Ocp deployment using acm UI (#5562)

Implement OCP DR clusters deployment using ACM UI

Signed-off-by: Shylesh Kumar Mohan <[email protected]>

conf/ocsci/multicluster_acm_ocp_deployment.yaml

@@ -0,0 +1,4 @@
+ENV_DATA:
+  acm_ocp_deployment: True
+UI_SELENIUM:
+  headless: False

conf/ocsci/multicluster_mode_rdr.yaml

@@ -0,0 +1,3 @@
+MULTICLUSTER:
+  # Other modes could be managed services
+  multicluster_mode: "regional-dr"

ocs_ci/deployment/deployment.py

@@ -154,12 +154,12 @@ class OCPDeployment(BaseOCPDeployment):
 
         pass
 
-    def deploy_cluster(self, log_cli_level="DEBUG"):
+    def do_deploy_ocp(self, log_cli_level):
         """
-        We are handling both OCP and OCS deployment here based on flags
-
+        Deploy OCP
         Args:
-            log_cli_level (str): log level for installer (default: DEBUG)
+            log_cli_level (str): log level for the installer
+
         """
         if not config.ENV_DATA["skip_ocp_deployment"]:
             if is_cluster_running(self.cluster_path):
@@ -175,46 +175,22 @@ def deploy_cluster(self, log_cli_level="DEBUG"):
                         collect_ocs_logs("deployment", ocs=False)
                     raise
 
-        # Deployment of network split scripts via machineconfig API happens
-        # before OCS deployment.
-        if config.DEPLOYMENT.get("network_split_setup"):
-            master_zones = config.ENV_DATA.get("master_availability_zones")
-            worker_zones = config.ENV_DATA.get("worker_availability_zones")
-            # special external zone, which is directly defined by ip addr list,
-            # such zone could represent external services, which we could block
-            # access to via ax-bx-cx network split
-            if config.DEPLOYMENT.get("network_split_zonex_addrs") is not None:
-                x_addr_list = config.DEPLOYMENT["network_split_zonex_addrs"].split(",")
-            else:
-                x_addr_list = None
-            if config.DEPLOYMENT.get("arbiter_deployment"):
-                arbiter_zone = self.get_arbiter_location()
-                logger.debug("detected arbiter zone: %s", arbiter_zone)
-            else:
-                arbiter_zone = None
-            # TODO: use temporary directory for all temporary files of
-            # ocs-deployment, not just here in this particular case
-            tmp_path = Path(tempfile.mkdtemp(prefix="ocs-ci-deployment-"))
-            logger.debug("created temporary directory %s", tmp_path)
-            setup_netsplit(
-                tmp_path, master_zones, worker_zones, x_addr_list, arbiter_zone
-            )
-        ocp_version = version.get_semantic_ocp_version_from_config()
-        if (
-            config.ENV_DATA.get("deploy_acm_hub_cluster")
-            and ocp_version >= version.VERSION_4_9
-        ):
-            try:
-                self.deploy_acm_hub()
-            except Exception as e:
-                logger.error(e)
+    def do_deploy_submariner(self):
+        """
+        Deploy Submariner operator
 
+        """
         # Multicluster operations
         if config.multicluster:
             # Configure submariner only on non-ACM clusters
             submariner = Submariner()
             submariner.deploy()
 
+    def do_deploy_ocs(self):
+        """
+        Deploy OCS/ODF and run verification as well
+
+        """
         if not config.ENV_DATA["skip_ocs_deployment"]:
             for i in range(config.nclusters):
                 if config.multicluster and config.get_acm_index() == i:
@@ -234,7 +210,6 @@ def deploy_cluster(self, log_cli_level="DEBUG"):
                         collect_ocs_logs("deployment", ocp=False)
                     raise
             config.reset_ctx()
-
             # Run ocs_install_verification here only in case of multicluster.
             # For single cluster, test_deployment will take care.
             if config.multicluster:
@@ -251,12 +226,63 @@ def deploy_cluster(self, log_cli_level="DEBUG"):
         else:
             logger.warning("OCS deployment will be skipped")
 
