def test_add_node(self):
"""
Test for adding worker nodes to the cluster while IOs
"""
new_nodes = 3
if config.ENV_DATA['platform'].lower() in constants.CLOUD_PLATFORMS:
dt = config.ENV_DATA['deployment_type']
if dt == 'ipi':
machines = machine_utils.get_machinesets()
logger.info(
f'The worker nodes number before expansion {len(helpers.get_worker_nodes())}'
)
for machine in machines:
add_new_node_and_label_it(machine)
logger.info(
f'The worker nodes number after expansion {len(helpers.get_worker_nodes())}'
)
else:
logger.info(
f'The worker nodes number before expansion {len(helpers.get_worker_nodes())}'
)
if config.ENV_DATA.get('rhel_workers'):
node_type = constants.RHEL_OS
else:
node_type = constants.RHCOS
assert add_new_node_and_label_upi(node_type,
new_nodes), "Add node failed"
logger.info(
f'The worker nodes number after expansion {len(helpers.get_worker_nodes())}'
)
elif config.ENV_DATA['platform'].lower() == constants.VSPHERE_PLATFORM:
pytest.skip("Skipping add node in Vmware platform due to "
"https://bugzilla.redhat.com/show_bug.cgi?id=1844521")
def test_add_node(self):
"""
Test for adding worker nodes to the cluster while IOs
"""
new_nodes = 3
if config.ENV_DATA["platform"].lower() in constants.CLOUD_PLATFORMS:
dt = config.ENV_DATA["deployment_type"]
if dt == "ipi":
machines = machine_utils.get_machinesets()
logger.info(
f"The worker nodes number before expansion {len(node.get_worker_nodes())}"
)
for machine in machines:
add_new_node_and_label_it(machine)
logger.info(
f"The worker nodes number after expansion {len(node.get_worker_nodes())}"
)
else:
logger.info(
f"The worker nodes number before expansion {len(node.get_worker_nodes())}"
)
if config.ENV_DATA.get("rhel_workers"):
node_type = constants.RHEL_OS
else:
node_type = constants.RHCOS
assert add_new_node_and_label_upi(node_type,
new_nodes), "Add node failed"
logger.info(
f"The worker nodes number after expansion {len(node.get_worker_nodes())}"
)
elif config.ENV_DATA["platform"].lower() == constants.VSPHERE_PLATFORM:
pytest.skip("Skipping add node in Vmware platform due to "
"https://bugzilla.redhat.com/show_bug.cgi?id=1844521")
# Issue to remove skip code https://github.com/red-hat-storage/ocs-ci/issues/2403
# logger.info(f'The worker nodes number before expansion {len(node.get_worker_nodes())}')
# if config.ENV_DATA.get('rhel_user'):
# pytest.skip("Skipping add RHEL node, code unavailable")
# node_type = constants.RHCOS
# assert add_new_node_and_label_upi(node_type, new_nodes), "Add node failed"
# logger.info(f'The worker nodes number after expansion {len(node.get_worker_nodes())}')
ceph_cluster_obj = CephCluster()
assert ceph_cluster_obj.wait_for_rebalance(
timeout=3600), "Data re-balance failed to complete"
def test_add_node_aws(self):
"""
Test for adding worker nodes to the cluster while IOs
"""
dt = config.ENV_DATA['deployment_type']
if dt == 'ipi':
machines = machine_utils.get_machinesets()
logger.info(f'The worker nodes number before expansion {len(helpers.get_worker_nodes())}')
for machine in machines:
add_new_node_and_label_it(machine)
logger.info(f'The worker nodes number after expansion {len(helpers.get_worker_nodes())}')
else:
new_nodes = 3
logger.info(f'The worker nodes number before expansion {len(helpers.get_worker_nodes())}')
if config.ENV_DATA.get('rhel_workers'):
node_type = constants.RHEL_OS
else:
node_type = constants.RHCOS
assert add_new_node_and_label_upi(node_type, new_nodes), "Add node failed"
logger.info(f'The worker nodes number after expansion {len(helpers.get_worker_nodes())}')
def test_add_node(self):
"""
Test for adding worker nodes to the cluster while IOs
"""
dt = config.ENV_DATA['deployment_type']
if dt == 'ipi':
before_replica_counts = dict()
count = 2
machines = machine_utils.get_machinesets()
for machine in machines:
before_replica_counts.update(
{machine: machine_utils.get_replica_count(machine)})
worker_nodes_before = helpers.get_worker_nodes()
logger.info(
f'The worker nodes number before adding a new node is {len(worker_nodes_before)}'
)
after_replica_counts = dict()
for machine in machines:
machine_utils.add_node(machine, count=count)
after_replica_counts.update(({
machine:
machine_utils.get_replica_count(machine)
}))
logger.info(after_replica_counts)
for sample in TimeoutSampler(timeout=300,
sleep=3,
func=helpers.get_worker_nodes):
if len(sample) == count * len(machines):
break
worker_nodes_after = helpers.get_worker_nodes()
logger.info(
f'The worker nodes number after adding a new node is {len(worker_nodes_after)}'
)
wait_for_nodes_status(node_names=worker_nodes_after,
status=constants.NODE_READY)
else:
pytest.skip("UPI not yet supported")
def factory(ocs_nodes=False, node_count=3, taint_label=None):
"""
Args:
ocs_nodes (bool): True if new nodes are OCS, False otherwise
node_count (int): Number of nodes to be added
taint_label (str): Taint label to be added
"""
new_nodes = []
if config.ENV_DATA["platform"].lower() in constants.CLOUD_PLATFORMS:
dt = config.ENV_DATA["deployment_type"]
if dt == "ipi":
machines = machine_utils.get_machinesets()
log.info(
f"The worker nodes number before expansion {len(get_worker_nodes())}"
)
for machine in machines:
new_nodes.append(
add_new_node_and_label_it(
machine, mark_for_ocs_label=ocs_nodes))
log.info(
f"The worker nodes number after expansion {len(get_worker_nodes())}"
)
else:
log.info(
f"The worker nodes number before expansion {len(get_worker_nodes())}"
)
if config.ENV_DATA.get("rhel_workers"):
node_type = constants.RHEL_OS
else:
node_type = constants.RHCOS
new_nodes.append(
add_new_node_and_label_upi(node_type,
node_count,
mark_for_ocs_label=ocs_nodes))
log.info(
f"The worker nodes number after expansion {len(get_worker_nodes())}"
)
elif config.ENV_DATA["platform"].lower() == constants.VSPHERE_PLATFORM:
log.info(
f"The worker nodes number before expansion {len(get_worker_nodes())}"
)
if config.ENV_DATA.get("rhel_user"):
node_type = constants.RHEL_OS
else:
node_type = constants.RHCOS
new_nodes.append(
add_new_node_and_label_upi(node_type,
node_count,
mark_for_ocs_label=ocs_nodes))
log.info(
f"The worker nodes number after expansion {len(get_worker_nodes())}"
)
nodes = [node for sublist in new_nodes for node in sublist]
if taint_label:
taint_nodes(nodes=nodes,
taint_label=taint_label), "Failed to taint nodes"
log.info(f"Successfully Tainted nodes {new_nodes} with {taint_label}")
def factory(ocs_nodes=False, node_count=3, taint_label=None):
"""
Args:
ocs_nodes (bool): True if new nodes are OCS, False otherwise
node_count (int): Number of nodes to be added
taint_label (str): Taint label to be added
"""
new_nodes = []
if config.ENV_DATA["platform"].lower() in constants.CLOUD_PLATFORMS:
dt = config.ENV_DATA["deployment_type"]
if dt == "ipi":
machines = machine_utils.get_machinesets()
log.info(
f"The worker nodes number before expansion {len(get_worker_nodes())}"
)
for machine in machines:
new_nodes.append(
add_new_node_and_label_it(
machine, mark_for_ocs_label=ocs_nodes))
log.info(
f"The worker nodes number after expansion {len(get_worker_nodes())}"
)
else:
log.info(
f"The worker nodes number before expansion {len(get_worker_nodes())}"
)
if config.ENV_DATA.get("rhel_workers"):
node_type = constants.RHEL_OS
else:
node_type = constants.RHCOS
new_nodes.append(
add_new_node_and_label_upi(node_type,
node_count,
mark_for_ocs_label=ocs_nodes))
log.info(
f"The worker nodes number after expansion {len(get_worker_nodes())}"
)
elif config.ENV_DATA["platform"].lower() == constants.VSPHERE_PLATFORM:
pytest.skip("Skipping add node in Vmware platform due to "
"https://bugzilla.redhat.com/show_bug.cgi?id=1844521")
