feat!: Adds telemetry client calls to worker (#42)

Signed-off-by: Caden Marofke <marofke@amazon.com>

pyproject.toml

@@ -8,7 +8,7 @@ dynamic = ["version"]
 dependencies = [
     "requests ~= 2.29",
     "boto3 ~= 1.26",
-    "deadline == 0.28.*",
+    "deadline == 0.29.*",
     "openjd-sessions == 0.2.*",
     # tomli became tomllib in standard library in Python 3.11
     "tomli >= 1.1.0 ; python_version<'3.11'",

src/deadline_worker_agent/aws/deadline/__init__.py

@@ -9,7 +9,10 @@
 from botocore.retries.standard import RetryContext
 from botocore.exceptions import ClientError
 
-from ...startup.config import Configuration
+from deadline.client.api import get_telemetry_client, TelemetryClient
+
+from ..._version import __version__ as version  # noqa
+from ...startup.config import Configuration, Capabilities
 from ...boto import DeadlineClient, NoOverflowExponentialBackoff as Backoff
 from ...api_models import (
     AssumeFleetRoleForWorkerResponse,
@@ -735,3 +738,15 @@ def update_worker_schedule(
             raise DeadlineRequestUnrecoverableError(e)
 
     return response
+
+
+def _get_deadline_telemetry_client() -> TelemetryClient:
+    """Wrapper around the Deadline Client Library telemetry client, in order to set package-specific information"""
+    return get_telemetry_client("deadline-cloud-worker-agent", version)
+
+
+def record_worker_start_event(capabilities: Capabilities) -> None:
+    """Calls the telemetry client to record an event capturing generic machine information."""
+    _get_deadline_telemetry_client().record_event(
+        event_type="com.amazon.rum.deadline.worker_agent.start", event_details=capabilities.dict()
+    )

src/deadline_worker_agent/installer/__init__.py

@@ -23,7 +23,7 @@ def install() -> None:
 
     arg_parser = get_argument_parser()
     args = arg_parser.parse_args(namespace=ParsedCommandLineArguments)
-    worker_agent_program = Path(sysconfig.get_path("scripts")) / "deadline-worker-agent"
+    scripts_path = Path(sysconfig.get_path("scripts"))
 
     cmd = [
         "sudo",
@@ -36,8 +36,8 @@ def install() -> None:
         args.region,
         "--user",
         args.user,
-        "--worker-agent-program",
-        str(worker_agent_program),
+        "--scripts-path",
+        str(scripts_path),
     ]
     if args.group:
         cmd += ["--group", args.group]
@@ -49,6 +49,8 @@ def install() -> None:
         cmd.append("--allow-shutdown")
     if not args.install_service:
         cmd.append("--no-install-service")
+    if args.telemetry_opt_out:
+        cmd.append("--telemetry-opt-out")
 
     try:
         run(
@@ -72,6 +74,7 @@ class ParsedCommandLineArguments(Namespace):
     service_start: bool
     allow_shutdown: bool
     install_service: bool
+    telemetry_opt_out: bool
 
 
 def get_argument_parser() -> ArgumentParser:  # pragma: no cover
@@ -122,6 +125,11 @@ def get_argument_parser() -> ArgumentParser:  # pragma: no cover
         action="store_false",
         dest="install_service",
     )
+    parser.add_argument(
+        "--telemetry-opt-out",
+        help="Opts out of telemetry data collection",
+        action="store_true",
+    )
     parser.add_argument(
         "--yes",
         "-y",

src/deadline_worker_agent/installer/install.sh

@@ -37,17 +37,20 @@ fleet_id=unset
 wa_user=$default_wa_user
 confirm=""
 region="us-west-2"
-worker_agent_program="unset"
+scripts_path="unset"
+worker_agent_program="deadline-worker-agent"
+client_library_program="deadline"
 allow_shutdown="no"
 no_install_service="no"
 start_service="no"
+telemetry_opt_out="no"
 warning_lines=()
 
