Fixed issues with telemetry.

ptlinux/telemetry.py

@@ -34,6 +34,8 @@ DEFAULT_TELEMETRY_ENDPOINT = "https://pristontale.eu/api/launcher/telemetry"
 TELEMETRY_ENDPOINT_VARIABLE = "PTLINUX_TELEMETRY_ENDPOINT"
 TELEMETRY_KEY_VARIABLE = "PTLINUX_TELEMETRY_KEY"
 TELEMETRY_DEBUG_VARIABLE = "PTLINUX_TELEMETRY_DEBUG"
+TELEMETRY_PROCESS_ROLE_VARIABLE = "PTLINUX_TELEMETRY_PROCESS_ROLE"
+TELEMETRY_OPERATION_PROCESS_ROLE = "operation"
 
 MAX_REQUEST_BYTES = 64 * 1024
 MAX_MESSAGE_BYTES = 10 * 1024
@@ -263,9 +265,14 @@ class TelemetryService:
 		self._session_metadata = dict(session_metadata or {})
 		self._monotonic = monotonic
 		self._started_at = monotonic()
+		self._reports_launcher_lifecycle = (
+			self._environment.get(TELEMETRY_PROCESS_ROLE_VARIABLE, "").strip().casefold()
+			!= TELEMETRY_OPERATION_PROCESS_ROLE
+		)
 		self._enabled = bool(enabled)
 		self._active = False
 		self._accepting = False
+		self._closing = False
 		self._ready = False
 		self._first_run = False
 		self._installation_id: str | None = None
@@ -322,6 +329,7 @@ class TelemetryService:
 				self._base_payload = self._collect_common_payload(first_seen)
 				self._stop.clear()
 				self._accepting = True
+				self._closing = False
 				self._active = True
 				worker = threading.Thread(
 					target=self._worker_loop,
@@ -338,15 +346,16 @@ class TelemetryService:
 				self._debug_log(f"initialization failed: {type(exc).__name__}")
 				return
 
-		self.send_event(
-			"launcher_started",
-			metadata={"cold_start": True, **self._session_metadata},
-		)
-		if self._first_run:
+		if self._reports_launcher_lifecycle:
 			self.send_event(
-				"first_run_started",
-				metadata={"install_type": self._install_type()},
+				"launcher_started",
+				metadata={"cold_start": True, **self._session_metadata},
 			)
+			if self._first_run:
+				self.send_event(
+					"first_run_started",
+					metadata={"install_type": self._install_type()},
+				)
 		self._start_environment_collection()
 
 	def set_enabled(self, enabled: bool) -> None:
@@ -368,6 +377,8 @@ class TelemetryService:
 			if self._ready or not self._active:
 				return
 			self._ready = True
+			if not self._reports_launcher_lifecycle:
+				return
 		duration_ms = round((self._monotonic() - self._started_at) * 1000)
 		self.send_event(
 			"launcher_ready",
@@ -674,11 +685,12 @@ class TelemetryService:
 			self._send_repeated_error(state)
 
 	def shutdown(self, timeout_seconds: float = SHUTDOWN_WAIT_SECONDS) -> None:
-		"""Queue the closing event and give the daemon worker a bounded flush."""
+		"""Finish this process role and give the daemon worker a bounded flush."""
 
 		with self._state_lock:
-			if not self._active:
+			if not self._active or self._closing:
 				return
+			self._closing = True
 		deadline = time.monotonic() + max(0.0, timeout_seconds)
 		environment_worker = self._environment_worker
 		if isinstance(environment_worker, threading.Thread) and (
@@ -692,17 +704,18 @@ class TelemetryService:
 			)
 		self._flush_repeated_errors()
 		runtime_seconds = max(0, round(self._monotonic() - self._started_at))
-		if self._first_run and not self._ready:
+		if self._reports_launcher_lifecycle:
+			if self._first_run and not self._ready:
+				self.send_event(
+					"first_run_failed",
+					severity="error",
+					message="Launcher did not reach the ready state",
+					metadata={"duration_ms": runtime_seconds * 1000},
+				)
 			self.send_event(
-				"first_run_failed",
-				severity="error",
-				message="Launcher did not reach the ready state",
-				metadata={"duration_ms": runtime_seconds * 1000},
+				"launcher_closed",
+				metadata={"runtime_seconds": runtime_seconds, **self._session_metadata},
 			)
-		self.send_event(
-			"launcher_closed",
-			metadata={"runtime_seconds": runtime_seconds, **self._session_metadata},
-		)
 		with self._state_lock:
 			self._accepting = False
 			self._stop.set()
@@ -719,6 +732,7 @@ class TelemetryService:
 		with self._state_lock:
 			self._enabled = False
 			self._accepting = False
+			self._closing = True
 			self._stop.set()
 			self._repeated_errors.clear()
 		while True:
@@ -795,6 +809,11 @@ class TelemetryService:
 					self._base_payload[key] = {**existing, **value}
 				else:
 					self._base_payload[key] = value
+			if not self._reports_launcher_lifecycle:
+				# Dashboard operation subprocesses enrich their own events with the
+				# detected environment, while the dashboard remains the sole source
+				# of launcher lifecycle and automatic compatibility events.
+				return
 
 		runtime = sanitized.get("runtime")
 		dependencies = sanitized.get("dependencies")
@@ -816,14 +835,38 @@ class TelemetryService:
 			context={"operation": "launcher", "stage": "environment_detection"},
 		)
 		missing_required = dependency_details.get("missing_required")
-		if dependency_details.get("healthy") is False:
+		generic_dependency_failures = (
+			[
+				str(value)
+				for value in missing_required
+				if isinstance(value, str) and value
+			]
+			if isinstance(missing_required, list)
+			else []
+		)
+		if (
+			dependency_details.get("container_ready") is False
+			and "container_host" not in generic_dependency_failures
+		):
+			generic_dependency_failures.append("container_host")
+		if (
+			dependency_details.get("healthy") is False
+			and generic_dependency_failures
+		):
 			self.send_event(
 				"dependency_check_failed",
 				severity="warning",
 				message="One or more launcher dependencies are unavailable",
 				metadata={
-					"dependencies": (
-						missing_required if isinstance(missing_required, list) else []
+					"dependencies": generic_dependency_failures,
+					"required_shared_libraries_present": dependency_details.get(
+						"required_shared_libraries_present"
+					),
+					"vulkan_64bit_available": dependency_details.get(
+						"vulkan_64bit_available"
+					),
+					"vulkan_32bit_available": dependency_details.get(
+						"vulkan_32bit_available"
 					),
 				},
 				context={"operation": "launcher", "stage": "dependency_check"},
@@ -846,7 +889,17 @@ class TelemetryService:
 			self.send_event(
 				"vulkan_32bit_check_failed",
 				severity="warning",
-				metadata={"dependency": "vulkan_32bit", "check": "32bit_vulkan"},
+				metadata={
+					"dependency": "vulkan_32bit",
+					"check": "32bit_vulkan",
+					"loader_available": runtime_details.get(
+						"vulkan_32bit_loader_available"
+					),
+					"icd_available": runtime_details.get(
+						"vulkan_32bit_icd_available"
+					),
+					"detection": runtime_details.get("vulkan_32bit_detection"),
+				},
 				context={"operation": "launcher", "stage": "vulkan_check"},
 			)