Add max three app name status for agent control health check

newrelic/config.py

@@ -42,14 +42,17 @@
 from newrelic.common.log_file import initialize_logging
 from newrelic.common.object_names import callable_name, expand_builtin_exception_name
 from newrelic.core import trace_cache
+from newrelic.core.agent_control_health import (
+    HealthStatus,
+    agent_control_health_instance,
+    agent_control_healthcheck_loop,
+)
 from newrelic.core.config import (
     Settings,
     apply_config_setting,
     default_host,
     fetch_config_setting,
 )
-from newrelic.core.agent_control_health import HealthStatus, agent_control_health_instance, agent_control_healthcheck_loop
-
 
 __all__ = ["initialize", "filter_app_factory"]
 
@@ -597,12 +600,16 @@ def _process_app_name_setting():
     # primary application name and link it with the other applications.
     # When activating the application the linked names will be sent
     # along to the core application where the association will be
-    # created if the do not exist.
+    # created if it does not exist.
 
-    name = _settings.app_name.split(";")[0].strip() or "Python Application"
+    app_name_list = _settings.app_name.split(";")
+    name = app_name_list[0].strip() or "Python Application"
+
+    if len(app_name_list) > 3:
+        agent_control_health.set_health_status(HealthStatus.MAX_APP_NAME.value)
 
     linked = []
-    for altname in _settings.app_name.split(";")[1:]:
+    for altname in app_name_list[1:]:
         altname = altname.strip()
         if altname:
             linked.append(altname)
@@ -4833,7 +4840,9 @@ def _setup_agent_console():
         newrelic.core.agent.Agent.run_on_startup(_startup_agent_console)
 
 
-agent_control_health_thread = threading.Thread(name="Agent-Control-Health-Main-Thread", target=agent_control_healthcheck_loop)
+agent_control_health_thread = threading.Thread(
+    name="Agent-Control-Health-Main-Thread", target=agent_control_healthcheck_loop
+)
 agent_control_health_thread.daemon = True
 
 

newrelic/core/agent_control_health.py

@@ -21,8 +21,8 @@
 from enum import IntEnum
 from pathlib import Path
 from urllib.parse import urlparse
-from newrelic.core.config import _environ_as_bool, _environ_as_int
 
+from newrelic.core.config import _environ_as_bool, _environ_as_int
 
 _logger = logging.getLogger(__name__)
 
@@ -33,6 +33,7 @@ class HealthStatus(IntEnum):
     MISSING_LICENSE = 2
     FORCED_DISCONNECT = 3
     HTTP_ERROR = 4
+    MAX_APP_NAME = 6
     PROXY_ERROR = 7
     AGENT_DISABLED = 8
     FAILED_NR_CONNECTION = 9
@@ -47,6 +48,7 @@ class HealthStatus(IntEnum):
     HealthStatus.MISSING_LICENSE.value: "License key missing in configuration",
     HealthStatus.FORCED_DISCONNECT.value: "Forced disconnect received from New Relic (HTTP status code 410)",
     HealthStatus.HTTP_ERROR.value: "HTTP error response code {response_code} received from New Relic while sending data type {info}",
+    HealthStatus.MAX_APP_NAME.value: "The maximum number of configured app names (3) exceeded",
     HealthStatus.PROXY_ERROR.value: "HTTP Proxy configuration error; response code {response_code}",
     HealthStatus.AGENT_DISABLED.value: "Agent is disabled via configuration",
     HealthStatus.FAILED_NR_CONNECTION.value: "Failed to connect to New Relic data collector",
@@ -61,6 +63,8 @@ class HealthStatus(IntEnum):
 )
 LICENSE_KEY_ERROR_CODES = frozenset([HealthStatus.INVALID_LICENSE.value, HealthStatus.MISSING_LICENSE.value])
 
+NR_CONNECTION_ERROR_CODES = frozenset([HealthStatus.FAILED_NR_CONNECTION.value, HealthStatus.FORCED_DISCONNECT.value])
+
 
 def is_valid_file_delivery_location(file_uri):
     # Verify whether file directory provided to agent via env var is a valid file URI to determine whether health
@@ -150,7 +154,10 @@ def set_health_status(self, status_code, response_code=None, info=None):
         previous_status_code = self.status_code
 
         if status_code == HealthStatus.FAILED_NR_CONNECTION.value and previous_status_code in LICENSE_KEY_ERROR_CODES:
-            # Do not update to failed connection status when license key is the issue
+            # Do not update to failed connection status when license key is the issue so the more descriptive status is not overridden
+            return
+        elif status_code in NR_CONNECTION_ERROR_CODES and previous_status_code == HealthStatus.MAX_APP_NAME:
+            # Do not let NR connection error override the max app name status
             return
         elif status_code == HealthStatus.AGENT_SHUTDOWN.value and not self.is_healthy:
             # Do not override status with agent_shutdown unless the agent was previously healthy

tests/agent_features/test_agent_control_health_check.py

@@ -12,17 +12,23 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import os
-import time
 import re
-import pytest
 import threading
+import time
 
-from newrelic.core.config import finalize_application_settings
+import pytest
+from testing_support.fixtures import initialize_agent
 from testing_support.http_client_recorder import HttpClientRecorder
-from newrelic.core.agent_control_health import HealthStatus, is_valid_file_delivery_location, agent_control_health_instance
-from newrelic.config import initialize, _reset_configuration_done
+
+from newrelic.config import _reset_configuration_done, initialize
+from newrelic.core.agent_control_health import (
+    HealthStatus,
+    agent_control_health_instance,
+    is_valid_file_delivery_location,
+)
 from newrelic.core.agent_protocol import AgentProtocol
 from newrelic.core.application import Application
+from newrelic.core.config import finalize_application_settings, global_settings
 from newrelic.network.exceptions import DiscardDataForRequest
 
 
@@ -34,6 +40,7 @@ def get_health_file_contents(tmp_path):
         contents = f.readlines()
         return contents
 
+
 @pytest.mark.parametrize("file_uri", ["", "file://", "/test/dir", "foo:/test/dir"])
 def test_invalid_file_directory_supplied(file_uri):
     assert not is_valid_file_delivery_location(file_uri)
@@ -168,6 +175,31 @@ def test_proxy_error_status(monkeypatch, tmp_path):
     assert contents[4] == "last_error: NR-APM-007\n"
 
 
+def test_multiple_activations_running_threads(monkeypatch, tmp_path):
+    # Setup expected env vars to run agent control health check
+    monkeypatch.setenv("NEW_RELIC_AGENT_CONTROL_ENABLED", True)
+    file_path = tmp_path.as_uri()
+    monkeypatch.setenv("NEW_RELIC_AGENT_CONTROL_HEALTH_DELIVERY_LOCATION", file_path)
+
+    _reset_configuration_done()
+    initialize()
+
+    application_1 = Application("Test App 1")
+    application_2 = Application("Test App 2")
+
+    application_1.activate_session()
+    application_2.activate_session()
+
+    running_threads = threading.enumerate()
+
+    # 6 threads expected: One main agent thread, two active session threads, one main health check thread, and two
+    # active session health threads
+    assert len(running_threads) == 6
+    assert running_threads[1].name == "Agent-Control-Health-Main-Thread"
+    assert running_threads[2].name == "Agent-Control-Health-Session-Thread"
+    assert running_threads[4].name == "Agent-Control-Health-Session-Thread"
+
+
 def test_update_to_healthy(monkeypatch, tmp_path):
     # Setup expected env vars to run agent control health check
     monkeypatch.setenv("NEW_RELIC_AGENT_CONTROL_ENABLED", True)
@@ -200,26 +232,25 @@ def test_update_to_healthy(monkeypatch, tmp_path):
     assert contents[1] == "status: Healthy\n"
 
 
-def test_multiple_activations_running_threads(monkeypatch, tmp_path):
+def test_max_app_name_status(monkeypatch, tmp_path):
     # Setup expected env vars to run agent control health check
     monkeypatch.setenv("NEW_RELIC_AGENT_CONTROL_ENABLED", True)
     file_path = tmp_path.as_uri()
     monkeypatch.setenv("NEW_RELIC_AGENT_CONTROL_HEALTH_DELIVERY_LOCATION", file_path)
 
     _reset_configuration_done()
-    initialize()
-
-    application_1 = Application("Test App 1")
-    application_2 = Application("Test App 2")
+    initialize_agent(app_name="test1;test2;test3;test4")
+    # Give time for the scheduler to kick in and write to the health file
+    time.sleep(5)
 
-    application_1.activate_session()
-    application_2.activate_session()
+    contents = get_health_file_contents(tmp_path)
 
-    running_threads = threading.enumerate()
+    # Assert on contents of health file
+    assert len(contents) == 5
+    assert contents[0] == "healthy: False\n"
+    assert contents[1] == "status: The maximum number of configured app names (3) exceeded\n"
+    assert contents[4] == "last_error: NR-APM-006\n"
 
-    # 6 threads expected: One main agent thread, two active session threads, one main health check thread, and two
-    # active session health threads
-    assert len(running_threads) == 6
-    assert running_threads[1].name == "Agent-Control-Health-Main-Thread"
-    assert running_threads[2].name == "Agent-Control-Health-Session-Thread"
-    assert running_threads[4].name == "Agent-Control-Health-Session-Thread"
+    # Set app name back to original name specific
+    settings = global_settings()
+    settings.app_name = "Python Agent Test (agent_features)"