+    def do_deploy_rdr(self):
+        """
+        Call Regional DR deploy
+
+        """
         # Multicluster: Handle all ODF multicluster DR ops
         if config.multicluster:
             dr_conf = self.get_rdr_conf()
             deploy_dr = MultiClusterDROperatorsDeploy(dr_conf)
             deploy_dr.deploy()
 
+    def deploy_cluster(self, log_cli_level="DEBUG"):
+        """
+        We are handling both OCP and OCS deployment here based on flags
+
+        Args:
+            log_cli_level (str): log level for installer (default: DEBUG)
+        """
+        self.do_deploy_ocp(log_cli_level)
+        # Deployment of network split scripts via machineconfig API happens
+        # before OCS deployment.
+        if config.DEPLOYMENT.get("network_split_setup"):
+            master_zones = config.ENV_DATA.get("master_availability_zones")
+            worker_zones = config.ENV_DATA.get("worker_availability_zones")
+            # special external zone, which is directly defined by ip addr list,
+            # such zone could represent external services, which we could block
+            # access to via ax-bx-cx network split
+            if config.DEPLOYMENT.get("network_split_zonex_addrs") is not None:
+                x_addr_list = config.DEPLOYMENT["network_split_zonex_addrs"].split(",")
+            else:
+                x_addr_list = None
+            if config.DEPLOYMENT.get("arbiter_deployment"):
+                arbiter_zone = self.get_arbiter_location()
+                logger.debug("detected arbiter zone: %s", arbiter_zone)
+            else:
+                arbiter_zone = None
+            # TODO: use temporary directory for all temporary files of
+            # ocs-deployment, not just here in this particular case
+            tmp_path = Path(tempfile.mkdtemp(prefix="ocs-ci-deployment-"))
+            logger.debug("created temporary directory %s", tmp_path)
+            setup_netsplit(
+                tmp_path, master_zones, worker_zones, x_addr_list, arbiter_zone
+            )
+        ocp_version = version.get_semantic_ocp_version_from_config()
+        if (
+            config.ENV_DATA.get("deploy_acm_hub_cluster")
+            and ocp_version >= version.VERSION_4_9
+        ):
+            try:
+                self.deploy_acm_hub()
+            except Exception as e:
+                logger.error(e)
+
+        self.do_deploy_submariner()
+        self.do_deploy_ocs()
+        self.do_deploy_rdr()
+
     def get_rdr_conf(self):
         """
         Aggregate important Regional DR parameters in the dictionary

ocs_ci/deployment/factory.py

@@ -13,7 +13,10 @@ class DeploymentFactory(object):
     """
 
     def __init__(self):
-        self.deployment_platform = config.ENV_DATA["platform"].lower()
+        if config.ENV_DATA.get("acm_ocp_deployment"):
+            self.deployment_platform = constants.ACM_OCP_DEPLOYMENT
+        else:
+            self.deployment_platform = config.ENV_DATA["platform"].lower()
         self.cls_map = {}
         # A map of all existing deployments and respective classes
         # should be put here, but only in the condition if that platform is used.
@@ -76,6 +79,10 @@ def __init__(self):
             from .rhv import RHVIPI
 
             self.cls_map["rhv_ipi"] = RHVIPI
+        elif self.deployment_platform == constants.ACM_OCP_DEPLOYMENT:
+            from .multicluster_deployment import OCPDeployWithACM
+
+            self.cls_map["acm_ocp_deployment"] = OCPDeployWithACM
 
