fix: Improve error handling and reliability

4. Pool exhaustion - explicit error
   - Add NoAvailablePoolError exception class
   - find_available_pool() now raises instead of returning None
   - haproxy_add_domain() catches and returns user-friendly error

5. haproxy_add_server - disk-first pattern
   - Save to config FIRST, then update HAProxy
   - If HAProxy update fails, rollback config automatically
   - Prevents inconsistency between disk and runtime

6. Wildcard removal - log failures
   - Changed silent pass to logger.warning()
   - Failures now visible in logs for debugging
   - Does not block domain removal operation

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

mcp/server.py

@@ -90,6 +90,11 @@ class HaproxyError(Exception):
     pass
 
 
+class NoAvailablePoolError(HaproxyError):
+    """All pool backends are in use."""
+    pass
+
+
 # CSV field indices for HAProxy stats (show stat command)
 class StatField:
     """HAProxy CSV stat field indices."""
@@ -358,11 +363,14 @@ def get_map_contents() -> List[Tuple[str, str]]:
     return entries
 
 
-def find_available_pool() -> Optional[str]:
+def find_available_pool() -> str:
     """Find first unused pool from pool_1 to pool_{POOL_COUNT}.
 
     Returns:
-        Pool name (e.g., 'pool_1') if available, None if all pools are used
+        Pool name (e.g., 'pool_1') if available
+
+    Raises:
+        NoAvailablePoolError: If all pools are in use
     """
     used_pools: Set[str] = set()
     for domain, backend in get_map_contents():
@@ -373,7 +381,8 @@ def find_available_pool() -> Optional[str]:
         pool_name = f"pool_{i}"
         if pool_name not in used_pools:
             return pool_name
-    return None
+
+    raise NoAvailablePoolError(f"All {POOL_COUNT} pool backends are in use")
 
 
 def get_domain_backend(domain: str) -> Optional[str]:
@@ -746,9 +755,10 @@ def haproxy_add_domain(domain: str, ip: str = "", http_port: int = 80) -> str:
         return f"Error: Domain {domain} already exists (mapped to {existing_backend})"
 
     # Find available pool
+    try:
         pool = find_available_pool()
-    if not pool:
-        return f"Error: No available pools (all {POOL_COUNT} pools are in use)"
+    except NoAvailablePoolError as e:
+        return f"Error: {e}"
 
     try:
         # Save to disk first (atomic write for persistence)
@@ -816,7 +826,10 @@ def haproxy_remove_domain(domain: str) -> str:
 
         # Clear map entries via Runtime API (immediate effect)
         haproxy_cmd(f"del map {MAP_FILE_CONTAINER} {domain}")
+        try:
             haproxy_cmd(f"del map {MAP_FILE_CONTAINER} .{domain}")
+        except HaproxyError as e:
+            logger.warning("Failed to remove wildcard entry for %s: %s", domain, e)
 
         # Disable all servers in the pool (reset to 0.0.0.0:0)
         for slot in range(1, MAX_SLOTS + 1):
@@ -921,6 +934,10 @@ def haproxy_add_server(domain: str, slot: int, ip: str, http_port: int = 80) ->
     try:
         backend, server_prefix = get_backend_and_prefix(domain)
 
+        # Save to persistent config FIRST (disk-first pattern)
+        add_server_to_config(domain, slot, ip, http_port)
+
+        try:
             results = []
             for suffix, port in get_server_suffixes(http_port):
                 server = f"{server_prefix}{suffix}_{slot}"
@@ -928,11 +945,12 @@ def haproxy_add_server(domain: str, slot: int, ip: str, http_port: int = 80) ->
                 haproxy_cmd_checked(f"set server {backend}/{server} state ready")
                 results.append(f"{server} → {ip}:{port}")
 
-        # Save to persistent config
-        add_server_to_config(domain, slot, ip, http_port)
-
             return f"Added to {domain} ({backend}) slot {slot}:\n" + "\n".join(results)
-    except (HaproxyError, ValueError, IOError) as e:
+        except HaproxyError as e:
+            # Rollback config on HAProxy failure
+            remove_server_from_config(domain, slot)
+            return f"Error: {e}"
+    except (ValueError, IOError) as e:
         return f"Error: {e}"