Provider mode deployment with Hosted clusters (#10120)

Changes:
    fix leftover, to prevent issues such as https://ocs4-jenkins-csb-odf-qe.apps.ocp-c1.prod.psi.redhat.com/job/qe-deploy-ocs-cluster/39701/consoleFull (metallb installation and latest OCP fetch from release notes)
    changed made ODF and Hosted cluster installation done by completing jenkins stage 1. Deploy OCP cluster stage 2. Install OD, dependencies and Hosted Cluster
    clean ODF provider and native client validation. Fixed some issues failing on waiting just deployed resources
    ODF with "latest-4.16" now can be a version, set in

clusters: 
   hcp416-bm2-a:
     hosted_odf_version: "latest-4.16" 

"--olm-disable-default-sources" parametrization added
FIXES
prevent failures on converged deployment Provider and Hosted cluster #10160

Signed-off-by: Daniel Osypenko <[email protected]>

ocs_ci/deployment/deployment.py

@@ -580,6 +580,19 @@ def do_deploy_odf_provider_mode(self):
         ):
             storage_client_deployment_obj.provider_and_native_client_installation()
 
+    def do_deploy_cnv(self):
+        """
+        Deploy CNV
+        We run it in OCP deployment stage, hence `ship_ocs_deployment` is set True.
+        When we run it in OCS deployment stage, the `skip_ocs_deployment` is set to False automatically and
+        second installation does not happen.
+        """
+        if (
+            config.DEPLOYMENT.get("cnv_deployment")
+            and config.ENV_DATA["skip_ocs_deployment"]
+        ):
+            CNVInstaller().deploy_cnv()
+
     def do_deploy_hosted_clusters(self):
         """
         Deploy Hosted cluster(s)
@@ -656,8 +669,7 @@ def deploy_cluster(self, log_cli_level="DEBUG"):
         self.do_deploy_rdr()
         self.do_deploy_fusion()
         self.do_deploy_odf_provider_mode()
-        if config.DEPLOYMENT.get("cnv_deployment"):
-            CNVInstaller().deploy_cnv()
+        self.do_deploy_cnv()
         self.do_deploy_hosted_clusters()
 
     def get_rdr_conf(self):
@@ -921,7 +933,7 @@ def wait_for_subscription(self, subscription_name, namespace=None):
         if not namespace:
             namespace = self.namespace
 
-        if config.multicluster:
+        if self.muliclusterhub_running():
             resource_kind = constants.SUBSCRIPTION_WITH_ACM
         else:
             resource_kind = constants.SUBSCRIPTION
@@ -1054,7 +1066,7 @@ def deploy_ocs_via_operator(self, image=None):
             worker_nodes = get_worker_nodes()
             node_obj = ocp.OCP(kind="node")
             platform = config.ENV_DATA.get("platform").lower()
-            if platform != constants.BAREMETAL_PLATFORM:
+            if platform not in [constants.BAREMETAL_PLATFORM, constants.HCI_BAREMETAL]:
                 for node in worker_nodes:
                     for interface in interfaces:
                         ip_link_cmd = f"ip link set promisc on {interface}"
@@ -1297,7 +1309,10 @@ def deploy_ocs_via_operator(self, image=None):
             cluster_data["spec"]["storageDeviceSets"][0]["replica"] = 1
 
         # set size of request for storage
-        if self.platform.lower() == constants.BAREMETAL_PLATFORM:
+        if self.platform.lower() in [
+            constants.BAREMETAL_PLATFORM,
+            constants.HCI_BAREMETAL,
+        ]:
             pv_size_list = helpers.get_pv_size(
                 storageclass=self.DEFAULT_STORAGECLASS_LSO
             )
@@ -2321,6 +2336,28 @@ def deploy_multicluster_hub(self):
             logger.error(f"Failed to install MultiClusterHub. Exception is: {ex}")
             return False
 
+    def muliclusterhub_running(self):
+        """
+        Check if MultiCluster Hub is running
+
+        Returns:
+            bool: True if MultiCluster Hub is running, False otherwise
+        """
+        ocp_obj = OCP(
+            kind=constants.ACM_MULTICLUSTER_HUB, namespace=constants.ACM_HUB_NAMESPACE
+        )
+        try:
+            mch_running = ocp_obj.wait_for_resource(
+                condition=constants.STATUS_RUNNING,
+                resource_name=constants.ACM_MULTICLUSTER_RESOURCE,
+                column="STATUS",
+                timeout=6,
+                sleep=3,
+            )
+        except CommandFailed:
+            mch_running = False
+        return mch_running
+
 
 def create_external_pgsql_secret():
     """
@@ -2359,8 +2396,8 @@ def validate_acm_hub_install():
         condition=constants.STATUS_RUNNING,
         resource_name=constants.ACM_MULTICLUSTER_RESOURCE,
         column="STATUS",
-        timeout=720,
-        sleep=5,
+        timeout=1200,
+        sleep=30,
     )
     logger.info("MultiClusterHub Deployment Succeeded")
 

