Checking common sockets before failing

Signed-off-by: Fiona Waters <[email protected]>

Author: Fiona-Waters

README.md

@@ -94,8 +94,6 @@ client = TrainerClient(backend_config=ContainerBackendConfig())
 job_id = client.train(trainer=CustomTrainer(func=train_fn))
 ```
 
-For detailed configuration options and platform-specific setup (macOS, Linux), see the [ContainerBackend documentation](kubeflow/trainer/backends/container/README.md).
-
 ## Supported Kubeflow Projects
 
 | Project                     | Status           | Version Support | Description                                                          |

kubeflow/trainer/backends/container/backend.py

@@ -130,7 +130,8 @@ def _create_adapter(self) -> BaseContainerClientAdapter:
         Create the appropriate container client adapter.
 
         Tries Docker first, then Podman if Docker fails, unless a specific
-        runtime is requested in the config.
+        runtime is requested in the config. Automatically tries common socket
+        locations (e.g., Colima for Docker on macOS, user socket for Podman).
 
         Raises RuntimeError if neither Docker nor Podman are available.
         """
@@ -144,19 +145,15 @@ def _create_adapter(self) -> BaseContainerClientAdapter:
             [self.cfg.container_runtime] if self.cfg.container_runtime else ["docker", "podman"]
         )
 
+        attempted_connections = []
         last_error = None
+
         for runtime_name in runtimes_to_try:
             if runtime_name not in runtime_map:
                 continue
 
-            try:
-                adapter = runtime_map[runtime_name](self.cfg.container_host)
-                adapter.ping()
-                logger.debug(f"Using {runtime_name} as container runtime")
-                return adapter
-            except Exception as e:
-                logger.debug(f"{runtime_name} initialization failed: {e}")
-                last_error = e
+            # Try common socket locations for this runtime
+            socket_locations = self._get_common_socket_locations(runtime_name)
 
             for host in socket_locations:
                 try: