timeout ad modification for node health to fix flaky tests

src/node/waku_node.py

@@ -5,6 +5,7 @@ import random
 import re
 import shutil
 import string
+import time
 import pytest
 import requests
 from src.libs.common import delay
@@ -295,6 +296,30 @@ class WakuNode:
             check_healthy()
         check_ready()
 
+    def wait_for_connected(self, timeout=30):
+        deadline = time.time() + timeout
+
+        while time.time() < deadline:
+            try:
+                raw = self.health()
+                health = json.loads(raw) if isinstance(raw, (str, bytes)) else raw
+                status = health.get("connectionStatus", "")
+
+                if status == "Connected":
+                    logger.debug("Node reached connectionStatus=Connected")
+                    return
+
+            except Exception as exc:
+                logger.debug("Health poll failed ")
+
+            time.sleep(1)
+
+        raise TimeoutError(f"Node did not reach connectionStatus='Connected' within {timeout}s.")
+
+    def get_node_health(self):
+        raw = self.health()
+        return json.loads(raw) if isinstance(raw, (str, bytes)) else raw
+
     def get_id(self):
         try:
             return self.info_response["listenAddresses"][0].split("/")[-1]
@@ -517,7 +542,6 @@ class WakuNode:
         return self._container
 
     def generate_random_nodekey(self):
-        # Define the set of hexadecimal characters
         hex_chars = string.hexdigits.lower()
         # Generate a random 64-character string from the hex characters
         random_key = "".join(random.choice(hex_chars) for _ in range(64))

src/steps/common.py

@@ -22,12 +22,17 @@ class StepsCommon:
         if not hasattr(self, "test_content_topic"):
             self.test_content_topic = "/test/1/default/proto"
 
+    @allure.step
+    def wait_for_node_connected(self, node, timeout=30):
+        node.wait_for_connected(timeout=timeout)
+
     @allure.step
     @retry(stop=stop_after_delay(20), wait=wait_fixed(0.5), reraise=True)
     def add_node_peer(self, node, multiaddr_list, shards=[0, 1, 2, 3, 4, 5, 6, 7, 8]):
         if node.is_nwaku():
             for multiaddr in multiaddr_list:
                 node.add_peers([multiaddr])
+        self.wait_for_node_connected(node)
 
     @allure.step
     @retry(stop=stop_after_delay(70), wait=wait_fixed(1), reraise=True)

tests/filter/test_subscribe_create.py

@@ -75,9 +75,10 @@ class TestFilterSubscribeCreate(StepsFilter):
         assert not failed_content_topics, f"ContentTopics failed: {failed_content_topics}"
         try:
             self.create_filter_subscription({"requestId": "1", "contentFilters": ["rate limited"], "pubsubTopic": self.test_pubsub_topic})
-            raise AssertionError("The 30th subscribe call was not rate limited!!!")
         except Exception as ex:
             assert "subscription failed" in str(ex) or "rate limit exceeded" in str(ex)
+        else:
+            raise AssertionError("Expected extra filter subscription to be rate limited, but it succeeded")
 
     def test_filter_subscribe_to_101_content_topics(self, subscribe_main_nodes):
         try: