metrics tests (#68)

src/node/waku_node.py

@@ -302,11 +302,14 @@ class WakuNode:
         ws_address = next((addr for addr in addresses if "/ws" not in addr), None)
         if ws_address:
             identifier = ws_address.split("/p2p/")[-1]
-            new_address = f"/ip4/{self._ext_ip}/tcp/{self._tcp_port}/p2p/{identifier}"
+            new_address = f"{self.get_tcp_address()}/p2p/{identifier}"
             return new_address
         else:
             raise AttributeError("No '/ws' address found")
 
+    def get_tcp_address(self):
+        return f"/ip4/{self._ext_ip}/tcp/{self._tcp_port}"
+
     def info(self):
         return self._api.info()
 

src/steps/metrics.py

@@ -1,14 +1,17 @@
+import re
 from src.libs.custom_logger import get_custom_logger
 import allure
 from tenacity import retry, stop_after_delay, wait_fixed
 
+from src.test_data import METRICS_WITH_INITIAL_VALUE_ZERO
+
 
 logger = get_custom_logger(__name__)
 
 
 class StepsMetrics:
     @allure.step
-    def check_metric(self, node, metric_name, expected_value):
+    def check_metric(self, node, metric_name, expected_value, exact=False):
         logger.debug(f"Checking metric: {metric_name} has {expected_value}")
         response = node.get_metrics()
         lines = response.split("\n")
@@ -22,7 +25,10 @@ class StepsMetrics:
         if actual_value is None:
             raise AttributeError(f"Metric '{metric_name}' not found")
         logger.debug(f"Found metric: {metric_name} with value {actual_value}")
-        assert actual_value == expected_value, f"Expected value for '{metric_name}' is {expected_value}, but got {actual_value}"
+        if exact:
+            assert actual_value == expected_value, f"Expected value for '{metric_name}' is {expected_value}, but got {actual_value}"
+        else:
+            assert actual_value >= expected_value, f"Expected value for '{metric_name}' is >= {expected_value}, but got {actual_value}"
 
     @allure.step
     def wait_for_metric(self, node, metric_name, expected_value, timeout_duration=90):
@@ -31,3 +37,32 @@ class StepsMetrics:
             self.check_metric(node, metric_name, expected_value)
 
         check_metric_with_retry()
+
+    def validate_initial_metrics(self, node):
+        metrics_data = node.get_metrics()
+
+        # Regular expression to match metric lines, accounting for optional labels
+        metric_pattern = re.compile(r"^(?P<metric_name>[a-zA-Z0-9_:]+(?:{[^}]+})?)\s+(?P<value>[0-9]+\.?[0-9]*)$", re.MULTILINE)
+
+        # Dictionary to store the metrics and their values
+        metrics_dict = {}
+        for match in metric_pattern.finditer(metrics_data):
+            metric_name = match.group("metric_name")
+            value = float(match.group("value"))
+            metrics_dict[metric_name] = value
+
+        errors = []
+        # Assert that specific metrics have a value of 0.0
+        for metric in METRICS_WITH_INITIAL_VALUE_ZERO:
+            if metric not in metrics_dict:
+                errors.append(f"Metric {metric} is missing from the metrics data")
+            elif metrics_dict[metric] != 0.0:
+                errors.append(f"Expected {metric} to be 0.0, but got {metrics_dict[metric]}")
+
+        # Assert that all other metrics have a value greater than 0.0
+        for metric, value in metrics_dict.items():
+            if metric not in METRICS_WITH_INITIAL_VALUE_ZERO and value <= 0.0:
+                errors.append(f"Expected {metric} to have a positive value, but got {value}")
+
+        assert not errors, f"Metrics validation failed:\n" + "\n".join(errors)
+        logger.debug(f"All metrics are present and have valid values.")

src/steps/store.py

@@ -33,9 +33,10 @@ class StepsStore(StepsCommon):
 
     @pytest.fixture(scope="function", autouse=False)
     def node_setup(self, store_setup):
+        logger.debug(f"Running fixture setup: {inspect.currentframe().f_code.co_name}")
         self.setup_first_publishing_node(store="true", relay="true")
         self.setup_first_store_node(store="true", relay="true")
-        self.subscribe_to_pubsub_topics_via_relay()
+        self.subscribe_to_pubsub_topics_via_relay(node=self.main_publishing_nodes)
 
     @allure.step
     def start_publishing_node(self, image, node_index, **kwargs):

src/test_data.py

@@ -177,7 +177,6 @@ LOG_ERROR_KEYWORDS = [
     "abort",
     "segfault",
     "corrupt",
-    "unreachable",
     "terminated",
     "oom",
     "unhandled",
@@ -195,3 +194,183 @@ LOG_ERROR_KEYWORDS = [
     "race condition",
     "double free",
 ]
+
+METRICS_WITH_INITIAL_VALUE_ZERO = [
+    "libp2p_peers",
+    "libp2p_failed_upgrades_incoming_total",
+    "libp2p_failed_upgrades_outgoing_total",
+    "libp2p_total_dial_attempts_total",
+    "libp2p_successful_dials_total",
+    "libp2p_failed_dials_total",
+    "waku_rln_messages_total_total",
+    "waku_rln_spam_messages_total_total",
+    "waku_rln_valid_messages_total_sum",
+    "waku_rln_valid_messages_total_count",
+    'waku_rln_valid_messages_total_bucket{le="1.0"}',
+    'waku_rln_valid_messages_total_bucket{le="2.0"}',
+    'waku_rln_valid_messages_total_bucket{le="3.0"}',
+    'waku_rln_valid_messages_total_bucket{le="4.0"}',
+    'waku_rln_valid_messages_total_bucket{le="5.0"}',
+    'waku_rln_valid_messages_total_bucket{le="+Inf"}',
+    "waku_rln_proof_verification_total_total",
+    "waku_rln_number_registered_memberships",
+    "waku_rln_proof_verification_duration_seconds",
+    "waku_rln_proof_generation_duration_seconds",
+    "waku_rln_instance_creation_duration_seconds",
+    "waku_rln_membership_insertion_duration_seconds",
+    "waku_rln_membership_credentials_import_duration_seconds",
+    "libp2p_pubsub_sig_verify_success_total",
+    "libp2p_pubsub_sig_verify_failure_total",
+    "libp2p_pubsub_disconnects_over_non_priority_queue_limit_total",
+    "libp2p_pubsub_peers",
+    "libp2p_pubsub_topics",
+    "libp2p_pubsub_subscriptions_total",
+    "libp2p_pubsub_unsubscriptions_total",
+    "libp2p_pubsub_validation_success_total",
+    "libp2p_pubsub_validation_failure_total",
+    "libp2p_pubsub_validation_ignore_total",
+    "libp2p_pubsub_broadcast_iwant_total",
+    "libp2p_pubsub_received_iwant_total",
+    "libp2p_gossipsub_cache_window_size",
+    'libp2p_gossipsub_peers_per_topic_mesh{topic="other"}',
+    'libp2p_gossipsub_peers_per_topic_fanout{topic="other"}',
+    'libp2p_gossipsub_peers_per_topic_gossipsub{topic="other"}',
+    "libp2p_gossipsub_under_dout_topics",
+    "libp2p_gossipsub_no_peers_topics",
+    "libp2p_gossipsub_low_peers_topics",
+    "libp2p_gossipsub_healthy_peers_topics",
+    "libp2p_gossipsub_failed_publish_total",
+    "libp2p_gossipsub_invalid_topic_subscription_total",
+    "libp2p_gossipsub_duplicate_during_validation_total",
+    "libp2p_gossipsub_idontwant_saved_messages_total",
+    "libp2p_gossipsub_duplicate_total",
+    "libp2p_gossipsub_received_total",
+    "libp2p_rendezvous_register_total",
+    "libp2p_rendezvous_discover_total",
+    "libp2p_rendezvous_registered",
+    "libp2p_rendezvous_namespaces",
+    "waku_peer_store_size",
+    "waku_total_unique_peers",
+    "presto_server_missing_requests_count",
+    "presto_server_invalid_requests_count",
+    'waku_archive_messages{type="stored"}',
+    "waku_archive_queries",
+    "waku_archive_insert_duration_seconds_sum",
+    "waku_archive_insert_duration_seconds_count",
+    'waku_archive_insert_duration_seconds_bucket{le="0.005"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.01"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.025"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.05"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.075"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.1"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.25"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.5"}',
+    'waku_archive_insert_duration_seconds_bucket{le="0.75"}',
+    'waku_archive_insert_duration_seconds_bucket{le="1.0"}',
+    'waku_archive_insert_duration_seconds_bucket{le="2.5"}',
+    'waku_archive_insert_duration_seconds_bucket{le="5.0"}',
+    'waku_archive_insert_duration_seconds_bucket{le="7.5"}',
+    'waku_archive_insert_duration_seconds_bucket{le="10.0"}',
+    'waku_archive_insert_duration_seconds_bucket{le="+Inf"}',
+    "waku_archive_query_duration_seconds_sum",
+    "waku_archive_query_duration_seconds_count",
+    'waku_archive_query_duration_seconds_bucket{le="0.005"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.01"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.025"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.05"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.075"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.1"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.25"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.5"}',
+    'waku_archive_query_duration_seconds_bucket{le="0.75"}',
+    'waku_archive_query_duration_seconds_bucket{le="1.0"}',
+    'waku_archive_query_duration_seconds_bucket{le="2.5"}',
+    'waku_archive_query_duration_seconds_bucket{le="5.0"}',
+    'waku_archive_query_duration_seconds_bucket{le="7.5"}',
+    'waku_archive_query_duration_seconds_bucket{le="10.0"}',
+    'waku_archive_query_duration_seconds_bucket{le="+Inf"}',
+    "waku_legacy_archive_queries",
+    "waku_legacy_archive_insert_duration_seconds_sum",
+    "waku_legacy_archive_insert_duration_seconds_count",
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.005"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.01"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.025"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.05"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.075"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.1"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.25"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.5"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="0.75"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="1.0"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="2.5"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="5.0"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="7.5"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="10.0"}',
+    'waku_legacy_archive_insert_duration_seconds_bucket{le="+Inf"}',
+    "waku_legacy_archive_query_duration_seconds_sum",
+    "waku_legacy_archive_query_duration_seconds_count",
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.005"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.01"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.025"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.05"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.075"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.1"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.25"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.5"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="0.75"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="1.0"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="2.5"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="5.0"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="7.5"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="10.0"}',
+    'waku_legacy_archive_query_duration_seconds_bucket{le="+Inf"}',
+    "waku_legacy_store_queries",
+    "waku_store_queries",
+    "waku_filter_subscriptions",
+    "waku_filter_handle_message_duration_seconds_sum",
+    "waku_filter_handle_message_duration_seconds_count",
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.005"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.01"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.025"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.05"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.075"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.1"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.25"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.5"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="0.75"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="1.0"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="2.5"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="5.0"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="7.5"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="10.0"}',
+    'waku_filter_handle_message_duration_seconds_bucket{le="+Inf"}',
+    "discovery_session_lru_cache_hits_total",
+    "discovery_session_lru_cache_misses_total",
+    "discovery_session_decrypt_failures_total",
+    "discovery_unsolicited_messages_total",
+    "discovery_enr_auto_update_total",
+    "waku_discv5_discovered",
+    "waku_px_peers_received_total",
+    "waku_px_peers_received_unknown",
+    "waku_px_peers_sent",
+    "waku_px_peers_cached",
+    "waku_histogram_message_size_sum",
+    "waku_histogram_message_size_count",
+    'waku_histogram_message_size_bucket{le="0.0"}',
+    'waku_histogram_message_size_bucket{le="5.0"}',
+    'waku_histogram_message_size_bucket{le="15.0"}',
+    'waku_histogram_message_size_bucket{le="50.0"}',
+    'waku_histogram_message_size_bucket{le="75.0"}',
+    'waku_histogram_message_size_bucket{le="100.0"}',
+    'waku_histogram_message_size_bucket{le="125.0"}',
+    'waku_histogram_message_size_bucket{le="150.0"}',
+    'waku_histogram_message_size_bucket{le="300.0"}',
+    'waku_histogram_message_size_bucket{le="700.0"}',
+    'waku_histogram_message_size_bucket{le="1000.0"}',
+    'waku_histogram_message_size_bucket{le="+Inf"}',
+    "waku_lightpush_peers",
+    "waku_filter_peers",
+    "waku_store_peers",
+    "waku_px_peers",
+    "waku_dnsdisc_discovered",
+]

tests/metrics/test_metrics.py

@@ -0,0 +1,114 @@
+import pytest
+from src.env_vars import DEFAULT_NWAKU
+from src.libs.common import delay
+from src.node.waku_node import WakuNode
+from src.steps.filter import StepsFilter
+from src.steps.light_push import StepsLightPush
+from src.steps.metrics import StepsMetrics
+from src.steps.relay import StepsRelay
+from src.steps.store import StepsStore
+
+
+class TestMetrics(StepsRelay, StepsMetrics, StepsFilter, StepsLightPush, StepsStore):
+    def test_metrics_initial_value(self):
+        node = WakuNode(DEFAULT_NWAKU, f"node1_{self.test_id}")
+        node.start(relay="true", filter="true", store="true", lightpush="true")
+        delay(5)
+        self.validate_initial_metrics(node)
+
+    @pytest.mark.usefixtures("setup_main_relay_nodes", "subscribe_main_relay_nodes", "relay_warm_up")
+    def test_metrics_after_relay_publish(self):
+        self.node1.send_relay_message(self.create_message(), self.test_pubsub_topic)
+        delay(0.5)
+        self.node2.get_relay_messages(self.test_pubsub_topic)
+        delay(5)
+        for node in self.main_nodes:
+            if node.is_nwaku():
+                self.check_metric(node, "libp2p_peers", 1)
+                self.check_metric(node, "libp2p_pubsub_peers", 1)
+                self.check_metric(node, "libp2p_pubsub_topics", 1)
+                self.check_metric(node, "libp2p_pubsub_subscriptions_total", 1)
+                self.check_metric(node, 'libp2p_gossipsub_peers_per_topic_mesh{topic="other"}', 1)
+                self.check_metric(node, "waku_peer_store_size", 1)
+                self.check_metric(node, "waku_histogram_message_size_count", 1)
+                self.check_metric(node, 'waku_node_messages_total{type="relay"}', 1)
+
+    @pytest.mark.usefixtures("setup_main_relay_node", "setup_main_filter_node", "subscribe_main_nodes")
+    def test_metrics_after_filter_get(self):
+        message = self.create_message()
+        self.node1.send_relay_message(message, self.test_pubsub_topic)
+        delay(0.5)
+        self.get_filter_messages(message["contentTopic"], pubsub_topic=self.test_pubsub_topic, node=self.node2)
+        delay(5)
+        self.check_metric(self.node1, "libp2p_peers", 1)
+        self.check_metric(self.node1, "libp2p_pubsub_peers", 1)
+        self.check_metric(self.node1, "libp2p_pubsub_topics", 1)
+        self.check_metric(self.node1, "libp2p_pubsub_subscriptions_total", 1)
+        self.check_metric(self.node1, "waku_peer_store_size", 1)
+        self.check_metric(self.node1, "waku_histogram_message_size_count", 1)
+        self.check_metric(self.node1, 'waku_node_messages_total{type="relay"}', 1)
+        self.check_metric(self.node1, 'waku_filter_requests{type="SUBSCRIBE"}', 1)
+        if self.node2.is_nwaku():
+            self.check_metric(
+                self.node2, f'waku_service_peers{{protocol="/vac/waku/filter-subscribe/2.0.0-beta1",peerId="{self.node1.get_tcp_address()}"}}', 1
+            )
+            self.check_metric(self.node2, "libp2p_peers", 1)
+            self.check_metric(self.node2, "libp2p_total_dial_attempts_total", 1)
+            self.check_metric(self.node2, "waku_peer_store_size", 1)
+
+    def test_metrics_after_light_push(self):
+        self.setup_first_receiving_node()
+        self.setup_second_receiving_node(lightpush="false", relay="true")
+        self.setup_first_lightpush_node()
+        self.subscribe_to_pubsub_topics_via_relay()
+        payload = self.create_payload(self.test_pubsub_topic)
+        self.light_push_node1.send_light_push_message(payload)
+        delay(0.5)
+        self.receiving_node1.get_relay_messages(self.test_pubsub_topic)
+        delay(5)
+        if self.light_push_node1.is_nwaku():
+            self.check_metric(
+                self.light_push_node1,
+                f'waku_service_peers{{protocol="/vac/waku/lightpush/2.0.0-beta1",peerId="{self.receiving_node1.get_tcp_address()}"}}',
+                1,
+            )
+            self.check_metric(self.light_push_node1, "libp2p_peers", 1)
+            self.check_metric(self.light_push_node1, "waku_peer_store_size", 1)
+        if self.receiving_node1.is_nwaku():
+            self.check_metric(self.receiving_node1, "libp2p_peers", 1)
+            self.check_metric(self.receiving_node1, "libp2p_pubsub_peers", 1)
+            self.check_metric(self.receiving_node1, "libp2p_pubsub_topics", 1)
+            self.check_metric(self.receiving_node1, "libp2p_pubsub_subscriptions_total", 1)
+            self.check_metric(self.receiving_node1, "waku_peer_store_size", 1)
+            self.check_metric(self.receiving_node1, "waku_histogram_message_size_count", 1)
+            self.check_metric(self.receiving_node1, 'waku_node_messages_total{type="relay"}', 1)
+
+    def test_metrics_after_store_get(self, node_setup):
+        self.publish_message(message=self.create_message())
+        self.check_published_message_is_stored(page_size=50, ascending="true")
+        delay(5)
+        self.check_metric(self.publishing_node1, "libp2p_peers", 1)
+        self.check_metric(self.publishing_node1, "libp2p_pubsub_peers", 1)
+        self.check_metric(self.publishing_node1, "libp2p_pubsub_topics", 1)
+        self.check_metric(self.publishing_node1, "libp2p_pubsub_subscriptions_total", 1)
+        self.check_metric(self.publishing_node1, "waku_peer_store_size", 1)
+        self.check_metric(self.publishing_node1, "waku_histogram_message_size_count", 1)
+        self.check_metric(self.publishing_node1, 'waku_node_messages_total{type="relay"}', 1)
+        if self.store_node1.is_nwaku():
+            self.check_metric(
+                self.store_node1,
+                f'waku_service_peers{{protocol="/vac/waku/store/2.0.0-beta4",peerId="{self.publishing_node1.get_tcp_address()}"}}',
+                1,
+            )
+            self.check_metric(
+                self.store_node1,
+                f'waku_service_peers{{protocol="/vac/waku/store-query/3.0.0",peerId="{self.publishing_node1.get_tcp_address()}"}}',
+                1,
+            )
+            self.check_metric(self.store_node1, "libp2p_peers", 1)
+            self.check_metric(self.store_node1, "libp2p_pubsub_peers", 1)
+            self.check_metric(self.store_node1, "libp2p_pubsub_topics", 1)
+            self.check_metric(self.store_node1, "libp2p_pubsub_subscriptions_total", 1)
+            self.check_metric(self.store_node1, "waku_peer_store_size", 1)
+            self.check_metric(self.store_node1, "waku_histogram_message_size_count", 1)
+            self.check_metric(self.store_node1, 'waku_node_messages_total{type="relay"}', 1)