add smoke test for crash logs from python [INPLAT-36] (#3243)

Co-authored-by: Tony Hsu <tony.hsu@datadoghq.com>
Co-authored-by: brettlangdon <me@brett.is>
Co-authored-by: Brett Langdon <brett.langdon@datadoghq.com>

lib-injection/build/docker/python/dd-lib-python-init-test-django/django_app.py

@@ -24,6 +24,13 @@ def index(request):
     return HttpResponse("test")
 
 
+def crashme(request):
+    import ctypes
+
+    ctypes.string_at(0)
+
+
 urlpatterns = [
     path("", index),
+    path("crashme", crashme),
 ]

manifests/dd_apm_inject.yml

@@ -1,4 +1,6 @@
 tests/:
   docker_ssi/:
     test_docker_ssi.py:
-      TestDockerSSIFeatures: v0.19.1
+      TestDockerSSIFeatures: v0.19.1
+    test_docker_ssi_crash.py:
+      TestDockerSSICrash: v0.19.1

tests/docker_ssi/test_docker_ssi_crash.py

@@ -0,0 +1,49 @@
+from urllib.parse import urlparse
+
+from utils import (
+    bug,
+    scenarios,
+    features,
+    context,
+    interfaces,
+)
+from utils import weblog
+from utils.tools import logger
+
+
+@scenarios.docker_ssi
+class TestDockerSSICrash:
+    """Test the ssi in a simulated host injection environment (docker container + test agent)
+    We test scenarios when the application crashes and sends a crash report."""
+
+    _r = None
+
+    def setup_crash(self):
+        if TestDockerSSICrash._r is None:
+            parsed_url = urlparse(context.scenario.weblog_url + "/crashme")
+            logger.info(f"Setting up Docker SSI installation WEBLOG_URL {context.scenario.weblog_url}")
+            TestDockerSSICrash._r = weblog.request(
+                "GET", parsed_url.path, domain=parsed_url.hostname, port=parsed_url.port
+            )
+            logger.info(f"Setup Docker SSI installation {TestDockerSSICrash._r}")
+
+        self.r = TestDockerSSICrash._r
+
+    @features.ssi_crashtracking
+    @bug(condition=context.library != "python", reason="INPLAT-11")
+    def test_crash(self):
+        """Validate that a crash report is generated when the application crashes"""
+        logger.info(f"Testing Docker SSI crash tracking: {context.scenario.library.library}")
+        assert (
+            self.r.status_code is None
+        ), f"Response from request {context.scenario.weblog_url + '/crashme'} was supposed to fail: {self.r}"
+
+        # No traces should have been generated
+        assert not interfaces.test_agent.get_traces(
+            self.r
+        ), f"Traces found for request {context.scenario.weblog_url + '/crashme'}"
+
+        # Crash report should have been generated
+        crash_reports = interfaces.test_agent.get_crash_reports()
+        assert crash_reports, "No crash report found"
+        assert len(crash_reports) == 1, "More than one crash report found"

utils/_context/_scenarios/docker_ssi.py

@@ -141,9 +141,13 @@ def fill_context(self, json_tested_components):
             logger.stdout(f"{key}: {self._tested_components[key]}")
 
     def post_setup(self):
-        logger.stdout("--- Waiting for all traces to be sent to test agent ---")
-        time.sleep(5)  # wait for the traces to be sent to the test agent
-        interfaces.test_agent.collect_data(f"{self.host_log_folder}/interfaces/test_agent")
+        logger.stdout("--- Waiting for all traces and telemetry to be sent to test agent ---")
+        data = None
+        attempts = 0
+        while attempts < 30 and not data:
+            attempts += 1
+            data = interfaces.test_agent.collect_data(f"{self.host_log_folder}/interfaces/test_agent")
+            time.sleep(5)
 
     @property
     def library(self):

utils/_features.py

@@ -2407,6 +2407,16 @@ def ssi_guardrails(test_object):
         pytest.mark.features(feature_id=322)(test_object)
         return test_object
 
+    @staticmethod
+    def ssi_crashtracking(test_object):
+        """
+        Docker ssi crashtracking
+
+        https://feature-parity.us1.prod.dog/#/?feature=340
+        """
+        pytest.mark.features(feature_id=340)(test_object)
+        return test_object
+
     @staticmethod
     def ssi_service_naming(test_object):
         """

utils/interfaces/_test_agent.py

@@ -56,6 +56,11 @@ def get_telemetry_for_runtime(self, runtime_id):
 
         return telemetry_msgs
 
+    def get_crashlog_for_runtime(self, runtime_id):
+        logger.debug(f"Try to find a crashlog related to runtime-id {runtime_id}")
+        assert runtime_id is not None, "Runtime ID not found"
+        return [l for l in self.get_telemetry_logs() if l["runtime_id"] == runtime_id]
+
     def get_telemetry_for_autoinject(self):
         logger.debug("Try to find telemetry data related to autoinject")
         injection_metrics = []
@@ -79,3 +84,16 @@ def get_telemetry_for_autoinject_library_entrypoint(self):
             if str(series["metric"]).startswith("library_entrypoint.")
         ]
         return injection_metrics
+
+    def get_telemetry_logs(self):
+        logger.debug("Try to find telemetry data related to logs")
+        return [t for t in self._data_telemetry_list if t["request_type"] == "logs"]
+
+    def get_crash_reports(self):
+        logger.debug("Try to find telemetry data related to crash reports")
+        return [
+            p
+            for t in self.get_telemetry_logs()
+            for p in t["payload"]
+            if "signame" in p.get("tags", "") or "signum" in p.get("tags", "")
+        ]