NR Control Health Checks

newrelic/config.py

@@ -48,7 +48,7 @@
     default_host,
     fetch_config_setting,
 )
-from newrelic.core.super_agent_health import super_agent_health_instance, super_agent_healthcheck_loop
+from newrelic.core.super_agent_health import HealthStatus, super_agent_health_instance, super_agent_healthcheck_loop
 
 
 __all__ = ["initialize", "filter_app_factory"]
@@ -104,7 +104,7 @@ def _map_aws_account_id(s):
 # all the settings have been read.
 
 _cache_object = []
-super_agent_instance = super_agent_health_instance()
+super_agent_health = super_agent_health_instance()
 
 
 def _reset_config_parser():
@@ -1038,22 +1038,25 @@ def _load_configuration(
 
     # Now read in the configuration file. Cache the config file
     # name in internal settings object as indication of succeeding.
-    if config_file.endswith(".toml"):
-        try:
-            import tomllib
-        except ImportError:
-            raise newrelic.api.exceptions.ConfigurationError(
-                "TOML configuration file can only be used if tomllib is available (Python 3.11+)."
-            )
-        with open(config_file, "rb") as f:
-            content = tomllib.load(f)
-            newrelic_section = content.get("tool", {}).get("newrelic")
-            if not newrelic_section:
-                raise newrelic.api.exceptions.ConfigurationError("New Relic configuration not found in TOML file.")
-            _config_object.read_dict(_toml_config_to_configparser_dict(newrelic_section))
-    elif not _config_object.read([config_file]):
-        super_agent_instance.set_health_status("invalid_config")
-        raise newrelic.api.exceptions.ConfigurationError(f"Unable to open configuration file {config_file}.")
+    try:
+        if config_file.endswith(".toml"):
+            try:
+                import tomllib
+            except ImportError:
+                raise newrelic.api.exceptions.ConfigurationError(
+                    "TOML configuration file can only be used if tomllib is available (Python 3.11+)."
+                )
+            with open(config_file, "rb") as f:
+                content = tomllib.load(f)
+                newrelic_section = content.get("tool", {}).get("newrelic")
+                if not newrelic_section:
+                    raise newrelic.api.exceptions.ConfigurationError("New Relic configuration not found in TOML file.")
+                _config_object.read_dict(_toml_config_to_configparser_dict(newrelic_section))
+        elif not _config_object.read([config_file]):
+            raise newrelic.api.exceptions.ConfigurationError(f"Unable to open configuration file {config_file}.")
+    except Exception:
+        super_agent_health.set_health_status(HealthStatus.INVALID_CONFIG.value)
+        raise
 
     _settings.config_file = config_file
 
@@ -4829,7 +4832,7 @@ def _setup_super_agent_health():
     if super_agent_health_thread.is_alive():
         return
 
-    if super_agent_instance.health_check_enabled:
+    if super_agent_health.health_check_enabled:
         super_agent_health_thread.start()
 
 
@@ -4840,7 +4843,7 @@ def initialize(
     log_file=None,
     log_level=None,
 ):
-    super_agent_instance.start_time_unix_nano = time.time_ns()
+    super_agent_health.start_time_unix_nano = time.time_ns()
 
     if config_file is None:
         config_file = os.environ.get("NEW_RELIC_CONFIG_FILE", None)
@@ -4857,7 +4860,7 @@ def initialize(
 
     if _settings.monitor_mode:
         if not _settings.license_key:
-            super_agent_instance.set_health_status("missing_license")
+            super_agent_health.set_health_status(HealthStatus.MISSING_LICENSE.value)
 
     if _settings.monitor_mode or _settings.developer_mode:
         _settings.enabled = True
@@ -4867,7 +4870,7 @@ def initialize(
         _setup_agent_console()
     else:
         _settings.enabled = False
-        super_agent_instance.set_health_status("agent_disabled")
+        super_agent_health.set_health_status(HealthStatus.AGENT_DISABLED.value)
 
 
 def filter_app_factory(app, global_conf, config_file, environment=None):

newrelic/console.py

@@ -36,7 +36,7 @@
 from newrelic.common.object_wrapper import ObjectProxy
 from newrelic.core.agent import agent_instance
 from newrelic.core.config import flatten_settings, global_settings
-from newrelic.core.super_agent_health import super_agent_health_instance
+from newrelic.core.super_agent_health import HealthStatus, super_agent_health_instance
 from newrelic.core.trace_cache import trace_cache
 
 
@@ -514,7 +514,7 @@ def __init__(self, config_file, stdin=None, stdout=None, log=None):
 
         if not self.__config_object.read([config_file]):
             super_agent_instance = super_agent_health_instance()
-            super_agent_instance.set_health_status("invalid_config")
+            super_agent_instance.set_health_status(HealthStatus.INVALID_CONFIG.value)
             raise RuntimeError(f"Unable to open configuration file {config_file}.")
 
