def setUp(self):
""" setup for the test """
self.agent_sessions = None
# get paths from the build_vars generated by build
with open('../../../.build_vars.json') as build_file:
build_paths = json.load(build_file)
self.context = DaosContext(build_paths['PREFIX'] + '/lib/')
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
# generate a hostfile
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
# fire up the DAOS servers
self.server_group = self.params.get("name", '/run/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers, self.server_group,
build_paths['PREFIX'] + '/../')
# create a pool to test with
createmode = self.params.get("mode", '/run/pool/createmode/')
createuid = self.params.get("uid", '/run/pool/createuid/')
creategid = self.params.get("gid", '/run/pool/creategid/')
createsetid = self.params.get("setname", '/run/pool/createset/')
createsize = self.params.get("size", '/run/pool/createsize/')
self.pool = DaosPool(self.context)
self.pool.create(createmode, createuid, creategid, createsize,
createsetid)
self.pool.get_uuid_str()
time.sleep(2)
def setUp(self):
"""Set up each test case."""
super(TestWithServers, self).setUp()
self.server_group = self.params.get("name", "/server_config/",
"daos_server")
# Determine which hosts to use as servers and optionally clients.
# Support the use of a host type count to test with subsets of the
# specified hosts lists
test_machines = self.params.get("test_machines", "/run/hosts/*")
test_servers = self.params.get("test_servers", "/run/hosts/*")
test_clients = self.params.get("test_clients", "/run/hosts/*")
server_count = self.params.get("server_count", "/run/hosts/*")
client_count = self.params.get("client_count", "/run/hosts/*")
# Supported combinations of yaml hosts arguments:
# - test_machines [+ server_count]
# - test_servers [+ server_count]
# - test_servers [+ server_count] + test_clients [+ client_count]
if test_machines:
self.hostlist_servers = test_machines[:server_count]
elif test_servers and test_clients:
self.hostlist_servers = test_servers[:server_count]
self.hostlist_clients = test_clients[:client_count]
elif test_servers:
self.hostlist_servers = test_servers[:server_count]
self.log.info("hostlist_servers: %s", self.hostlist_servers)
self.log.info("hostlist_clients: %s", self.hostlist_clients)
# If a specific count is specified, verify enough servers/clients are
# specified to satisy the count
host_count_checks = (
("server", server_count,
len(self.hostlist_servers) if self.hostlist_servers else 0),
("client", client_count,
len(self.hostlist_clients) if self.hostlist_clients else 0))
for host_type, expected_count, actual_count in host_count_checks:
if expected_count:
self.assertEqual(
expected_count, actual_count,
"Test requires {} {}; {} specified".format(
expected_count, host_type, actual_count))
# Start the servers and clients
# Create host files
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir, self.hostfile_servers_slots)
if self.hostlist_clients:
self.hostfile_clients = write_host_file.write_host_file(
self.hostlist_clients, self.workdir,
self.hostfile_clients_slots)
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers,
self.hostlist_clients)
if self.setup_start_servers:
self.start_servers()
def setUp(self):
self.agent_sessions = None
self.hostlist_servers = None
# get paths from the build_vars generated by build
with open(
os.path.join(os.path.dirname(os.path.realpath(__file__)),
"../../../../.build_vars.json")) as build_file:
build_paths = json.load(build_file)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
# NULL is causing connect to blow up so skip that test for now
uuidlist = self.params.get("uuid", '/run/connecttests/UUID/*/')
connectuuid = uuidlist[0]
if connectuuid == 'NULLPTR':
self.cancel("skipping null pointer test until DAOS-1781 is fixed")
# launch the server
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers, server_group,
self.basepath)
def setUp(self):
self.agent_sessions = None
# get paths from the build_vars generated by build
with open('../../../.build_vars.json') as build_file:
build_paths = json.load(build_file)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.server_group = self.params.get("name",
'/server_config/',
'daos_server')
# setup the DAOS python API
self.context = DaosContext(build_paths['PREFIX'] + '/lib/')
self.pool = None
self.d_log = DaosLog(self.context)
# getting hostfile
self.hostfile_servers = None
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/*')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
print ("Host file is: {}".format(self.hostfile_servers))
self.container = None
# starting server
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers, self.server_group,
self.basepath)
def setUp(self):
self.agent_sessions = None
# get paths from the build_vars generated by build
with open('../../../.build_vars.json') as build_file:
build_paths = json.load(build_file)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.server_group = self.params.get("name", '/server_config/',
'daos_server')
# setup the DAOS python API
self.context = DaosContext(build_paths['PREFIX'] + '/lib/')
self.hostlist_servers = self.params.get("test_servers",
'/run/hosts/test_machines/*')
self.hostfile_servers = (write_host_file.write_host_file(
self.hostlist_servers, self.workdir))
print("Host file servers is: {}".format(self.hostfile_servers))
self.hostlist_clients = self.params.get("test_clients",
'/run/hosts/test_machines/*')
self.slots = self.params.get("slots", '/run/ior/clientslots/*')
self.hostfile_clients = (write_host_file.write_host_file(
self.hostlist_clients, self.workdir, self.slots))
print("Host file clients is: {}".format(self.hostfile_clients))
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers,
self.hostlist_clients)
server_utils.run_server(self.hostfile_servers, self.server_group,
self.basepath)
if not distutils.spawn.find_executable("ior") and \
int(str(self.name).split("-")[0]) == 1:
ior_utils.build_ior(self.basepath)
def setUp(self):
# get paths from the build_vars generated by build
with open('../../../.build_vars.json') as build_file:
build_paths = json.load(build_file)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.server_group = self.params.get("name", '/server_config/',
'daos_server')
# setup the DAOS python API
self.context = DaosContext(build_paths['PREFIX'] + '/lib/')
self.hostlist_servers = self.params.get("test_servers", '/run/hosts/*')
hostfile_servers = (write_host_file.write_host_file(
self.hostlist_servers, self.workdir))
print("Host file servers is: {}".format(hostfile_servers))
self.hostlist_clients = self.params.get("test_clients", '/run/hosts/*')
self.slots = self.params.get("slots", '/run/ior/clientslots/*')
self.hostfile_clients = (write_host_file.write_host_file(
self.hostlist_clients, self.workdir, self.slots))
print("Host file clients is: {}".format(self.hostfile_clients))
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers,
self.hostlist_clients)
server_utils.run_server(hostfile_servers, self.server_group,
self.basepath)
def setUp(self):
# get paths from the build_vars generated by build
with open('../../../.build_vars.json') as build_file:
build_paths = json.load(build_file)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.server_group = self.params.get("name", '/server_config/',
'daos_server')
self.daosctl = self.basepath + '/install/bin/daosctl'
# setup the DAOS python API
self.context = DaosContext(build_paths['PREFIX'] + '/lib/')
self.pool = None
self.hostlist_servers = self.params.get("test_servers",
'/run/hosts/test_machines/*')
self.hostfile_servers = (write_host_file.write_host_file(
self.hostlist_servers, self.workdir))
print("Host file servers is: {}".format(self.hostfile_servers))
self.hostlist_clients = self.params.get("test_clients",
'/run/hosts/test_machines/*')
self.hostfile_clients = (write_host_file.write_host_file(
self.hostlist_clients, self.workdir, None))
print("Host file clientsis: {}".format(self.hostfile_clients))
# set ior_flags and object class to be used by test
self.ior_flags = self.params.get("F", '/run/ior/iorflags/')
self.object_class = self.params.get("o", '/run/ior/objectclass/')
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers,
self.hostlist_clients)
server_utils.run_server(self.hostfile_servers, self.server_group,
self.basepath)
def test_destroy_connect(self):
"""
Test destroying a pool that has a connected client with force == false.
Should fail.
:avocado: tags=pool,pooldestroy,x
"""
host = self.hostlist_servers[0]
try:
# write out a hostfile_servers and start the servers with it
self.hostlist_servers = self.params.get("test_machines1",
'/run/hosts/')
hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.tmp)
self.agent_sessions = agent_utils.run_agent(
self.basepath, self.hostlist_servers)
server_utils.run_server(hostfile_servers, self.server_group,
self.basepath)
# parameters used in pool create
createmode = self.params.get("mode", '/run/poolparams/createmode/')
createuid = self.params.get("uid", '/run/poolparams/createuid/')
creategid = self.params.get("gid", '/run/poolparams/creategid/')
createsetid = self.params.get("setname",
'/run/poolparams/createset/')
createsize = self.params.get("size", '/run/poolparams/createsize/')
# initialize a python pool object then create the underlying
# daos storage
pool = DaosPool(self.context)
pool.create(createmode, createuid, creategid, createsize,
createsetid, None)
# need a connection to create container
pool.connect(1 << 1)
# destroy pool with connection open
pool.destroy(0)
# should throw an exception and not hit this
self.fail("Shouldn't hit this line.\n")
except DaosApiError as excep:
print("got exception which is expected so long as it is BUSY")
print(excep)
print(traceback.format_exc())
# pool should still be there
exists = check_for_pool.check_for_pool(host, pool.get_uuid_str)
if exists != 0:
self.fail("Pool gone, but destroy should have failed.\n")
# no matter what happens cleanup
finally:
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions)
server_utils.stop_server(hosts=self.hostlist_servers)
os.remove(hostfile_servers)
def setUp(self):
super(BadExcludeTest, self).setUp()
self.agent_sessions = None
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(self,
self.hostlist_servers)
server_utils.run_server(self, self.hostfile_servers, server_group)
def test_many_servers(self):
"""
Test destroy on a large (relative) number of servers.
:avocado: tags=pool,pooldestroy,destroybig
"""
try:
# write out a hostfile_servers and start the servers with it
self.hostlist_servers = self.params.get("test_machines6",
'/run/hosts/')
hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.tmp)
self.agent_sessions = agent_utils.run_agent(
self.basepath, self.hostlist_servers)
server_utils.run_server(hostfile_servers, self.server_group,
self.basepath)
# parameters used in pool create
createmode = self.params.get("mode", '/run/poolparams/createmode/')
createuid = self.params.get("uid", '/run/poolparams/createuid/')
creategid = self.params.get("gid", '/run/poolparams/creategid/')
createsetid = self.params.get("setname",
'/run/poolparams/createset/')
createsize = self.params.get("size", '/run/poolparams/createsize/')
# initialize a python pool object then create the underlying
# daos storage
pool = DaosPool(self.context)
pool.create(createmode, createuid, creategid, createsize,
createsetid, None)
time.sleep(1)
# okay, get rid of it
pool.destroy(1)
except DaosApiError as excep:
print(excep)
print(traceback.format_exc())
self.fail("6 server test failed.\n")
except Exception as excep:
self.fail("Daos code segfaulted most likely. Error: %s" % excep)
# no matter what happens cleanup
finally:
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions)
server_utils.stop_server(hosts=self.hostlist_servers)
os.remove(hostfile_servers)
def setUp(self):
self.agent_sessions = None
self.pool = None
self.container = None
self.obj = None
self.ioreq = None
self.hostlist = None
self.hostfile = None
self.no_of_dkeys = None
self.no_of_akeys = None
self.array_size = None
self.record_length = None
with open('../../.build_vars.json') as json_f:
build_paths = json.load(json_f)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.context = DaosContext(build_paths['PREFIX'] + '/lib64/')
self.d_log = DaosLog(self.context)
self.hostlist = self.params.get("test_machines", '/run/hosts/*')
self.hostfile = write_host_file.write_host_file(
self.hostlist, self.workdir)
self.no_of_dkeys = self.params.get("no_of_dkeys", '/run/dkeys/*')[0]
self.no_of_akeys = self.params.get("no_of_akeys", '/run/akeys/*')[0]
self.array_size = self.params.get("size", '/array_size/')
self.record_length = self.params.get("length", '/run/record/*')
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist)
server_utils.run_server(self, self.hostfile, server_group)
self.pool = DaosPool(self.context)
self.pool.create(self.params.get("mode", '/run/pool/createmode/*'),
os.geteuid(), os.getegid(),
self.params.get("size", '/run/pool/createsize/*'),
self.params.get("setname", '/run/pool/createset/*'),
None)
self.pool.connect(2)
self.container = DaosContainer(self.context)
self.container.create(self.pool.handle)
self.container.open()
self.obj = DaosObj(self.context, self.container)
self.obj.create(objcls=1)
self.obj.open()
self.ioreq = IORequest(self.context,
self.container,
self.obj,
objtype=4)
def setUp(self):
super(PoolSvc, self).setUp()
self.pool = None
self.hostfile_servers = None
self.hostlist_servers = self.params.get("test_machines",
'/run/hosts/*')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
print("Host file is: {}".format(self.hostfile_servers))
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers, self.server_group,
self.basepath)
def setUp(self):
"""Define test setup to be done."""
self.log.info("<<setUp Started>> at %s", time.ctime())
# Start the daos_agents in the job scripts
self.setup_start_servers = True
self.setup_start_agents = False
super(Soak, self).setUp()
# Initialize loop param for all tests
self.loop = 1
self.exclude_slurm_nodes = []
# Setup logging directories for soak logfiles
# self.output dir is an avocado directory .../data/
self.log_dir = self.params.get("logdir", "/run/*")
self.outputsoakdir = self.outputdir + "/soak"
# Create the remote log directories on all client nodes
self.test_log_dir = self.log_dir + "/pass" + str(self.loop)
self.local_pass_dir = self.outputsoakdir + "/pass" + str(self.loop)
# Fail if slurm partition daos_client is not defined
if not self.partition_clients:
raise SoakTestError(
"<<FAILED: Partition is not correctly setup for daos "
"slurm partition>>")
# Check if the server nodes are in the client list;
# this will happen when only one partition is specified
for host_server in self.hostlist_servers:
if host_server in self.hostlist_clients:
self.hostlist_clients.remove(host_server)
self.exclude_slurm_nodes.append(host_server)
self.log.info("<<Updated hostlist_clients %s >>",
self.hostlist_clients)
# include test node for log cleanup; remove from client list
test_node = [socket.gethostname().split('.', 1)[0]]
if test_node[0] in self.hostlist_clients:
self.hostlist_clients.remove(test_node[0])
self.exclude_slurm_nodes.append(test_node[0])
self.log.info("<<Updated hostlist_clients %s >>",
self.hostlist_clients)
if not self.hostlist_clients:
self.fail("There are no nodes that are client only;"
"check if the partition also contains server nodes")
self.node_list = self.hostlist_clients + test_node
# Start agent on test node - need daos_agent yaml file
self.agent_sessions = run_agent(self, self.hostlist_servers, test_node)
def test_delete_doesnt_exist(self):
"""
Test destroying a pool uuid that doesn't exist.
:avocado: tags=pool,pooldestroy
"""
self.hostlist_servers = self.params.get("test_machines1",
'/run/hosts/')
hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.tmp)
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(hostfile_servers, self.server_group,
self.basepath)
setid = self.params.get("setname", '/run/setnames/validsetname/')
host = self.hostlist_servers[0]
try:
# randomly selected uuid, that is exceptionally unlikely to exist
bogus_uuid = '81ef94d7-a59d-4a5e-935b-abfbd12f2105'
# TODO make these params in the yaml
daosctl = self.basepath + '/install/bin/daosctl'
delete_cmd = ('{0} destroy-pool -i {1} -s {2}'.format(
daosctl, bogus_uuid, setid))
process.system(delete_cmd)
# the above command should fail resulting in an exception so if
# we get here the test has failed
self.fail("Pool {0} found on host {1} when not expected.\n".format(
bogus_uuid, host))
except Exception as _excep:
# expecting an exception so catch and pass the test
pass
# no matter what happens shutdown the server
finally:
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions)
server_utils.stop_server(hosts=self.hostlist_servers)
os.remove(hostfile_servers)
def setUp(self):
super(CartSelfTest, self).setUp()
self.agent_sessions = None
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
context = DaosContext(self.prefix + '/lib/')
self.d_log = DaosLog(context)
# self_test params
self.self_test_bin = os.path.join(self.prefix, "bin/self_test")
self.endpoint = self.params.get("endpoint", "/run/testparams/")
self.max_rpcs = self.params.get("max_inflight_rpcs",
"/run/testparams/")
self.repetitions = self.params.get("repetitions", "/run/testparams/")
self.message_size = (self.params.get(
"size", "/run/muxtestparams/message_size/*")[0])
self.share_addr = self.params.get("val",
"/run/muxtestparams/share_addr/*")[0]
self.env_dict = {
"CRT_PHY_ADDR_STR": "ofi+sockets",
"CRT_CTX_NUM": "8",
"OFI_INTERFACE": "eth0",
"CRT_CTX_SHARE_ADDR": str(self.share_addr)
}
self.env_list = []
for key, val in self.env_dict.items():
self.env_list.append("-x")
self.env_list.append("{0}={1}".format(key, val))
# daos server params
self.server_group = self.params.get("name", 'server_config',
'daos_server')
self.uri_file = os.path.join(self.basepath, "install", "tmp",
"uri.txt")
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers,
self.server_group,
self.basepath,
uri_path=self.uri_file,
env_dict=self.env_dict)
def setUp(self):
super(ObjectDataValidation, self).setUp()
self.agent_sessions = None
self.pool = None
self.container = None
self.obj = None
self.ioreq = None
self.hostlist = None
self.hostfile = None
self.no_of_dkeys = None
self.no_of_akeys = None
self.array_size = None
self.record_length = None
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.hostlist = self.params.get("test_servers", '/run/hosts/*')
self.hostfile = write_host_file.write_host_file(
self.hostlist, self.workdir)
self.no_of_dkeys = self.params.get("no_of_dkeys", '/run/dkeys/*')[0]
self.no_of_akeys = self.params.get("no_of_akeys", '/run/akeys/*')[0]
self.array_size = self.params.get("size", '/array_size/')
self.record_length = self.params.get("length", '/run/record/*')
self.agent_sessions = agent_utils.run_agent(self, self.hostlist)
server_utils.run_server(self, self.hostfile, server_group)
self.pool = DaosPool(self.context)
self.pool.create(self.params.get("mode", '/run/pool/createmode/*'),
os.geteuid(), os.getegid(),
self.params.get("size", '/run/pool/createsize/*'),
self.params.get("setname", '/run/pool/createset/*'),
None)
self.pool.connect(2)
self.container = DaosContainer(self.context)
self.container.create(self.pool.handle)
self.container.open()
self.obj = DaosObj(self.context, self.container)
self.obj.create(objcls=1)
self.obj.open()
self.ioreq = IORequest(self.context,
self.container,
self.obj,
objtype=4)
def setUp(self):
self.agent_sessions = None
# there is a presumption that this test lives in a specific spot
# in the repo
# get paths from the build_vars generated by build
self.hostlist_servers = None
with open('../../../.build_vars.json') as f_open:
build_paths = json.load(f_open)
basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
self.hostfile = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile, server_group, basepath)
self.dmg = basepath + '/install/bin/dmg'
def setUp(self):
super(BadConnectTest, self).setUp()
self.agent_sessions = None
self.hostlist_servers = None
self.hostlist_servers = self.params.get("test_servers", '/run/hosts/')
# NULL is causing connect to blow up so skip that test for now
uuidlist = self.params.get("uuid", '/run/connecttests/UUID/*/')
connectuuid = uuidlist[0]
if connectuuid == 'NULLPTR':
self.cancel("skipping null pointer test until DAOS-1781 is fixed")
# launch the server
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(self,
self.hostlist_servers)
server_utils.run_server(self, self.hostfile_servers, server_group)
def setUp(self):
# there is a presumption that this test lives in a specific
# spot in the repo
with open('../../../.build_vars.json') as build_file:
build_paths = json.load(build_file)
basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers, server_group, basepath)
self.daosctl = basepath + '/install/bin/daosctl'
def setUp(self):
self.agent_sessions = None
# get paths from the build_vars generated by build
with open(os.path.join(os.path.dirname(os.path.realpath(__file__)),
"../../../../.build_vars.json")) as build_file:
build_paths = json.load(build_file)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
