update

workspace/sprint_1_2/CODEBASE/deps/implementation/backend_deploy/gitea_modal_build.py

@@ -32,6 +32,7 @@ DEFAULT_IMAGE_NAME = "msk-lumina-backend"
 DEFAULT_TAG = "latest"
 APP_NAME = "msk-lumina-backend-builder"
 
+
 # =============================================================================
 # Modal Image Definition
 # =============================================================================
@@ -78,6 +79,10 @@ def build_and_push(
     platform: str = "linux/amd64",
 ) -> dict:
     import boto3
+    import base64
+    import time
+    from botocore.exceptions import ClientError
+
     full_image = f"{registry}/{image_name}:{tag}"
     print(f"Building {full_image} for {platform}")
 
@@ -96,11 +101,45 @@ def build_and_push(
         #     ecr = boto3.client("ecr-public", region_name=region)
         # else:
         #     ecr = boto3.client("ecr", region_name=region)
-        ecr = boto3.client("ecr-public", region_name=region)
-        auth = ecr.get_authorization_token()
-        token = auth["authorizationData"]["authorizationToken"]
-        username, password = base64.b64decode(token).decode().split(":")
 
+        
+        # ecr = boto3.client("ecr-public", region_name=region)
+        # auth = ecr.get_authorization_token()
+        # token = auth["authorizationData"]["authorizationToken"]
+
+        def get_ecr_public_credentials(region="us-east-1", max_retries=3, initial_delay=2):
+            ecr = boto3.client("ecr-public", region_name=region)
+            delay = initial_delay
+
+            for attempt in range(max_retries):
+                try:
+                    auth = ecr.get_authorization_token()
+                    token = auth["authorizationData"]["authorizationToken"]
+                    username, password = base64.b64decode(token).decode().split(":")
+                    return username, password
+                    
+                except ClientError as e:
+                    error_code = e.response['Error']['Code']
+                    
+                    # If it's a permanent access issue, don't waste time retrying
+                    if error_code in ["AccessDeniedException", "UnrecognizedClientException"]:
+                        print(f"Permanent permission error: {e}. Fixing config needed.")
+                        raise e
+                        
+                    print(f"Attempt {attempt + 1} failed with transient error ({error_code}). Retrying in {delay}s...")
+                    
+                except Exception as e:
+                    print(f"Attempt {attempt + 1} failed with unexpected error: {e}. Retrying in {delay}s...")
+
+                # Wait before trying again (doubles the wait time each loop)
+                time.sleep(delay)
+                delay *= 2
+
+            # If all retries fail, raise a final clean exception
+            raise RuntimeError(f"Failed to fetch ECR Public token after {max_retries} attempts.")
+       
+        # username, password = base64.b64decode(token).decode().split(":")
+        username, password = get_ecr_public_credentials()
         config = {
             "auths": {
                 registry: {
@@ -138,7 +177,7 @@ def build_and_push(
         "--context", str(workspace_root),
         "--dockerfile", str(dockerfile_path),
         "--destination", full_image,
-        "--custom-platform", platform,
+        "--custom-platform", platform, 
         "--verbosity", "info", 
     ]