ocs_ci/deployment/helpers/hypershift_base.py

@@ -238,6 +238,7 @@ def create_kubevirt_ocp_cluster(
         root_volume_size: str = 40,
         ocp_version=None,
         cp_availability_policy=None,
+        disable_default_sources=None,
     ):
         """
         Create HyperShift hosted cluster. Default parameters have minimal requirements for the cluster.
@@ -251,6 +252,7 @@ def create_kubevirt_ocp_cluster(
             root_volume_size (str): Root volume size of the cluster, default 40 (Gi is not required)
             cp_availability_policy (str): Control plane availability policy, default HighlyAvailable, if no value
             provided and argument is not used in the command the single replica mode cluster will be created
+            disable_default_sources (bool): Disable default sources on hosted cluster, such as 'redhat-operators'
         Returns:
             str: Name of the hosted cluster
         """
@@ -308,6 +310,9 @@ def create_kubevirt_ocp_cluster(
                 f" --control-plane-availability-policy {cp_availability_policy} "
             )
 
+        if disable_default_sources:
+            create_hcp_cluster_cmd += " --olm-disable-default-sources"
+
         logger.info("Creating HyperShift hosted cluster")
         exec_cmd(create_hcp_cluster_cmd)
 

ocs_ci/deployment/helpers/lso_helpers.py

@@ -212,7 +212,7 @@ def setup_local_storage(storageclass):
         # extra_disks is used in vSphere attach_disk() method
         storage_class_device_count = config.ENV_DATA.get("extra_disks", 1)
     expected_pvs = len(worker_names) * storage_class_device_count
-    if platform == constants.BAREMETAL_PLATFORM:
+    if platform in [constants.BAREMETAL_PLATFORM, constants.HCI_BAREMETAL]:
         verify_pvs_created(expected_pvs, storageclass, False)
     else:
         verify_pvs_created(expected_pvs, storageclass)

ocs_ci/deployment/hosted_cluster.py

@@ -371,13 +371,19 @@ def deploy_ocp(
         cp_availability_policy = (
             config.ENV_DATA["clusters"].get(self.name).get("cp_availability_policy")
         )
+        disable_default_sources = (
+            config.ENV_DATA["clusters"]
+            .get(self.name)
+            .get("disable_default_sources", False)
+        )
         return self.create_kubevirt_ocp_cluster(
             name=self.name,
             nodepool_replicas=nodepool_replicas,
             cpu_cores=cpu_cores_per_hosted_cluster,
             memory=memory_per_hosted_cluster,
             ocp_version=ocp_version,
             cp_availability_policy=cp_availability_policy,
+            disable_default_sources=disable_default_sources,
         )
 
     def deploy_dependencies(
@@ -523,7 +529,12 @@ def __init__(self, name: str):
         self.timeout_check_resources_exist_sec = 6
         self.timeout_wait_csvs_minutes = 20
         self.timeout_wait_pod_minutes = 30
-        self.storage_client_name = None
+
+        # default cluster name picked from the storage client yaml
+        storage_client_data = templating.load_yaml(
+            constants.PROVIDER_MODE_STORAGE_CLIENT
+        )
+        self.storage_client_name = storage_client_data["metadata"]["name"]
 
     @kubeconfig_exists_decorator
     def exec_oc_cmd(self, cmd, timeout=300, ignore_error=False, **kwargs):
@@ -622,19 +633,9 @@ def do_deploy(self):
         """
         Deploy ODF client on hosted OCP cluster
         """
-        logger.info(f"Deploying ODF client on hosted OCP cluster '{self.name}'")
-        hosted_odf_version = get_semantic_version(
-            config.ENV_DATA.get("clusters").get(self.name).get("hosted_odf_version"),
-            only_major_minor=True,
+        logger.info(
+            f"Deploying ODF client on hosted OCP cluster '{self.name}'. Creating ODF client namespace"
         )
-
-        no_network_policy_version = version.VERSION_4_16
-
-        if hosted_odf_version < no_network_policy_version:
-            logger.info("Applying network policy")
-            self.apply_network_policy()
-
-        logger.info("Creating ODF client namespace")
         self.create_ns()
 
         if self.odf_csv_installed():
@@ -1086,12 +1087,13 @@ def create_subscription(self):
         subscription_data = templating.load_yaml(constants.PROVIDER_MODE_SUBSCRIPTION)
 
         # since we are allowed to install N+1 on hosted clusters we can not rely on PackageManifest default channel
-        odf_version = get_semantic_version(
-            config.ENV_DATA.get("clusters").get(self.name).get("hosted_odf_version"),
-            only_major_minor=True,
+        hosted_odf_version = (
+            config.ENV_DATA.get("clusters").get(self.name).get("hosted_odf_version")
         )
+        if "latest" in hosted_odf_version:
+            hosted_odf_version = hosted_odf_version.split("-")[-1]
 