         listener_socket = self.__config_object.get("newrelic", "console.listener_socket") % {"pid": "*"}

newrelic/core/agent.py

@@ -35,6 +35,8 @@
 from newrelic.samplers.cpu_usage import cpu_usage_data_source
 from newrelic.samplers.gc_data import garbage_collector_data_source
 from newrelic.samplers.memory_usage import memory_usage_data_source
+from newrelic.core.super_agent_health import HealthStatus, super_agent_health_instance
+
 
 _logger = logging.getLogger(__name__)
 
@@ -217,6 +219,7 @@ def __init__(self, config):
         self._scheduler = sched.scheduler(self._harvest_timer, self._harvest_shutdown.wait)
 
         self._process_shutdown = False
+        self._super_agent = super_agent_health_instance()
 
         self._lock = threading.Lock()
 
@@ -734,6 +737,11 @@ def shutdown_agent(self, timeout=None):
         if self._harvest_shutdown_is_set():
             return
 
+        self._super_agent.set_health_status(HealthStatus.AGENT_SHUTDOWN.value)
+
+        if self._super_agent.health_check_enabled:
+            self._super_agent.write_to_health_file()
+
         if timeout is None:
             timeout = self._config.shutdown_timeout
 

newrelic/core/agent_protocol.py

@@ -47,7 +47,7 @@
     NetworkInterfaceException,
     RetryDataForRequest,
 )
-from newrelic.core.super_agent_health import super_agent_health_instance
+from newrelic.core.super_agent_health import HealthStatus, super_agent_health_instance
 
 _logger = logging.getLogger(__name__)
 
@@ -251,21 +251,21 @@ def send(
                 # initialize function doesn't get overridden with invalid_license as a missing license key is also
                 # treated as a 401 status code
                 if not self._license_key:
-                    self.super_agent.set_health_status("missing_license")
+                    self.super_agent.set_health_status(HealthStatus.MISSING_LICENSE.value)
                 else:
-                    self.super_agent.set_health_status("invalid_license")
+                    self.super_agent.set_health_status(HealthStatus.INVALID_LICENSE.value)
 
             if status == 407:
-                self.super_agent.set_health_status("proxy_error", status)
+                self.super_agent.set_health_status(HealthStatus.PROXY_ERROR.value, status)
 
             if status == 410:
-                self.super_agent.set_health_status("forced_disconnect")
+                self.super_agent.set_health_status(HealthStatus.FORCED_DISCONNECT.value)
 
             level, message = self.LOG_MESSAGES.get(status, self.LOG_MESSAGES["default"])
 
             # If the default error message was used, then we know we have a general HTTP error
             if message.startswith("Received a non 200 or 202"):
-                self.super_agent.set_health_status("http_error", status, method)
+                self.super_agent.set_health_status(HealthStatus.HTTP_ERROR.value, status, method)
 
             _logger.log(
                 level,

newrelic/core/application.py

@@ -49,7 +49,7 @@
     RetryDataForRequest,
 )
 from newrelic.samplers.data_sampler import DataSampler
-from newrelic.core.super_agent_health import super_agent_healthcheck_loop, super_agent_health_instance
+from newrelic.core.super_agent_health import HealthStatus, super_agent_healthcheck_loop, super_agent_health_instance
 
 _logger = logging.getLogger(__name__)
 
@@ -369,7 +369,7 @@ def connect_to_data_collector(self, activate_agent):
                         None, self._app_name, self.linked_applications, environment_settings()
                     )
                 except ForceAgentDisconnect:
-                    self._super_agent.set_health_status("failed_nr_connection")
+                    self._super_agent.set_health_status(HealthStatus.FAILED_NR_CONNECTION.value)
                     # Any disconnect exception means we should stop trying to connect
                     _logger.error(
                         "The New Relic service has requested that the agent "
@@ -380,7 +380,7 @@ def connect_to_data_collector(self, activate_agent):
                     )
                     return
                 except NetworkInterfaceException:
-                    self._super_agent.set_health_status("failed_nr_connection")
+                    self._super_agent.set_health_status(HealthStatus.FAILED_NR_CONNECTION.value)
                     active_session = None
                 except Exception:
                     # If an exception occurs after agent has been flagged to be
@@ -390,7 +390,7 @@ def connect_to_data_collector(self, activate_agent):
                     # the application is still running.
 
                     if not self._agent_shutdown and not self._pending_shutdown:
-                        self._super_agent.set_health_status("failed_nr_connection")
+                        self._super_agent.set_health_status(HealthStatus.FAILED_NR_CONNECTION.value)
                         _logger.exception(
                             "Unexpected exception when registering "
                             "agent with the data collector. If this problem "
@@ -1700,7 +1700,7 @@ def internal_agent_shutdown(self, restart=False):
         optionally triggers activation of a new session.
 
         """
-        self._super_agent.set_health_status("agent_shutdown")
+        self._super_agent.set_health_status(HealthStatus.AGENT_SHUTDOWN.value)
         if self._super_agent.health_check_enabled:
             self._super_agent.write_to_health_file()