# Issue to remove skip code https://github.com/red-hat-storage/ocs-ci/issues/2403
# log.info(
# f"The worker nodes number before expansion {len(get_worker_nodes())}"
# )
# if config.ENV_DATA.get("rhel_user"):
# pytest.skip("Skipping add RHEL node, code unavailable")
# node_type = constants.RHCOS
# new_nodes.append(
# add_new_node_and_label_upi(node_type, num_nodes=node_count)
# )
# log.info(
# f"The worker nodes number after expansion {len(get_worker_nodes())}"
# )
nodes = [node for sublist in new_nodes for node in sublist]
if taint_label:
taint_nodes(nodes=nodes,
taint_label=taint_label), "Failed to taint nodes"
log.info(f"Successfully Tainted nodes {new_nodes} with {taint_label}")
def setup(
self,
request,
scenario,
nodes,
multi_pvc_factory,
service_account_factory,
dc_pod_factory,
):
"""
Identify the nodes and start multiple dc pods for the test
Args:
scenario (str): Scenario of app pods running on OCS or dedicated nodes
(eg., 'colocated', 'dedicated')
nodes: A fixture to get instance of the relevant platform nodes class
multi_pvc_factory: A fixture create a set of new PVCs
service_account_factory: A fixture to create a service account
dc_pod_factory: A fixture to create dc pod
Returns:
list: dc pod objs
"""
worker_nodes = node.get_worker_nodes()
ocs_nodes = machine.get_labeled_nodes(constants.OPERATOR_NODE_LABEL)
non_ocs_nodes = list(set(worker_nodes) - set(ocs_nodes))
def finalizer():
helpers.remove_label_from_worker_node(node_list=worker_nodes,
label_key="nodetype")
# Check ceph health
ceph_health_check(tries=80)
request.addfinalizer(finalizer)
if (scenario == "dedicated") and len(non_ocs_nodes) == 0:
if config.ENV_DATA.get("deployment_type").lower() == "ipi":
machines = machine.get_machinesets()
node.add_new_node_and_label_it(machines[0],
num_nodes=1,
mark_for_ocs_label=False)
else:
if (config.ENV_DATA.get("platform").lower() ==
constants.VSPHERE_PLATFORM):
pytest.skip(
"Skipping add node in VSPHERE due to https://bugzilla.redhat.com/show_bug.cgi?id=1844521"
)
is_rhel = config.ENV_DATA.get(
"rhel_workers") or config.ENV_DATA.get("rhel_user")
node_type = constants.RHEL_OS if is_rhel else constants.RHCOS
node.add_new_node_and_label_upi(node_type=node_type,
num_nodes=1,
mark_for_ocs_label=False)
non_ocs_nodes = list(set(node.get_worker_nodes()) - set(ocs_nodes))
app_pod_nodes = ocs_nodes if (scenario
== "colocated") else non_ocs_nodes
# Label nodes to be able to run app pods
helpers.label_worker_node(node_list=app_pod_nodes,
label_key="nodetype",
label_value="app-pod")
access_modes_rbd = [
constants.ACCESS_MODE_RWO,
f"{constants.ACCESS_MODE_RWX}-Block",
]
access_modes_cephfs = [
constants.ACCESS_MODE_RWO, constants.ACCESS_MODE_RWX
]
pvcs_rbd = multi_pvc_factory(
interface=constants.CEPHBLOCKPOOL,
size=self.pvc_size,
access_modes=access_modes_rbd,
status=constants.STATUS_BOUND,
num_of_pvc=len(access_modes_rbd),
)
project = pvcs_rbd[0].project
pvcs_cephfs = multi_pvc_factory(
interface=constants.CEPHFILESYSTEM,
project=project,
size=self.pvc_size,
access_modes=access_modes_cephfs,
status=constants.STATUS_BOUND,
num_of_pvc=len(access_modes_cephfs),
)
pvcs = pvcs_cephfs + pvcs_rbd
# Set volume mode on PVC objects
for pvc_obj in pvcs:
pvc_info = pvc_obj.get()
setattr(pvc_obj, "volume_mode", pvc_info["spec"]["volumeMode"])
sa_obj = service_account_factory(project=project)
pods = []
# Create pods
for pvc_obj in pvcs:
if constants.CEPHFS_INTERFACE in pvc_obj.storageclass.name:
interface = constants.CEPHFILESYSTEM
else:
interface = constants.CEPHBLOCKPOOL
num_pods = 2 if pvc_obj.access_mode == constants.ACCESS_MODE_RWX else 1
logger.info("Creating app pods")
for _ in range(num_pods):
pods.append(
dc_pod_factory(
interface=interface,
pvc=pvc_obj,
node_selector={"nodetype": "app-pod"},
raw_block_pv=pvc_obj.volume_mode == "Block",
sa_obj=sa_obj,
))
logger.info(
f"Created {len(pods)} pods using {len(pvcs_cephfs)} cephfs, {len(pvcs_rbd)} rbd PVCs."
)
return pods