-        subscription_data["spec"]["channel"] = f"stable-{str(odf_version)}"
+        subscription_data["spec"]["channel"] = f"stable-{str(hosted_odf_version)}"
 
         subscription_file = tempfile.NamedTemporaryFile(
             mode="w+", prefix="subscription", delete=False
@@ -1140,15 +1142,14 @@ def storage_claim_exists_cephfs(self):
         Returns:
             bool: True if storage class claim for CephFS exists, False otherwise
         """
-        if (
-            get_semantic_version(
-                config.ENV_DATA.get("clusters")
-                .get(self.name)
-                .get("hosted_odf_version"),
-                only_major_minor=True,
-            )
-            < version.VERSION_4_16
-        ):
+
+        hosted_odf_version = (
+            config.ENV_DATA.get("clusters").get(self.name).get("hosted_odf_version")
+        )
+        if "latest" in hosted_odf_version:
+            hosted_odf_version = hosted_odf_version.split("-")[-1]
+
+        if get_semantic_version(hosted_odf_version, True) < version.VERSION_4_16:
             ocp = OCP(
                 kind=constants.STORAGECLASSCLAIM,
                 namespace=self.namespace_client,
@@ -1228,15 +1229,14 @@ def storage_claim_exists_rbd(self):
         Returns:
              bool: True if storage class claim for RBD exists, False otherwise
         """
-        if (
-            get_semantic_version(
-                config.ENV_DATA.get("clusters")
-                .get(self.name)
-                .get("hosted_odf_version"),
-                True,
-            )
-            < version.VERSION_4_16
-        ):
+
+        hosted_odf_version = (
+            config.ENV_DATA.get("clusters").get(self.name).get("hosted_odf_version")
+        )
+        if "latest" in hosted_odf_version:
+            hosted_odf_version = hosted_odf_version.split("-")[-1]
+
+        if get_semantic_version(hosted_odf_version, True) < version.VERSION_4_16:
             ocp = OCP(
                 kind=constants.STORAGECLASSCLAIM,
                 namespace=self.namespace_client,

ocs_ci/deployment/metallb.py

@@ -500,9 +500,9 @@ def deploy_lb(self):
         logger.info(
             f"Deploying MetalLB and dependant resources to namespace: '{self.namespace_lb}'"
         )
-        # icsp mirrors necessary to download packages for the downstream version of metallb
-        self.apply_icsp()
 
+        if self.apply_icsp():
+            logger.info("ICSP brew-registry applied successfully")
         if self.create_metallb_namespace():
             logger.info(f"Namespace {self.namespace_lb} created successfully")
         if self.create_catalog_source():
@@ -669,10 +669,26 @@ def wait_csv_installed(self):
                 break
         return True
 
+    def icsp_brew_registry_exists(self):
+        """
+        Check if the ICSP Brew registry exists
+
+        Returns:
+            bool: True if the ICSP Brew registry exists, False otherwise
+        """
+        return OCP(
+            kind="ImageContentSourcePolicy", resource_name="brew-registry"
+        ).check_resource_existence(
+            timeout=self.timeout_check_resources_existence, should_exist=True
+        )
+
     def apply_icsp(self):
         """
         Apply the ICSP to the cluster
         """
+        if self.icsp_brew_registry_exists():
+            logger.info("ICSP Brew registry already exists")
+            return
         icsp_data = templating.load_yaml(constants.SUBMARINER_DOWNSTREAM_BREW_ICSP)
         icsp_data_yaml = tempfile.NamedTemporaryFile(
             mode="w+", prefix="acm_icsp", delete=False
@@ -681,3 +697,4 @@ def apply_icsp(self):
         exec_cmd(f"oc create -f {icsp_data_yaml.name}", timeout=300)
         wait_for_machineconfigpool_status(node_type="all")
         logger.info("ICSP applied successfully")
+        return self.icsp_brew_registry_exists()

ocs_ci/helpers/managed_services.py

@@ -1,6 +1,7 @@
 """
 Managed Services related functionalities
 """
+
 import logging
 import re
 
@@ -236,7 +237,7 @@ def verify_osd_distribution_on_provider():
         # 4Ti is the size of OSD
         assert (
             osd_count == int(size) / 4
-        ), f"Zone {zone} does not have {size/4} osd, but {osd_count}"
+        ), f"Zone {zone} does not have {size / 4} osd, but {osd_count}"
 
 
 def verify_storageclient(
@@ -326,7 +327,7 @@ def get_all_storageclassclaims(namespace=None):
         sc_claim_obj = OCP(kind=constants.STORAGECLAIM, namespace=namespace)
     else:
         sc_claim_obj = OCP(kind=constants.STORAGECLASSCLAIM, namespace=namespace)
-    sc_claims_data = sc_claim_obj.get()["items"]
+    sc_claims_data = sc_claim_obj.get(retry=6, wait=30)["items"]
     log.info(f"storage claims: {sc_claims_data}")
     return [OCS(**claim_data) for claim_data in sc_claims_data]
 

ocs_ci/ocs/node.py

@@ -797,6 +797,7 @@ def get_compute_node_names(no_replace=False):
         constants.BAREMETAL_PLATFORM,
         constants.BAREMETALPSI_PLATFORM,
         constants.IBM_POWER_PLATFORM,
+        constants.HCI_BAREMETAL,
     ]:
         if no_replace:
             return [