     def get_deployment(self):
         """
@@ -84,6 +91,9 @@ def get_deployment(self):
         deployment_platform may look like 'aws', 'vmware', 'baremetal'
         deployment_type may be like 'ipi' or 'upi'
         """
+        if config.ENV_DATA.get("acm_ocp_deployment"):
+            logger.info("Deployment will be done through ACM platform")
+            return self.cls_map[constants.ACM_OCP_DEPLOYMENT]()
         deployment_type = config.ENV_DATA["deployment_type"]
         flexy_deployment = config.ENV_DATA["flexy_deployment"]
         deployment_cls_key = (

ocs_ci/deployment/multicluster_deployment.py

@@ -0,0 +1,108 @@
+import logging
+
+from ocs_ci.deployment.deployment import Deployment
+from ocs_ci.framework import config
+from ocs_ci.ocs.exceptions import ACMClusterDeployException
+from ocs_ci.ocs.ui import acm_ui
+from ocs_ci.ocs.utils import get_non_acm_cluster_config
+from ocs_ci.ocs.acm import acm
+from ocs_ci.ocs import constants
+
+
+logger = logging.getLogger(__name__)
+
+
+class OCPDeployWithACM(Deployment):
+    """
+    When we instantiate this class, the assumption is we already have
+    an OCP cluster with ACM installed and current context is ACM
+
+    """
+
+    def __init__(self):
+        """
+        When we init Deployment class it will have all the
+        ACM cluster's context
+        """
+        super().__init__()
+        self.multicluster_mode = config.MULTICLUSTER.get("multicluster_mode", None)
+        # Used for housekeeping during multiple OCP cluster deployments
+        self.deployment_cluster_list = list()
+        # Whether to start deployment in asynchronous mode or synchronous mode
+        # In async deploy mode, we will have a single wait method waiting for
+        # all the cluster deployments to finish
+        self.deploy_sync_mode = config.MULTICLUSTER.get("deploy_sync_mode", "async")
+        self.ui_driver = None
+
+    def do_deploy_ocp(self, log_cli_level="INFO"):
+        """
+        This function overrides the parent's function in order accomodate
+        ACM based OCP cluster deployments
+
+        """
+        if self.multicluster_mode == constants.RDR_MODE:
+            self.do_rdr_acm_ocp_deploy()
+
+    def do_rdr_acm_ocp_deploy(self):
+        """
+        Specific to regional DR OCP cluster deployments
+
+        """
+        factory = acm_ui.ACMOCPDeploymentFactory()
+        self.ui_driver = acm.login_to_acm()
+
+        if self.deploy_sync_mode == "async":
+            rdr_clusters = get_non_acm_cluster_config()
+            for c in rdr_clusters:
+                logger.info(f"{c.ENV_DATA['cluster_name']}")
+                logger.info(f"{c.ENV_DATA['platform']}")
+                logger.info(f"{c.ENV_DATA['deployment_type']}")
+            for cluster_conf in rdr_clusters:
+                deployer = factory.get_platform_instance(self.ui_driver, cluster_conf)
+                deployer.create_cluster_prereq()
+                deployer.create_cluster()
+                self.deployment_cluster_list.append(deployer)
+            # At this point deployment of all non-acm ocp clusters have been
+            # triggered, we need to wait for all of them to succeed
+            self.wait_for_all_clusters_async()
+            # Download kubeconfig to the respective directories
+            for cluster in self.deployment_cluster_list:
+                cluster.download_cluster_conf_files()
+
+    def wait_for_all_clusters_async(self):
+        # We will say done only when none of the clusters are in
+        # 'Creating' state. Either they have to be in 'Ready' state
+        # OR 'Failed' state
+        done_waiting = False
+        while not done_waiting:
+            done_waiting = True
+            for cluster in self.deployment_cluster_list:
+                if cluster.deployment_status not in ["ready", "failed"]:
+                    cluster.get_deployment_status()
+                    done_waiting = False
+        # We will fail even if one of the clusters failed to deploy
+        failed_list = list()
+        success_list = list()
+        for cluster in self.deployment_cluster_list:
+            if cluster.deployment_status == "failed":
+                failed_list.append(cluster)
+            else:
+                success_list.append(cluster)
+
+        if success_list:
+            logger.info("Deployment for following clusters Passed")
+            logger.info(f"{[c.cluster_name for c in success_list]}")
+        if failed_list:
+            logger.error("Deployment failed for following clusters")
+            logger.error(f"{[c.cluster_name for c in failed_list]}")
+            raise ACMClusterDeployException("one or more Cluster Deployment failed ")
+
+    def deploy_cluster(self, log_cli_level="INFO"):
+        """
+        We deploy new OCP clusters using ACM
+        Note: Importing cluster through ACM has been implemented as part
+        of Jenkins pipeline
+
+        """
+
+        super().deploy_cluster(log_cli_level=log_cli_level)

ocs_ci/ocs/constants.py

@@ -11,6 +11,7 @@
 
 import os
 
+
 # Logging
 LOG_FORMAT = "%(asctime)s - %(threadName)s - %(name)s - %(levelname)s - %(message)s"
 
@@ -654,6 +655,7 @@
 DR_RAMEN_HUB_OPERATOR_CONFIG = "ramen-hub-operator-config"
 DR_RAMEN_CLUSTER_OPERATOR_CONFIG = "ramen-dr-cluster-operator-config"
 ODF_MULTICLUSTER_ORCHESTRATOR_CONTROLLER_MANAGER = "odfmo-controller-manager"
+RDR_MODE = "regional-dr"
 
 # DR constants
 SUBMARINER_DOWNLOAD_URL = "https://get.submariner.io"
@@ -772,6 +774,7 @@
 OPENSHIFT_DEDICATED_PLATFORM = "openshiftdedicated"
 RHV_PLATFORM = "rhv"
 ROSA_PLATFORM = "rosa"
+ACM_OCP_DEPLOYMENT = "acm_ocp_deployment"
 ON_PREM_PLATFORMS = [
     VSPHERE_PLATFORM,
     BAREMETAL_PLATFORM,
@@ -1665,3 +1668,22 @@
 VAULT_TOKEN = "vaulttokens"
 VAULT_TENANT_SA = "vaulttenantsa"
 RBD_CSI_VAULT_TOKEN_REVIEWER_NAME = "rbd-csi-vault-token-review"
+# ACM UI related constants
+PLATFORM_XPATH_MAP = {
+    "vsphere": "cc_provider_vmware_vsphere",
+    "AWS": None,
+    "baremetal": None,
+    "azure": None,
+}
+ACM_PLATOFRM_VSPHERE_CRED_PREFIX = "vsphereacmocp-"
+# example release image url : quay.io/openshift-release-dev/ocp-release:4.9.23-x86_64
+ACM_OCP_RELEASE_IMG_URL_PREFIX = "registry.ci.openshift.org/ocp/release"
+ACM_VSPHERE_NETWORK = "VM Network"
+ACM_CLUSTER_DEPLOY_TIMEOUT = 2700  # 45 minutes
+ACM_CLUSTER_DEPLOYMENT_LABEL_KEY = "hive.openshift.io/cluster-deployment-name"
+ACM_CLUSTER_DEPLOYMENT_SECRET_TYPE_LABEL_KEY = "hive.openshift.io/secret-type"
+# Concatenated CA file for vcenter
+VSPHERE_CA_FILE_PATH = os.path.join(DATA_DIR, "vsphere_ca.crt")
+SSH_PRIV_KEY = os.path.expanduser(os.path.join(".ssh", "openshift-dev.pem"))
+SSH_PUB_KEY = os.path.expanduser(os.path.join(".ssh", "openshift-dev.pub"))
+SPACE = " "

ocs_ci/ocs/exceptions.py

@@ -472,6 +472,10 @@ class BenchmarkTestFailed(Exception):
     pass
 
 
+class ACMClusterDeployException(Exception):
+    pass
+
+
 class WrongVersionExpression(ValueError):
     pass