code clean up: rosa classic scripts (#345)

* added rosahcp script

* trimmed down extra char

* updated workload
HCP light-weight c-d-v2 iteration will be 7 iteration/node

* removed extra env vars

* increasing node ready timeout

* mgmt cluster env var

* scrapping metric for longer duration

* updated query logic to index MC stat

dags/openshift_nightlies/config/benchmarks/hcp-small-control-plane.json

@@ -0,0 +1,37 @@
+{
+    "benchmarks": [
+        {
+            "name": "node-density",
+            "workload": "kube-burner-ocp-wrapper",
+            "trigger_rule": "all_done",
+            "command": "./run.sh",
+            "env": {
+                "WORKLOAD": "node-density",
+                "LOG_LEVEL": "debug"
+            }
+        },
+        {
+            "name": "node-desnity-cni",
+            "workload": "kube-burner-ocp-wrapper",
+            "trigger_rule": "all_done",
+            "command": "./run.sh",
+            "env": {
+                "WORKLOAD": "node-density-cni",
+                "LOG_LEVEL": "debug"
+            }
+        },
+        {
+            "name": "cluster-density-v2",
+            "workload": "kube-burner-ocp-wrapper",
+            "trigger_rule": "all_done",
+            "command": "./run.sh",
+            "env": {
+                "WORKLOAD": "cluster-density-v2",
+                "ITERATIONS": "500",
+                "LOG_LEVEL": "debug",
+                "CHURN": "true",
+                "EXTRA_FLAGS": "--churn-duration=1h --churn-percent=10 --churn-delay=30s"
+            }
+        }                
+    ]
+}

dags/openshift_nightlies/config/benchmarks/hosted-control-plane-p75.json

@@ -1,13 +1,13 @@
 {
     "benchmarks": [
         {
-            "name": "cluster-density-ms-p75",
+            "name": "cluster-density-v2-p75",
             "workload": "kube-burner-ocp-wrapper",
             "trigger_rule": "all_done",
             "command": "./run.sh",
             "env": {
-                "WORKLOAD": "cluster-density-ms",
-                "ITERATIONS": "75",
+                "WORKLOAD": "cluster-density-v2",
+                "ITERATIONS": "63",
                 "LOG_LEVEL": "debug",
                 "EXTRA_FLAGS": "--churn-duration=1h --churn-percent=10 --churn-delay=30s"
             }

dags/openshift_nightlies/config/benchmarks/hosted-control-plane-p90.json

@@ -1,22 +1,15 @@
 {
     "benchmarks": [
         {
-            "name": "cluster-density-ms-p90",
-            "workload": "kube-burner",
+            "name": "cluster-density-v2-p90",
+            "workload": "kube-burner-ocp-wrapper",
+            "trigger_rule": "all_done",
             "command": "./run.sh",
             "env": {
-                "WORKLOAD": "cluster-density-ms",
-                "JOB_ITERATIONS": "100",
-                "JOB_TIMEOUT": "18000",
-                "STEP_SIZE": "2m",
-                "HYPERSHIFT": "true",
-                "METRICS_PROFILE": "metrics-profiles/hypershift-metrics.yaml",
-                "QPS": "20",
-                "BURST": "20",
-                "LOG_LEVEL": "info",
-                "PLATFORM_ALERTS": "false",
-                "CLEANUP_WHEN_FINISH": "true",
-                "CLEANUP": "true"
+                "WORKLOAD": "cluster-density-v2",
+                "ITERATIONS": "84",
+                "LOG_LEVEL": "debug",
+                "EXTRA_FLAGS": "--churn-duration=1h --churn-percent=10 --churn-delay=30s"
             }
         }
     ]

dags/openshift_nightlies/config/install/rosa/rosa-hcp-ovn-data-plane.json → dags/openshift_nightlies/config/install/rosa-hcp/data-plane.json

@@ -1,5 +1,4 @@
 {
-    "rosa_hcp": "true",
     "aws_profile": "",
     "aws_access_key_id": "",
     "aws_secret_access_key": "",

dags/openshift_nightlies/config/install/rosa/rosa-hcp-ovn.json → dags/openshift_nightlies/config/install/rosa-hcp/p75-control-plane.json

@@ -1,5 +1,4 @@
 {
-    "rosa_hcp": "true",
     "aws_profile": "",
     "aws_access_key_id": "",
     "aws_secret_access_key": "",
@@ -16,7 +15,7 @@
     "openshift_worker_instance_type": "m5.2xlarge",
     "machineset_metadata_label_prefix": "machine.openshift.io",
     "staging_mgmt_provisioner_shards": "b4bb294b-a76c-11ed-91b2-0a580a831ba1",
-    "number_of_hostedcluster": 2,
+    "number_of_hostedcluster": 10,
     "hcp_install_interval": 60,
     "extra_machinepool": [{
         "name": "infra",

dags/openshift_nightlies/config/install/rosa-hcp/small-control-plane.json

@@ -0,0 +1,27 @@
+{
+    "aws_profile": "",
+    "aws_access_key_id": "",
+    "aws_secret_access_key": "",
+    "aws_authentication_method": "sts",
+    "aws_region": "us-east-2",
+    "rosa_environment": "staging",
+    "rosa_cli_version": "container",
+    "ocm_cli_version": "container",
+    "ocm_environment": "stage",
+    "managed_channel_group": "nightly",
+    "managed_ocp_version": "latest",
+    "openshift_worker_count": 24,
+    "openshift_network_type": "OVNKubernetes",
+    "openshift_worker_instance_type": "m5.2xlarge",
+    "machineset_metadata_label_prefix": "machine.openshift.io",
+    "staging_mgmt_provisioner_shards": "b4bb294b-a76c-11ed-91b2-0a580a831ba1",
+    "number_of_hostedcluster": 1,
+    "hcp_install_interval": 10,
+    "extra_machinepool": [{
+        "name": "infra",
+        "replica": "1",
+        "instance_type": "r5.xlarge",
+        "labels": "node-role.kubernetes.io/infra=",
+        "taints": "node-role.kubernetes.io/infra=:NoSchedule"
+    }]
+}

dags/openshift_nightlies/dag.py

@@ -14,6 +14,7 @@
 from openshift_nightlies.tasks.install.openstack import jetpack
 from openshift_nightlies.tasks.install.baremetal import jetski, webfuse
 from openshift_nightlies.tasks.install.rosa import rosa
+from openshift_nightlies.tasks.install.rosahcp import rosahcp
 from openshift_nightlies.tasks.install.rogcp import rogcp
 from openshift_nightlies.tasks.install.hypershift import hypershift
 from openshift_nightlies.tasks.install.prebuilt import initialize_cluster
@@ -155,46 +156,50 @@ def _get_openshift_installer(self):
 class RosaNightlyDAG(AbstractOpenshiftNightlyDAG):
     def build(self):
         installer = self._get_openshift_installer()
-        if installer.get_type() == "rosa_hcp":
-            install_cluster = installer.get_install_hcp_task()
-            hosted_installer = self._get_hypershift_openshift_installer()
-            wait_task = hosted_installer.wait_task()
-            wait_before_cleanup = hosted_installer.wait_task(id="wait_before_cleanup")
-            for c_id, install_hc, postinstall_hc, cleanup_hc in install_cluster:
-                benchmark = self._add_benchmarks(task_group=c_id)
-                install_hc >> postinstall_hc >> wait_task >> benchmark >> wait_before_cleanup >> cleanup_hc
+        install_cluster = installer.get_install_task()
+        final_status = final_dag_status.get_task(self.dag)
+        with TaskGroup("benchmarks", prefix_group_id=False, dag=self.dag) as benchmarks:
+            must_gather = self._get_scale_ci_diagnosis().get_must_gather("must-gather")
+            benchmark_tasks = self._get_e2e_benchmarks().get_benchmarks()
+            chain(*benchmark_tasks)
+            # Configure must_gather as downstream of all benchmark tasks
+            for benchmark in benchmark_tasks:
+                benchmark >> must_gather
+        rosa_post_installation = self._get_rosa_postinstall_setup()._get_rosa_postinstallation()
+        if self.config.cleanup_on_success:
+            cleanup_cluster = installer.get_cleanup_task()
+            install_cluster >> rosa_post_installation >> benchmarks >> cleanup_cluster >> final_status
         else:
-            install_cluster = installer.get_install_task()
-            final_status = final_dag_status.get_task(self.dag)
-            with TaskGroup("benchmarks", prefix_group_id=False, dag=self.dag) as benchmarks:
-                must_gather = self._get_scale_ci_diagnosis().get_must_gather("must-gather")
-                benchmark_tasks = self._get_e2e_benchmarks().get_benchmarks()
-                chain(*benchmark_tasks)
-                # Configure must_gather as downstream of all benchmark tasks
-                for benchmark in benchmark_tasks:
-                    benchmark >> must_gather
-            rosa_post_installation = self._get_rosa_postinstall_setup()._get_rosa_postinstallation()
-            if self.config.cleanup_on_success:
-                cleanup_cluster = installer.get_cleanup_task()
-                install_cluster >> rosa_post_installation >> benchmarks >> cleanup_cluster >> final_status
-            else:
-                install_cluster >> rosa_post_installation >> benchmarks >> final_status
+            install_cluster >> rosa_post_installation >> benchmarks >> final_status
 
     def _get_openshift_installer(self):
         return rosa.RosaInstaller(self.dag, self.config, self.release)
 
     def _get_e2e_benchmarks(self, task_group="benchmarks"):
         return e2e.E2EBenchmarks(self.dag, self.config, self.release, task_group)
 
+class RosaHCPNightlyDAG(AbstractOpenshiftNightlyDAG):
+    def build(self):
+        installer = self._get_openshift_installer()
+        install_cluster = installer.get_install_hcp_task()
+        wait_task = installer.wait_task()
+        wait_before_cleanup = installer.wait_task(id="wait_before_cleanup")
+        for c_id, install_hc, postinstall_hc, cleanup_hc in install_cluster:
+            benchmark = self._add_benchmarks(task_group=c_id)
+            install_hc >> postinstall_hc >> wait_task >> benchmark >> wait_before_cleanup >> cleanup_hc
+
+    def _get_openshift_installer(self):
+        return rosahcp.RosaHCPInstaller(self.dag, self.config, self.release)
+
+    def _get_e2e_benchmarks(self, task_group="benchmarks"):
+        return e2e.E2EBenchmarks(self.dag, self.config, self.release, task_group)
+
     def _add_benchmarks(self, task_group):
         with TaskGroup(task_group, prefix_group_id=False, dag=self.dag) as benchmarks:
             benchmark_tasks = self._get_e2e_benchmarks(task_group).get_benchmarks()
             chain(*benchmark_tasks)
         return benchmarks
 
-    def _get_hypershift_openshift_installer(self):
-        return hypershift.HypershiftInstaller(self.dag, self.config, self.release)
-
 
 class RoGCPNightlyDAG(AbstractOpenshiftNightlyDAG):
     def build(self):
@@ -242,7 +247,7 @@ def build(self):
                 install_mgmt_cluster >> rosa_post_installation >> install_hc >> wait_task >> benchmark
 
     def _get_openshift_installer(self):
-        return rosa.RosaInstaller(self.dag, self.config, self.release)
+        return rosahcp.RosaHCPInstaller(self.dag, self.config, self.release)
 
     def _get_hypershift_openshift_installer(self):
         return hypershift.HypershiftInstaller(self.dag, self.config, self.release)
@@ -303,6 +308,8 @@ def build_releases():
             nightly = OpenstackNightlyDAG(openshift_release, dag_config)
         elif openshift_release.platform == "rosa":
             nightly = RosaNightlyDAG(openshift_release, dag_config)
+        elif openshift_release.platform == "rosahcp":
+            nightly = RosaHCPNightlyDAG(openshift_release, dag_config)            
         elif openshift_release.platform == "rogcp":
             nightly = RoGCPNightlyDAG(openshift_release, dag_config)
         elif openshift_release.platform == "hypershift":

dags/openshift_nightlies/manifest.yaml

@@ -103,16 +103,6 @@ platforms:
         config:
           install: rosa/iam-ovn.json
           benchmarks: control-plane.json
-      - name: osd-ovn-control-plane
-        schedule: "30 12 * * 1,3,5"
-        config:
-          install: rosa/ovn-osd.json
-          benchmarks: control-plane.json
-      - name: osd-ovn-data-plane
-        schedule: "30 1 * * 1,3,5" # an hour gap for OSD to avoid OsdCcsAdmin key limit
-        config:
-          install: rosa/ovn-osd.json
-          benchmarks: data-plane-mgs.json
       - name: ocm-api-load
         schedule: "None"
         config:
@@ -123,16 +113,25 @@ platforms:
         config:
           install: rosa/upgrade.json
           benchmarks: upgrade.json
-      - name: rosa-hcp-control-plane
+
+  rosahcp:
+    versions: ["4.12", "4.13"]
+    variants:
+      - name: p75-control-plane
         schedule:  "0 12 * * 3"
         config:
-          install: rosa/rosa-hcp-ovn.json
+          install: rosa-hcp/p75-control-plane.json
           benchmarks: hosted-control-plane-p75.json
-      - name: rosa-hcp-data-plane
+      - name: data-plane-v2
         schedule:  "1 12 * * 3"
         config:
-          install: rosa/rosa-hcp-ovn-data-plane.json
+          install: rosa-hcp/data-plane.json
           benchmarks: data-plane-v2.json
+      - name: small-control-plane
+        schedule:  "2 12 * * 3"
+        config:
+          install: rosa-hcp/small-control-plane.json
+          benchmarks: hcp-small-control-plane.json
 
   rogcp:
     versions: ["4.12", "4.13"]
@@ -148,33 +147,6 @@ platforms:
           install: rogcp/ovn.json
           benchmarks: data-plane-mgs.json
 
-  hypershift:
-    versions: ["4.12", "4.13"]
-    variants:
-      - name: management-control-plane
-        schedule: "30 3 * * 1,3,5" # an hour gap for OSD to avoid OsdCcsAdmin key limit
-        config:
-          install: hypershift/none-type.json
-          benchmarks: management-control-plane.json
-      - name: ovn-control-plane-p75
-        schedule: "30 4 * * 1,3,5" # an hour gap for OSD to avoid OsdCcsAdmin key limit
-        config:
-          install: hypershift/ovn-p75.json
-          benchmarks: hosted-control-plane-p75.json
-      - name: ovn-control-plane-p90
-        schedule: "30 5 * * 1,3,5" # an hour gap for OSD to avoid OsdCcsAdmin key limit
-        config:
-          install: hypershift/ovn-p90.json
-          benchmarks: hosted-control-plane-p90.json
-      - name: chaos-ovn-control-plane-p75
-        config:
-          install: hypershift/ovn-p75.json
-          benchmarks: hosted-control-plane-chaos-p75.json
-      - name: chaos-ovn-control-plane-p90
-        config:
-          install: hypershift/ovn-p90.json
-          benchmarks: hosted-control-plane-chaos-p90.json
-
   prebuilt:
     versions: ["4.x"]
     variants:

dags/openshift_nightlies/models/release.py

@@ -41,7 +41,7 @@ def _generate_cluster_name(self):
         else:
             cluster_name = f"{git_user}-{git_branch}-{release_name}"
 
-        if self.platform == 'rosa' or self.platform == 'rogcp' or self.platform == 'hypershift':
+        if self.platform == 'rosa' or self.platform == 'rogcp' or self.platform == 'hypershift' or self.platform == 'rosahcp':
             #Only 15 chars are allowed
             cluster_version = str(self.version).replace(".","")
             return "perf-"+md5(cluster_name.encode("ascii")).hexdigest()[:3]