Retries on esplora connection

Author: alexanderwiederin

.github/workflows/python.yml

@@ -26,9 +26,18 @@ jobs:
 
       - name: Wait for services to be fully initialized
         run: |
-          echo "Waiting for services to be ready..."
-          timeout 90 bash -c 'until docker compose ps | grep -q "bitcoin.*healthy" && docker compose ps | grep -q "electrs.*healthy"; do sleep 5; done'
-          echo "All services are healthy"
+          echo "Waiting for Bitcoin to be ready..."
+          timeout 60 bash -c 'until docker exec ldk-node-bitcoin-1 bitcoin-cli -regtest -rpcuser=user -rpcpassword=pass getblockchaininfo > /dev/null 2>&1; do echo "Waiting for Bitcoin..."; sleep 3; done'
+          echo "Bitcoin is ready"
+
+          echo "Waiting for Esplora to be ready..."
+          # Try to get Esplora's status with retries and a timeout
+          timeout 90 bash -c 'until curl -s http://127.0.0.1:3002/blocks/tip/hash > /dev/null 2>&1; do echo "Waiting for Esplora..."; sleep 3; done'
+          echo "Esplora is ready"
+
+          # This is critical - give Esplora extra time to fully index after it starts accepting connections
+          echo "Giving Esplora extra time to finish indexing..."
+          sleep 10
 
       - name: Install testing prerequisites
         run: |

bindings/python/src/ldk_node/test_ldk_node.py

@@ -51,14 +51,25 @@ def mine_and_wait(esplora_endpoint, blocks):
 def wait_for_block(esplora_endpoint, block_hash):
     url = esplora_endpoint + "/block/" + block_hash + "/status"
     esplora_picked_up_block = False
-    while not esplora_picked_up_block:
-        res = requests.get(url)
+    max_retries = 30
+    retry_count = 0
+
+    while not esplora_picked_up_block and retry_count < max_retries:
         try:
-            json = res.json()
-            esplora_picked_up_block = json['in_best_chain']
-        except:
-            pass
-        time.sleep(1)
+            res = requests.get(url, timeout=10)
+            try:
+                json = res.json()
+                esplora_picked_up_block = json['in_best_chain']
+            except:
+                pass
+        except Exception as e:
+            print(f"Request error: {e}, retrying ({retry_count+1}/{max_retries})...")
+
+        retry_count += 1
+        time.sleep(0.5)
+
+    if not esplora_picked_up_block:
+        raise TimeoutError(f"Block {block_hash} not found after {max_retries} attempts")
 
 def wait_for_tx(esplora_endpoint, txid):
     url = esplora_endpoint + "/tx/" + txid