Enable Hyperpod telemetry

Author: Roja Reddy Sareddy

setup.py

@@ -89,7 +89,6 @@
         "hyperpod-pytorch-job-template>=1.0.0, <2.0.0",
         "hyperpod-custom-inference-template>=1.0.0, <2.0.0",
         "hyperpod-jumpstart-inference-template>=1.0.0, <2.0.0",
-        "sagemaker",
     ],
     entry_points={
         "console_scripts": [

src/sagemaker/hyperpod/cli/commands/cluster.py

@@ -42,7 +42,7 @@
     TEMP_KUBE_CONFIG_FILE,
     OutputFormat,
 )
-from sagemaker.hyperpod.cli.telemetry.user_agent import (
+from sagemaker.hyperpod.common.telemetry.user_agent import (
     get_user_agent_extra_suffix,
 )
 from sagemaker.hyperpod.cli.service.list_pods import (

src/sagemaker/hyperpod/common/telemetry/constants.py

@@ -0,0 +1,65 @@
+from __future__ import absolute_import
+from enum import Enum
+
+
+class Feature(Enum):
+    """Enumeration of feature names used in telemetry."""
+
+    SDK_DEFAULTS = 1
+    LOCAL_MODE = 2
+    REMOTE_FUNCTION = 3
+    MODEL_TRAINER = 4
+    ESTIMATOR = 5
+    HYPERPOD = 6  # Added to support telemetry in sagemaker-hyperpod-cli
+
+    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

src/sagemaker/hyperpod/common/telemetry/telemetry_logging.py

@@ -10,13 +10,7 @@
 import re
 
 import boto3
-from sagemaker.telemetry.constants import Feature, Status, Region
-from sagemaker.telemetry.telemetry_logging import (
-    FEATURE_TO_CODE,
-    STATUS_TO_CODE,
-    _requests_helper,
-    _construct_url,
-)
+from sagemaker.hyperpod.common.telemetry.constants import Feature, Status, Region
 import importlib.metadata
 
 SDK_VERSION = importlib.metadata.version("sagemaker-hyperpod")
@@ -28,6 +22,20 @@
     sys.version_info.major, sys.version_info.minor, sys.version_info.micro
 )
 
+FEATURE_TO_CODE = {
+    str(Feature.SDK_DEFAULTS): 1,
+    str(Feature.LOCAL_MODE): 2,
+    str(Feature.REMOTE_FUNCTION): 3,
+    str(Feature.MODEL_TRAINER): 4,
+    str(Feature.ESTIMATOR): 5,
+    str(Feature.HYPERPOD): 6,  # Added to support telemetry in sagemaker-hyperpod-cli
+}
+
+STATUS_TO_CODE = {
+    str(Status.SUCCESS): 1,
+    str(Status.FAILURE): 0,
+}
+
 logger = logging.getLogger(__name__)
 
 
@@ -65,6 +73,43 @@ def get_region_and_account_from_current_context() -> Tuple[str, str]:
     return DEFAULT_AWS_REGION, "unknown"
 
 
+def _requests_helper(url, timeout):
+    """Make a GET request to the given URL"""
+
+    response = None
+    try:
+        response = requests.get(url, timeout)
+    except requests.exceptions.RequestException as e:
+        logger.exception("Request exception: %s", str(e))
+    return response
+
+
+def _construct_url(
+    accountId: str,
+    region: str,
+    status: str,
+    feature: str,
+    failure_reason: str,
+    failure_type: str,
+    extra_info: str,
+) -> str:
+    """Construct the URL for the telemetry request"""
+
+    base_url = (
+        f"https://sm-pysdk-t-{region}.s3.{region}.amazonaws.com/telemetry?"
+        f"x-accountId={accountId}"
+        f"&x-status={status}"
+        f"&x-feature={feature}"
+    )
+    logger.debug("Failure reason: %s", failure_reason)
+    if failure_reason:
+        base_url += f"&x-failureReason={failure_reason}"
+        base_url += f"&x-failureType={failure_type}"
+    if extra_info:
+        base_url += f"&x-extra={extra_info}"
+    return base_url
+
+
 def _send_telemetry_request(
     status: int,
     feature_list: List[int],

src/sagemaker/hyperpod/inference/hp_endpoint.py

@@ -14,7 +14,7 @@
 from sagemaker.hyperpod.common.telemetry.telemetry_logging import (
     _hyperpod_telemetry_emitter,
 )
-from sagemaker.telemetry.constants import Feature
+from sagemaker.hyperpod.common.telemetry.constants import Feature
 from sagemaker.hyperpod.inference.hp_endpoint_base import HPEndpointBase
 from typing import Dict, List, Optional
 from sagemaker_core.main.resources import Endpoint

src/sagemaker/hyperpod/inference/hp_endpoint_base.py

@@ -18,7 +18,7 @@
 from sagemaker.hyperpod.common.telemetry.telemetry_logging import (
     _hyperpod_telemetry_emitter,
 )
-from sagemaker.telemetry.constants import Feature
+from sagemaker.hyperpod.common.telemetry.constants import Feature
 
 
 class HPEndpointBase:

src/sagemaker/hyperpod/inference/hp_jumpstart_endpoint.py

@@ -19,7 +19,7 @@
 from sagemaker.hyperpod.common.telemetry.telemetry_logging import (
     _hyperpod_telemetry_emitter,
 )
-from sagemaker.telemetry.constants import Feature
+from sagemaker.hyperpod.common.telemetry.constants import Feature
 
 
 class HPJumpStartEndpoint(_HPJumpStartEndpoint, HPEndpointBase):

src/sagemaker/hyperpod/training/hyperpod_pytorch_job.py

@@ -16,7 +16,7 @@
 from sagemaker.hyperpod.common.telemetry.telemetry_logging import (
     _hyperpod_telemetry_emitter,
 )
-from sagemaker.telemetry.constants import Feature
+from sagemaker.hyperpod.common.telemetry.constants import Feature
 import yaml
 import logging