias should be put in a util file
user_name = user_name.split("@")[0].strip()
vc["user_status"].append({"userName":user_name, "userGPU":user_gpu.ToSerializable()})
ret = vc
break
return ret
def GetIdentityInfoFromDB(identityName):
lst = DataManager.GetIdentityInfo(identityName)
if lst:
return lst[0]
else:
logger.warn("GetIdentityInfo : Identity %s not found in DB" %
identityName)
info = {}
info["uid"] = INVALID_ID
info["gid"] = INVALID_ID
info["groups"] = [INVALID_ID]
return info
def getClusterVCs():
vcList = None
try:
with vc_cache_lock:
vcList = copy.deepcopy(vc_cache.values())
except Exception:
pass
if not vcList:
vcList = DataManager.ListVCs()
with vc_cache_lock:
for vc in vcList:
vc_cache[vc["vcName"]] = vc
return vcList
def _HasAccess(identityName, resourceAclPath, permissions):
start_time = timeit.default_timer()
requestedAccess = '%s;%s;%s' % (str(identityName), resourceAclPath,
str(permissions))
try:
identities = []
identities.extend(
IdentityManager.GetIdentityInfoFromDB(identityName)["groups"])
#TODO: handle isDeny
while resourceAclPath:
#logger.debug('resourceAclPath ' + resourceAclPath)
acl = DataManager.GetResourceAcl(resourceAclPath)
for ace in acl:
for identity in identities:
#logger.debug('identity %s' % identity)
if ace["identityName"] == identityName or (
str(ace["identityId"]) == str(identity) and
(int(identity) < INVALID_RANGE_START
or int(identity) > INVALID_RANGE_END)):
permissions = permissions & (~ace["permissions"])
if not permissions:
logger.info(
'Yes for %s in time %s' %
(requestedAccess,
str(timeit.default_timer() - start_time)))
return True
resourceAclPath = AuthorizationManager.__GetParentPath(
resourceAclPath)
logger.info(
'No for %s in time %s' %
(requestedAccess, str(timeit.default_timer() - start_time)))
return False
except Exception as e:
logger.error('Exception: ' + str(e))
logger.warn(
'No (exception) for %s in time %s' %
(requestedAccess, str(timeit.default_timer() - start_time)))
return False
def GetClusterStatus():
cluster_status,last_update_time = DataManager.GetClusterStatus()
return cluster_status,last_update_time
def InvalidateJobListCache(vcName):
CacheManager.Invalidate("GetAllPendingJobs", vcName)
DataManager.GetAllPendingJobs(vcName)