server_group = self.params.get("name",
'/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers, server_group,
self.basepath)
def setUp(self):
"""
test setup
"""
super(MultiServerCreateDeleteTest, self).setUp()
self.agent_sessions = None
# there is a presumption that this test lives in a specific spot
# in the repo
# get paths from the build_vars generated by build
self.hostlist_servers = None
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
self.hostfile = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.agent_sessions = agent_utils.run_agent(self,
self.hostlist_servers)
server_utils.run_server(self, self.hostfile, server_group)
self.dmg = self.basepath + '/install/bin/dmg'
def setUp(self):
self.agent_sessions = None
self.pool = None
self.hostlist_servers = None
self.hostfile_clients = None
self.hostfile_servers = None
self.out_queue = None
self.pool_connect = True
with open('../../../.build_vars.json') as json_f:
build_paths = json.load(json_f)
self.basepath = os.path.normpath(build_paths['PREFIX'] + "/../")
self.server_group = self.params.get("name", '/server_config/',
'daos_server')
self.context = DaosContext(build_paths['PREFIX'] + '/lib/')
self.d_log = DaosLog(self.context)
self.hostlist_servers = self.params.get("servers", '/run/hosts/*')
self.hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
self.hostlist_clients = self.params.get("clients", '/run/hosts/*')
self.hostfile_clients = write_host_file.write_host_file(
self.hostlist_clients, self.workdir)
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers,
self.hostlist_clients)
server_utils.run_server(self.hostfile_servers, self.server_group,
self.basepath)
self.pool = DaosPool(self.context)
self.pool.create(self.params.get("mode", '/run/pool/createmode/*'),
os.geteuid(),
os.getegid(),
self.params.get("size", '/run/pool/createsize/*'),
self.params.get("setname", '/run/pool/createset/*'),
nvme_size=self.params.get("size",
'/run/pool/nvmesize/*'))
def setUp(self):
super(Romio, self).setUp()
self.agent_sessions = None
# get paths from the build_vars generated by build
self.server_group = self.params.get("name", '/server_config/',
'daos_server')
self.hostlist_servers = self.params.get("test_servers", '/run/hosts/')
self.hostfile_servers = (write_host_file.write_host_file(
self.hostlist_servers, self.workdir))
print("Host file servers is: {}".format(self.hostfile_servers))
self.hostlist_clients = self.params.get("test_clients", '/run/hosts/')
self.hostfile_clients = (write_host_file.write_host_file(
self.hostlist_clients, self.workdir))
print("Host file clients is: {}".format(self.hostfile_clients))
# start servers
self.agent_sessions = agent_utils.run_agent(self,
self.hostlist_servers,
self.hostlist_clients)
server_utils.run_server(self, self.hostfile_servers, self.server_group)
self.mpio = None
def test_metadata_server_restart(self):
"""
Test ID: DAOS-1512
Test Description: This test will verify 2000 IOR small size container
after server restart. Test will write IOR in 5
different threads for faster execution time. Each
thread will create 400 (8bytes) containers to the
same pool. Restart the servers, read IOR container
file written previously and validate data integrity
by using IOR option "-R -G 1".
:avocado: tags=metadata,metadata_ior,nvme,small
"""
self.pool_connect = False
files_per_thread = 400
total_ior_threads = 5
threads = []
ior_args = {}
createsvc = self.params.get("svcn", '/run/pool/createsvc/')
svc_list = ""
for i in range(createsvc):
svc_list += str(int(self.pool.svc.rl_ranks[i])) + ":"
svc_list = svc_list[:-1]
ior_args['client_hostfile_servers'] = self.hostfile_clients
ior_args['pool_uuid'] = self.pool.get_uuid_str()
ior_args['svc_list'] = svc_list
ior_args['basepath'] = self.basepath
ior_args['server_group'] = self.server_group
ior_args['tmp_dir'] = self.workdir
ior_args['iorwriteflags'] = self.params.get("F",
'/run/ior/iorwriteflags/')
ior_args['iorreadflags'] = self.params.get("F",
'/run/ior/iorreadflags/')
ior_args['iteration'] = self.params.get("iter", '/run/ior/iteration/')
ior_args['stripe_size'] = self.params.get("s", '/run/ior/stripesize/*')
ior_args['stripe_count'] = self.params.get("c",
'/run/ior/stripecount/')
ior_args['async_io'] = self.params.get("a", '/run/ior/asyncio/')
ior_args['object_class'] = self.params.get("o",
'/run/ior/objectclass/')
ior_args['slots'] = self.params.get("slots", '/run/ior/clientslots/*')
ior_args['files_per_thread'] = files_per_thread
self.out_queue = Queue.Queue()
#IOR write threads
for i in range(total_ior_threads):
threads.append(
threading.Thread(target=ior_runner_thread,
args=(self.out_queue, "Thread-{}".format(i),
"write"),
kwargs=ior_args))
if self.thread_control(threads, "write") == "FAIL":
self.d_log.error(" IOR write Thread FAIL")
self.fail(" IOR write Thread FAIL")
#Server Restart
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions, self.hostlist_clients)
server_utils.stop_server(hosts=self.hostlist_servers)
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_clients,
self.hostlist_servers)
server_utils.run_server(self.hostfile_servers, self.server_group,
self.basepath)
#Read IOR with verification with same number of threads
threads = []
for i in range(total_ior_threads):
threads.append(
threading.Thread(target=ior_runner_thread,
args=(self.out_queue, "Thread-{}".format(i),
"read"),
kwargs=ior_args))
if self.thread_control(threads, "read") == "FAIL":
self.d_log.error(" IOR write Thread FAIL")
self.fail(" IOR read Thread FAIL")
def setUp(self):
"""Set up each test case."""
super(TestWithServers, self).setUp()
self.server_group = self.params.get("name", "/server_config/",
"daos_server")
# Determine which hosts to use as servers and optionally clients.
# Support the use of a host type count to test with subsets of the
# specified hosts lists
test_servers = self.params.get("test_servers", "/run/hosts/*")
test_clients = self.params.get("test_clients", "/run/hosts/*")
server_count = self.params.get("server_count", "/run/hosts/*")
client_count = self.params.get("client_count", "/run/hosts/*")
# If server or client host list are defined through valid slurm
# partition names override any hosts specified through lists.
test_servers, self.partition_servers = self.get_partition_hosts(
"server_partition", test_servers)
test_clients, self.partition_clients = self.get_partition_hosts(
"client_partition", test_clients)
# Supported combinations of yaml hosts arguments:
# - test_servers [+ server_count]
# - test_servers [+ server_count] + test_clients [+ client_count]
if test_servers and test_clients:
self.hostlist_servers = test_servers[:server_count]
self.hostlist_clients = test_clients[:client_count]
elif test_servers:
self.hostlist_servers = test_servers[:server_count]
self.log.info("hostlist_servers: %s", self.hostlist_servers)
self.log.info("hostlist_clients: %s", self.hostlist_clients)
# Find a configuration that meets the test requirements
self.config = Configuration(self.params,
self.hostlist_servers,
debug=self.debug)
if not self.config.set_config(self):
self.cancel("Test requirements not met!")
# If a specific count is specified, verify enough servers/clients are
# specified to satisy the count
host_count_checks = (
("server", server_count,
len(self.hostlist_servers) if self.hostlist_servers else 0),
("client", client_count,
len(self.hostlist_clients) if self.hostlist_clients else 0))
for host_type, expected_count, actual_count in host_count_checks:
if expected_count:
self.assertEqual(
expected_count, actual_count,
"Test requires {} {}; {} specified".format(
expected_count, host_type, actual_count))
# Create host files
if self.hostlist_clients:
self.hostfile_clients = write_host_file.write_host_file(
self.hostlist_clients, self.workdir,
self.hostfile_clients_slots)
# Start the clients (agents)
if self.setup_start_agents:
self.agent_sessions = agent_utils.run_agent(
self, self.hostlist_servers, self.hostlist_clients)
# Start the servers
if self.setup_start_servers:
self.start_servers()
def test_container_basics(self):
"""
Test basic container create/destroy/open/close/query. Nothing fancy
just making sure they work at a rudimentary level
:avocado: tags=container,containercreate,containerdestroy,basecont
"""
pool = None
hostlist = None
try:
hostlist = self.params.get("test_machines", '/run/hosts/*')
hostfile = write_host_file.write_host_file(hostlist,
self.workdir)
self.agent_sessions = agent_utils.run_agent(self.basepath, hostlist)
server_utils.run_server(hostfile, self.server_group, self.basepath)
# give it time to start
time.sleep(2)
# parameters used in pool create
createmode = self.params.get("mode", '/run/conttests/createmode/')
createuid = self.params.get("uid", '/run/conttests/createuid/')
creategid = self.params.get("gid", '/run/conttests/creategid/')
createsetid = self.params.get("setname",
'/run/conttests/createset/')
createsize = self.params.get("size", '/run/conttests/createsize/')
# initialize a python pool object then create the underlying
# daos storage
pool = DaosPool(self.context)
pool.create(createmode, createuid, creategid,
createsize, createsetid, None)
# need a connection to create container
pool.connect(1 << 1)
# create a container
container = DaosContainer(self.context)
container.create(pool.handle)
# now open it
container.open()
# do a query and compare the UUID returned from create with
# that returned by query
container.query()
if container.get_uuid_str() != c_uuid_to_str(
container.info.ci_uuid):
self.fail("Container UUID did not match the one in info'n")
container.close()
# wait a few seconds and then destroy
time.sleep(5)
container.destroy()
except DaosApiError as excep:
print(excep)
print(traceback.format_exc())
self.fail("Test was expected to pass but it failed.\n")
except Exception as excep:
self.fail("Daos code segfaulted most likely, error: %s" % excep)
finally:
# cleanup the pool
if pool is not None:
pool.disconnect()
pool.destroy(1)
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions)
server_utils.stop_server(hosts=hostlist)
def test_destroy_async(self):
"""
Performn destroy asynchronously, successful and failed.
:avocado: tags=pool,pooldestroy,destroyasync
"""
global GLOB_SIGNAL
global GLOB_RC
try:
# write out a hostfile_servers and start the servers with it
self.hostlist_servers = self.params.get("test_machines1",
'/run/hosts/')
hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.tmp)
self.agent_sessions = agent_utils.run_agent(
self.basepath, self.hostlist_servers)
server_utils.run_server(hostfile_servers, self.server_group,
self.basepath)
# parameters used in pool create
createmode = self.params.get("mode", '/run/poolparams/createmode/')
createuid = self.params.get("uid", '/run/poolparams/createuid/')
creategid = self.params.get("gid", '/run/poolparams/creategid/')
createsetid = self.params.get("setname",
'/run/poolparams/createset/')
createsize = self.params.get("size", '/run/poolparams/createsize/')
# initialize a python pool object then create the underlying
# daos storage
pool = DaosPool(self.context)
pool.create(createmode, createuid, creategid, createsize,
createsetid, None)
# allow the callback to tell us when its been called
GLOB_SIGNAL = threading.Event()
# blow it away but this time get return code via callback function
pool.destroy(1, cb_func)
# wait for callback
GLOB_SIGNAL.wait()
if GLOB_RC != 0:
self.fail("RC not as expected in async test")
# recreate the pool, reset the signal, shutdown the
# servers so call will fail and then check rc in the callback
pool.create(createmode, createuid, creategid, createsize,
createsetid, None)
GLOB_SIGNAL = threading.Event()
GLOB_RC = -9900000
server_utils.stop_server(hosts=self.hostlist_servers)
pool.destroy(1, cb_func)
# wait for callback, expecting a timeout since servers are down
GLOB_SIGNAL.wait()
if GLOB_RC != -1011:
self.fail("RC not as expected in async test")
except DaosApiError as excep:
print(excep)
print(traceback.format_exc())
self.fail("destroy async test failed.\n")
except Exception as excep:
self.fail("Daos code segfaulted most likely. Error: %s" % excep)
# no matter what happens cleanup
finally:
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions)
server_utils.stop_server(hosts=self.hostlist_servers)
os.remove(hostfile_servers)
def test_metadata_server_restart(self):
"""JIRA ID: DAOS-1512.
Test Description:
This test will verify 2000 IOR small size container after server
restart. Test will write IOR in 5 different threads for faster
execution time. Each thread will create 400 (8bytes) containers to
the same pool. Restart the servers, read IOR container file written
previously and validate data integrity by using IOR option
"-R -G 1".
Use Cases:
?
:avocado: tags=metadata,metadata_ior,nvme,small
"""
files_per_thread = 400
total_ior_threads = 5
self.out_queue = Queue.Queue()
processes = self.params.get("slots", "/run/ior/clientslots/*")
list_of_uuid_lists = [[
str(uuid.uuid4()) for _ in range(files_per_thread)
] for _ in range(total_ior_threads)]
# Launch threads to run IOR to write data, restart the agents and
# servers, and then run IOR to read the data
for operation in ("write", "read"):
# Create the IOR threads
threads = []
for index in range(total_ior_threads):
# Define the arguments for the ior_runner_thread method
ior_cmd = IorCommand()
ior_cmd.get_params(self)
ior_cmd.set_daos_params(self.server_group, self.pool)
ior_cmd.flags.value = self.params.get(
"F", "/run/ior/ior{}flags/".format(operation))
# Add a thread for these IOR arguments
threads.append(
threading.Thread(target=ior_runner_thread,
kwargs={
"ior_cmd": ior_cmd,
"uuids": list_of_uuid_lists[index],
"mgr": self.orterun,
"attach": self.tmp,
"hostfile": self.hostfile_clients,
"procs": processes,
"results": self.out_queue
}))
self.log.info("Creatied %s thread %s with container uuids %s",
operation, index, list_of_uuid_lists[index])
# Launch the IOR threads
if self.thread_control(threads, operation) == "FAIL":
self.d_log.error("IOR {} Thread FAIL".format(operation))
self.fail("IOR {} Thread FAIL".format(operation))
# Restart the agents and servers after the write / before the read
if operation == "write":
# Stop the agents and servers
if self.agent_sessions:
stop_agent(self.agent_sessions, self.hostlist_clients)
stop_server(hosts=self.hostlist_servers)
# Start the agents
self.agent_sessions = run_agent(self.basepath,
self.hostlist_clients,
self.hostlist_servers)
# Start the servers
run_server(self.hostfile_servers,
self.server_group,
self.basepath,
clean=False)
def test_rebuild_with_io(self):
"""
Test ID: Rebuild-003
Test Description: Trigger a rebuild while I/O is ongoing.
Use Cases:
-- single pool, single client performing continous read/write/verify
sequence while failure/rebuild is triggered in another process
:avocado: tags=pool,rebuild,rebuildwithio
"""
# the rebuild tests need to redo this stuff each time so not in setup
# as it usually would be
server_group = self.params.get("name", '/server_config/',
'daos_server')
self.hostlist_servers = self.params.get("test_machines", '/run/hosts/')
hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.workdir)
try:
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(hostfile_servers, server_group,
self.basepath)
# use the uid/gid of the user running the test, these should
# be perfectly valid
createuid = os.geteuid()
creategid = os.getegid()
# parameters used in pool create that are in yaml
createmode = self.params.get("mode", '/run/testparams/createmode/')
createsetid = self.params.get("setname",
'/run/testparams/createset/')
createsize = self.params.get("size", '/run/testparams/createsize/')
# initialize a python pool object then create the underlying
# daos storage
pool = DaosPool(self.context)
pool.create(createmode, createuid, creategid,
createsize, createsetid, None)
pool.connect(1 << 1)
container = DaosContainer(self.context)
container.create(pool.handle)
container.open()
# get pool status and make sure it all looks good before we start
pool.pool_query()
if pool.pool_info.pi_ndisabled != 0:
self.fail("Number of disabled targets reporting incorrectly.\n")
if pool.pool_info.pi_rebuild_st.rs_errno != 0:
self.fail("Rebuild error but rebuild hasn't run.\n")
if pool.pool_info.pi_rebuild_st.rs_done != 1:
self.fail("Rebuild is running but device hasn't failed yet.\n")
if pool.pool_info.pi_rebuild_st.rs_obj_nr != 0:
self.fail("Rebuilt objs not zero.\n")
if pool.pool_info.pi_rebuild_st.rs_rec_nr != 0:
self.fail("Rebuilt recs not zero.\n")
dummy_pool_version = pool.pool_info.pi_rebuild_st.rs_version
# do I/O for 30 seconds
dummy_bw = io_utilities.continuous_io(container, 30)
# trigger the rebuild
rank = self.params.get("rank", '/run/testparams/ranks/*')
server = DaosServer(self.context, server_group, rank)
server.kill(1)
pool.exclude([rank])
# do another 30 seconds of I/O,
# waiting for some improvements in server bootstrap
# at which point we can move the I/O to a separate client and
# really pound it with I/O
dummy_bw = io_utilities.continuous_io(container, 30)
# wait for the rebuild to finish
while True:
pool.pool_query()
if pool.pool_info.pi_rebuild_st.rs_done == 1:
break
else:
time.sleep(2)
# check rebuild statistics
if pool.pool_info.pi_ndisabled != 1:
self.fail("Number of disabled targets reporting incorrectly: {}"
.format(pool.pool_info.pi_ndisabled))
if pool.pool_info.pi_rebuild_st.rs_errno != 0:
self.fail("Rebuild error reported: {}".format(
pool.pool_info.pi_rebuild_st.rs_errno))
if pool.pool_info.pi_rebuild_st.rs_obj_nr <= 0:
self.fail("No objects have been rebuilt.")
if pool.pool_info.pi_rebuild_st.rs_rec_nr <= 0:
self.fail("No records have been rebuilt.")
except (ValueError, DaosApiError) as excep:
print(excep)
print(traceback.format_exc())
self.fail("Expecting to pass but test has failed.\n")
finally:
# wait for the I/O process to finish
try:
server_utils.stop_server(hosts=self.hostlist_servers)
os.remove(hostfile_servers)
# really make sure everything is gone
check_for_pool.cleanup_pools(self.hostlist_servers)
finally:
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions)
server_utils.kill_server(self.hostlist_servers)
def test_simple_delete(self):
"""
Test destroying a pool created on a single server, nobody is using
the pool, force is not needed.
:avocado: tags=pool,pooldestroy,quick
"""
self.hostlist_servers = self.params.get("test_machines1",
'/run/hosts/')
hostfile_servers = write_host_file.write_host_file(
self.hostlist_servers, self.tmp)
self.agent_sessions = agent_utils.run_agent(self.basepath,
self.hostlist_servers)
server_utils.run_server(hostfile_servers, self.server_group,
self.basepath)
setid = self.params.get("setname", '/run/setnames/validsetname/')
try:
# use the uid/gid of the user running the test, these should
# be perfectly valid
uid = os.geteuid()
gid = os.getegid()
# TODO make these params in the yaml
daosctl = self.basepath + '/install/bin/daosctl'
create_cmd = ('{0} create-pool -m {1} -u {2} -g {3} -s {4}'.format(
daosctl, 0x731, uid, gid, setid))
uuid_str = """{0}""".format(process.system_output(create_cmd))
print("uuid is {0}\n".format(uuid_str))
host = self.hostlist_servers[0]
exists = check_for_pool.check_for_pool(host, uuid_str)
if exists != 0:
self.fail("Pool {0} not found on host {1}.\n".format(
uuid_str, host))
delete_cmd = ('{0} destroy-pool -i {1} -s {2}'.format(
daosctl, uuid_str, setid))
process.system(delete_cmd)
exists = check_for_pool.check_for_pool(host, uuid_str)
if exists == 0:
self.fail(
"Pool {0} found on host {1} when not expected.\n".format(
uuid_str, host))
except Exception as excep:
print(excep)
print(traceback.format_exc())
self.fail("Expecting to pass but test has failed.\n")
# no matter what happens shutdown the server
finally:
try:
os.remove(hostfile_servers)
finally:
if self.agent_sessions:
agent_utils.stop_agent(self.agent_sessions)
server_utils.stop_server(hosts=self.hostlist_servers)