 usage()
 {
     echo "Usage: install.sh --farm-id FARM_ID --fleet-id FLEET_ID"
     echo "                  [--region REGION] [--user USER]"
-    echo "                  [--worker-agent-program WORKER_AGENT_PROGRAM]"
+    echo "                  [--scripts-path SCRIPTS_PATH]"
     echo "                  [-y]"
     echo ""
     echo "Arguments"
@@ -64,15 +67,18 @@ usage()
     echo "        A group name that the Worker Agent shares with the user(s) that Jobs will be running as."
     echo "        Do not use the primary/effective group of the Worker Agent user specifeid in --user as"
     echo "        this is not a secure configuration. Defaults to $default_job_group."
-    echo "    --worker-agent-program WORKER_AGENT_PROGRAM"
-    echo "        An optional path to the Worker Agent program. This is used as the program path"
-    echo "        when creating the systemd service. If not specified, the first program named"
-    echo "        deadline-worker-agent found in the search path will be used."
+    echo "    --scripts-path SCRIPTS_PATH"
+    echo "        An optional path to the directory that the Worker Agent and Deadline Cloud Library are"
+    echo "        installed. This is used as the program path when creating the systemd service for the "
+    echo "        Worker Agent. If not specified, the first program named 'deadline-worker-agent' and"
+    echo "        'deadline' found in the search path will be used."
     echo "    --allow-shutdown"
     echo "        Dictates whether a sudoers rule is created/deleted allowing the worker agent the"
     echo "        ability to shutdown the host system"
     echo "    --no-install-service"
     echo "        Skips the worker agent systemd service installation"
+    echo "    --telemetry-opt-out"
+    echo "        Opts out of telemetry collection for the worker agent"
     echo "    --start"
     echo "        Starts the systemd service as part of the installation. By default, the systemd"
     echo "        service is configured to start on system boot, but not started immediately."
@@ -104,7 +110,7 @@ validate_deadline_id() {
 }
 
 # Validate arguments
-PARSED_ARGUMENTS=$(getopt -n install.sh --longoptions farm-id:,fleet-id:,region:,user:,group:,worker-agent-program:,start,allow-shutdown,no-install-service -- "y" "$@")
+PARSED_ARGUMENTS=$(getopt -n install.sh --longoptions farm-id:,fleet-id:,region:,user:,group:,scripts-path:,start,allow-shutdown,no-install-service,telemetry-opt-out -- "y" "$@")
 VALID_ARGUMENTS=$?
 if [ "${VALID_ARGUMENTS}" != "0" ]; then
     usage
@@ -122,9 +128,10 @@ do
     --region)                region="$2"                ; shift 2 ;;
     --user)                  wa_user="$2"               ; shift 2 ;;
     --group)                 job_group="$2"             ; shift 2 ;;
-    --worker-agent-program)  worker_agent_program="$2"  ; shift 2 ;;
+    --scripts-path)          scripts_path="$2"          ; shift 2 ;;
     --allow-shutdown)        allow_shutdown="yes"       ; shift   ;;
     --no-install-service)    no_install_service="yes"   ; shift   ;;
+    --telemetry-opt-out)     telemetry_opt_out="yes"    ; shift   ;;
     --start)                 start_service="yes"        ; shift   ;;
     -y)                      confirm="$1"               ; shift   ;;
     # -- means the end of the arguments; drop this, and break out of the while loop
@@ -152,18 +159,38 @@ elif ! validate_deadline_id fleet "${fleet_id}"; then
     echo "ERROR: Non a valid value for --fleet-id: ${fleet_id}"
     usage
 fi
-if [[ "${worker_agent_program}" == "unset" ]]; then
+if [[ "${scripts_path}" == "unset" ]]; then
     set +e
     worker_agent_program=$(which deadline-worker-agent)
     if [[ "$?" != "0" ]]; then
         echo "ERROR: Could not find deadline-worker-agent in search path"
         exit 1
     fi
+    client_library_program=$(which deadline)
+    if [[ "$?" != "0" ]]; then
+        echo "ERROR: Could not find deadline in search path"
+        exit 1
+    fi
     set -e
-elif [[ ! -f "${worker_agent_program}" ]]; then
-    echo "ERROR: The specified Worker Agent path is not found: \"${worker_agent_program}\""
+elif [[ ! -d "${scripts_path}" ]]; then
+    echo "ERROR: The specified scripts path is not found: \"${scripts_path}\""
     usage
+else
+    set +e
+    # We have a provided scripts path, so we append it to the program paths
+    worker_agent_program="${scripts_path}"/deadline-worker-agent
+    if [[ ! -f "${worker_agent_program}" ]]; then
+        echo "ERROR: Could not find deadline-worker-agent in scripts path: \"${worker_agent_program}\""
+        exit 1
+    fi
+    client_library_program="${scripts_path}"/deadline
+    if [[ ! -f "${client_library_program}" ]]; then
+        echo "ERROR: Could not find deadline in scripts path: \"${client_library_program}\""
+        exit 1
+    fi
+    set -e
 fi
+
 if [[ ! -z "${region}" ]] && [[ ! "${region}" =~ ^[a-z]+-[a-z]+-[0-9]+$ ]]; then
     echo "ERROR: Not a valid value for --region: ${region}"
     usage
@@ -196,9 +223,12 @@ echo "Fleet ID: ${fleet_id}"
 echo "Region: ${region}"
 echo "Worker agent user: ${wa_user}"
 echo "Worker job group: ${job_group}"
+echo "Scripts path: ${scripts_path}"
 echo "Worker agent program path: ${worker_agent_program}"
+echo "Worker agent program path: ${client_library_program}"
 echo "Allow worker agent shutdown: ${allow_shutdown}"
 echo "Start systemd service: ${start_service}"
+echo "Telemetry opt-out: ${telemetry_opt_out}"
 
 # Confirmation prompt
 if [ -z "$confirm" ]; then
@@ -352,6 +382,12 @@ EOF
     fi
 fi
 
+if [[ "${telemetry_opt_out}" == "yes" ]]; then
+    # Set the Deadline Client Lib configuration setting
+    echo "Opting out of telemetry collection"
+    sudo -u $wa_user $client_library_program config set telemetry.opt_out true
+fi
+
 echo "Done"
 
 # Output warning lines if any

src/deadline_worker_agent/startup/entrypoint.py

@@ -29,6 +29,12 @@
 from .bootstrap import bootstrap_worker
 from .capabilities import AmountCapabilityName, AttributeCapabilityName, Capabilities
 from .config import Capabilities, Configuration, ConfigurationError
+from ..aws.deadline import (
+    DeadlineRequestError,
+    delete_worker,
+    update_worker,
+    record_worker_start_event,
+)
 
 __all__ = ["entrypoint"]
 _logger = logging.getLogger(__name__)
@@ -86,6 +92,7 @@ def entrypoint(cli_args: Optional[list[str]] = None) -> None:
         # if customer manually provided the capabilities (to be added in this function)
         # then we default to the customer provided ones
         system_capabilities = detect_system_capabilities()
+        record_worker_start_event(system_capabilities)
         config.capabilities = system_capabilities.merge(config.capabilities)
 
         # Log the configuration

test/unit/install/test_install.py

@@ -72,6 +72,11 @@ def allow_shutdown() -> bool:
     return False
 
 
+@pytest.fixture
+def telemetry_opt_out() -> bool:
+    return True
+
+
 @pytest.fixture
 def install_service() -> bool:
     return True
@@ -88,6 +93,7 @@ def parsed_args(
     confirmed: bool,
     allow_shutdown: bool,
     install_service: bool,
+    telemetry_opt_out: bool,
 ) -> ParsedCommandLineArguments:
     parsed_args = ParsedCommandLineArguments()
     parsed_args.farm_id = farm_id
@@ -99,6 +105,7 @@ def parsed_args(
     parsed_args.confirmed = confirmed
     parsed_args.allow_shutdown = allow_shutdown
     parsed_args.install_service = install_service
+    parsed_args.telemetry_opt_out = telemetry_opt_out
     return parsed_args
 
 
@@ -131,8 +138,8 @@ def expected_cmd(
         parsed_args.region,
         "--user",
         parsed_args.user,
-        "--worker-agent-program",
-        os.path.join(sysconfig.get_path("scripts"), "deadline-worker-agent"),
+        "--scripts-path",
+        sysconfig.get_path("scripts"),
     ]
     if parsed_args.group is not None:
         expected_cmd.extend(("--group", parsed_args.group))
@@ -142,6 +149,8 @@ def expected_cmd(
         expected_cmd.append("--start")
     if parsed_args.allow_shutdown:
         expected_cmd.append("--allow-shutdown")
+    if parsed_args.telemetry_opt_out:
+        expected_cmd.append("--telemetry-opt-out")
     return expected_cmd
 
 
@@ -218,6 +227,12 @@ def user(self, request: pytest.FixtureRequest) -> str:
     def allow_shutdown(self, request: pytest.FixtureRequest) -> bool:
         return request.param
 
+    @pytest.fixture(
+        params=(True, False),
+    )
+    def telemetry_opt_out(self, request: pytest.FixtureRequest) -> bool:
+        return request.param
+
     @pytest.fixture(
         params=(
             True,

test/unit/startup/test_entrypoint.py

@@ -148,6 +148,12 @@ def block_rich_import() -> Generator[None, None, None]:
         yield
 
 
+@pytest.fixture(autouse=True)
+def block_telemetry_client() -> Generator[MagicMock, None, None]:
+    with patch.object(entrypoint_mod, "record_worker_start_event") as telem_mock:
+        yield telem_mock
+
+
 def test_calls_worker_run(
     mock_worker_run: MagicMock,
 ) -> None: