From 06281de7c8f641e6b0f8caab012bb401e3e19386 Mon Sep 17 00:00:00 2001
From: Roman Zajic <zajic@zajic.net>
Date: Wed, 8 May 2024 20:50:05 +0800
Subject: [PATCH] chore: node readiness (#35)

* test: wait longer for info endpoint to start

* test: health endpoint response parsing

* fix: health endpoint response parsing

* fix: remove debug from health response parsing

* fix: mark on chain tests to skip again

* fix: check_healthy accepts both text and json response

* fix: delete test workflow

* fix: simplify if statements
---
 src/node/api_clients/rest.py |  4 ++--
 src/node/waku_node.py        | 31 ++++++++++++++++++++++++-------
 2 files changed, 26 insertions(+), 9 deletions(-)

diff --git a/src/node/api_clients/rest.py b/src/node/api_clients/rest.py
index daaedd0c64..c42566ec1c 100644
--- a/src/node/api_clients/rest.py
+++ b/src/node/api_clients/rest.py
@@ -25,8 +25,8 @@ class REST(BaseClient):
         return info_response.json()
 
     def health(self):
-        health_response = self.rest_call_text("get", "health")
-        return health_response.text()
+        health_response = self.rest_call("get", "health")
+        return health_response.content
 
     def get_peers(self):
         get_peers_response = self.rest_call("get", "admin/v1/peers")
diff --git a/src/node/waku_node.py b/src/node/waku_node.py
index 24e57bf2a4..44985ae4f8 100644
--- a/src/node/waku_node.py
+++ b/src/node/waku_node.py
@@ -47,8 +47,8 @@ class WakuNode:
         self._container = None
         logger.debug(f"WakuNode instance initialized with log path {self._log_path}")
 
-    @retry(stop=stop_after_delay(5), wait=wait_fixed(0.1), reraise=True)
-    def start(self, wait_for_node_sec=10, **kwargs):
+    @retry(stop=stop_after_delay(60), wait=wait_fixed(0.1), reraise=True)
+    def start(self, wait_for_node_sec=20, **kwargs):
         logger.debug("Starting Node...")
         self._docker_manager.create_network()
         self._ext_ip = self._docker_manager.generate_random_ext_ip()
@@ -188,17 +188,34 @@ class WakuNode:
 
     def ensure_ready(self, timeout_duration=10):
         @retry(stop=stop_after_delay(timeout_duration), wait=wait_fixed(0.1), reraise=True)
+        def check_healthy(node=self):
+            self.health_response = node.health()
+            if self.health_response == b"Node is healthy":
+                logger.info("Node is healthy !!")
+                return
+
+            try:
+                self.health_response = json.loads(self.health_response)
+            except Exception as ex:
+                raise AttributeError(f"Unknown health response format {ex}")
+
+            if self.health_response.get("nodeHealth") != "Ready":
+                raise AssertionError("Waiting for the node health status: Ready")
+
+            for p in self.health_response.get("protocolsHealth"):
+                if p.get("Rln Relay") != "Ready":
+                    raise AssertionError("Waiting for the Rln relay status: Ready")
+
+            logger.info("Node protocols are initialized !!")
+
+        @retry(stop=stop_after_delay(5), wait=wait_fixed(0.1), reraise=True)
         def check_ready(node=self):
             node.info_response = node.info()
             logger.info("REST service is ready !!")
 
+        check_healthy()
         check_ready()
 
-    @retry(stop=stop_after_delay(10), wait=wait_fixed(1), reraise=True)
-    def ensure_healthy(self):
-        self.health_response = self.health()
-        logger.info("Node is healthy !!")
-
     def get_enr_uri(self):
         try:
             return self.info_response["enrUri"]