Adding telemetry to release_v2 (#259)

* Added telemetry to release_v2

* Minor Code Clean-up

* Remove exiting system so that telemetry still sends on errors

* Adding hyperpod_v2 to names for clarity

Author: aviruthen

src/hyperpod_cli/commands/cluster.py

@@ -43,6 +43,8 @@
     TEMP_KUBE_CONFIG_FILE,
     OutputFormat,
 )
+from hyperpod_cli.telemetry import _hyperpod_telemetry_emitter
+from hyperpod_cli.telemetry.constants import Feature
 from hyperpod_cli.telemetry.user_agent import (
     get_user_agent_extra_suffix,
 )
@@ -107,6 +109,7 @@
     multiple=True,
     help="Optional. The namespace that you want to check the capacity for. Only SageMaker managed namespaces are supported.",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.get_clusters_cli")
 def get_clusters(
     region: Optional[str],
     orchestrator: Optional[str],
@@ -463,6 +466,7 @@ def _aggregate_nodes_info(
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.connect_cluster_cli")
 def connect_cluster(
     cluster_name: str,
     region: Optional[str],

src/hyperpod_cli/commands/job.py

@@ -50,6 +50,8 @@
     Volume,
     USER_NAME_LABEL_KEY,
 )
+from hyperpod_cli.telemetry import _hyperpod_telemetry_emitter
+from hyperpod_cli.telemetry.constants import Feature
 from hyperpod_cli.clients.kubernetes_client import (
     KubernetesClient,
 )
@@ -124,6 +126,7 @@
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.get_job_cli")
 def get_job(
     job_name: str,
     namespace: Optional[str],
@@ -144,9 +147,8 @@ def get_job(
         result = get_training_job_service.get_training_job(job_name, namespace, verbose)
         click.echo(result)
     except Exception as e:
-        sys.exit(
-            f"Unexpected error happens when trying to get training job {job_name} : {e}"
-        )
+        logger.error(f"Unexpected error happens when trying to get training job {job_name} : {e}")
+        raise
 
 
 @click.command()
@@ -186,6 +188,7 @@ def get_job(
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.list_jobs_cli")
 def list_jobs(
     namespace: Optional[str],
     all_namespaces: Optional[bool],
@@ -205,7 +208,8 @@ def list_jobs(
         )
         click.echo(result)
     except Exception as e:
-        sys.exit(f"Unexpected error happens when trying to list training job : {e}")
+        logger.error(f"Unexpected error happens when trying to list training job : {e}")
+        raise
 
 
 @click.command()
@@ -228,6 +232,7 @@ def list_jobs(
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.list_pods_cli")
 def list_pods(
     job_name: str,
     namespace: Optional[str],
@@ -246,9 +251,8 @@ def list_pods(
         result = list_pods_service.list_pods_for_training_job(job_name, namespace, True)
         click.echo(result)
     except Exception as e:
-        sys.exit(
-            f"Unexpected error happens when trying to list pods for training job {job_name} : {e}"
-        )
+        logger.error(f"Unexpected error happens when trying to list pods for training job {job_name} : {e}")
+        raise
 
 
 @click.command()
@@ -271,6 +275,7 @@ def list_pods(
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.cancel_job_cli")
 def cancel_job(
     job_name: str,
     namespace: Optional[str],
@@ -287,9 +292,8 @@ def cancel_job(
         result = cancel_training_job_service.cancel_training_job(job_name, namespace)
         click.echo(result)
     except Exception as e:
-        sys.exit(
-            f"Unexpected error happens when trying to cancel training job {job_name} : {e}"
-        )
+        logger.error(f"Unexpected error happens when trying to cancel training job {job_name} : {e}")
+        raise
 
 
 @click.command()
@@ -536,6 +540,7 @@ def cancel_job(
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.start_job_cli")
 def start_job(
     config_file: Optional[str],
     job_name: Optional[str],
@@ -876,6 +881,7 @@ def start_job(
     help="Optional. The namespace to use. If not specified, this command will first use the namespace wh connecting the cluster."
     "Otherwise if namespace is not configured when connecting to the cluster, a namespace that is managed by SageMaker will be auto discovered.",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.patch_job_cli")
 def patch_job(patch_type: str, job_name: str, namespace: Optional[str]):
 
     if patch_type not in JobPatchType.get_values():

src/hyperpod_cli/commands/pod.py

@@ -24,6 +24,8 @@
     setup_logger,
     set_logging_level,
 )
+from hyperpod_cli.telemetry import _hyperpod_telemetry_emitter
+from hyperpod_cli.telemetry.constants import Feature
 
 logger = setup_logger(__name__)
 
@@ -54,6 +56,7 @@
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.get_log_cli")
 def get_log(
     job_name: str,
     pod: str,
@@ -73,9 +76,8 @@ def get_log(
         )
         click.echo(result)
     except Exception as e:
-        sys.exit(
-            f"Unexpected error happens when trying to get logs for training job {job_name} : {e}"
-        )
+        logger.error(f"Unexpected error happens when trying to get logs for training job {job_name} : {e}")
+        raise
 
     try:
         cloudwatch_link = get_logs_service.generate_cloudwatch_link(pod, namespace=namespace)
@@ -148,6 +150,7 @@ def invoke(self, ctx):
     is_flag=True,
     help="Enable debug mode",
 )
+@_hyperpod_telemetry_emitter(Feature.HYPERPOD_V2, "hyperpod_v2.exec_cli")
 def exec(
     job_name: str,
     namespace: Optional[str],
@@ -173,6 +176,5 @@ def exec(
         )
         click.echo(result)
     except Exception as e:
-        sys.exit(
-            f"Unexpected error happens when trying to exec command for pod {pod} : {e}"
-        )
+        logger.error(f"Unexpected error happens when trying to exec command for pod {pod} : {e}")
+        raise

src/hyperpod_cli/telemetry/__init__.py

@@ -10,3 +10,5 @@
 # distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
 # ANY KIND, either express or implied. See the License for the specific
 # language governing permissions and limitations under the License.
+from __future__ import absolute_import
+from .telemetry_logging import _hyperpod_telemetry_emitter

src/hyperpod_cli/telemetry/constants.py

@@ -0,0 +1,60 @@
+from __future__ import absolute_import
+from enum import Enum
+
+
+class Feature(Enum):
+    """Enumeration of feature names used in telemetry."""
+
+    HYPERPOD_V2 = 10
+
+    def __str__(self):  # pylint: disable=E0307
+        """Return the feature name."""
+        return self.name
+
+
+class Status(Enum):
+    """Enumeration of status values used in telemetry."""
+
+    SUCCESS = 1
+    FAILURE = 0
+
+    def __str__(self):  # pylint: disable=E0307
+        """Return the status name."""
+        return self.name
+
+
+class Region(str, Enum):
+    """Telemetry: List of all supported AWS regions."""
+
+    # Classic
+    US_EAST_1 = "us-east-1"  # IAD
+    US_EAST_2 = "us-east-2"  # CMH
+    US_WEST_1 = "us-west-1"  # SFO
+    US_WEST_2 = "us-west-2"  # PDX
+    AP_NORTHEAST_1 = "ap-northeast-1"  # NRT
+    AP_NORTHEAST_2 = "ap-northeast-2"  # ICN
+    AP_NORTHEAST_3 = "ap-northeast-3"  # KIX
+    AP_SOUTH_1 = "ap-south-1"  # BOM
+    AP_SOUTHEAST_1 = "ap-southeast-1"  # SIN
+    AP_SOUTHEAST_2 = "ap-southeast-2"  # SYD
+    CA_CENTRAL_1 = "ca-central-1"  # YUL
+    EU_CENTRAL_1 = "eu-central-1"  # FRA
+    EU_NORTH_1 = "eu-north-1"  # ARN
+    EU_WEST_1 = "eu-west-1"  # DUB
+    EU_WEST_2 = "eu-west-2"  # LHR
+    EU_WEST_3 = "eu-west-3"  # CDG
+    SA_EAST_1 = "sa-east-1"  # GRU
+    # Opt-in
+    AP_EAST_1 = "ap-east-1"  # HKG
+    AP_SOUTHEAST_3 = "ap-southeast-3"  # CGK
+    AF_SOUTH_1 = "af-south-1"  # CPT
+    EU_SOUTH_1 = "eu-south-1"  # MXP
+    ME_SOUTH_1 = "me-south-1"  # BAH
+    MX_CENTRAL_1 = "mx-central-1"  # QRO
+    AP_SOUTHEAST_7 = "ap-southeast-7"  # BKK
+    AP_SOUTH_2 = "ap-south-2"  # HYD
+    AP_SOUTHEAST_4 = "ap-southeast-4"  # MEL
+    EU_CENTRAL_2 = "eu-central-2"  # ZRH
+    EU_SOUTH_2 = "eu-south-2"  # ZAZ
+    IL_CENTRAL_1 = "il-central-1"  # TLV
+    ME_CENTRAL_1 = "me-central-1"  # DXB