Stop long prunes failing with a bare 500 (0.60.1)

A prune that deletes a lot of data (image prune -a in particular) easily
runs past docker-py's 60s read timeout. requests' ReadTimeout is not a
DockerException, so it slipped past safe_call and the UI only showed
"Request failed with status code 500".

Prunes now use a client with a 540s timeout (below nginx's 600s
proxy_read_timeout), and safe_call turns transport timeouts and
connection errors into DockerError, so they surface as a 502 with a
message.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

backend/docker_client.py

@@ -9,6 +9,7 @@ import functools
 from typing import Any
 
 import docker
+import requests
 from docker.errors import APIError, DockerException, NotFound
 
 from config import settings
@@ -23,19 +24,37 @@ class DockerError(Exception):
         super().__init__(f"{error}: {detail}" if detail else error)
 
 
-@functools.lru_cache(maxsize=1)
-def get_client() -> docker.DockerClient:
+# Read timeout for get_long_client(). docker-py's default of 60s is fine for
+# interactive calls, but a prune that deletes gigabytes of volume data or image
+# layers routinely takes longer. Kept below nginx's proxy_read_timeout (600s)
+# so the caller still gets our error instead of a bare gateway timeout.
+LONG_TIMEOUT_SECONDS = 540
+
+
+def _connect(**kwargs) -> docker.DockerClient:
     try:
         base_url = f"unix://{settings.DOCKER_SOCKET}"
-        client = docker.DockerClient(base_url=base_url)
+        client = docker.DockerClient(base_url=base_url, **kwargs)
         client.ping()
         return client
     except DockerException as exc:  # pragma: no cover - environment dependent
         raise DockerError("docker_unavailable", str(exc)) from exc
 
 
+@functools.lru_cache(maxsize=1)
+def get_client() -> docker.DockerClient:
+    return _connect()
+
+
+@functools.lru_cache(maxsize=1)
+def get_long_client() -> docker.DockerClient:
+    """A client for calls that can legitimately run for minutes (prunes)."""
+    return _connect(timeout=LONG_TIMEOUT_SECONDS)
+
+
 def reset_client() -> None:
     get_client.cache_clear()
+    get_long_client.cache_clear()
 
 
 def safe_call(fn, *args, **kwargs) -> Any:
@@ -48,3 +67,12 @@ def safe_call(fn, *args, **kwargs) -> Any:
         raise DockerError("docker_api_error", str(exc)) from exc
     except DockerException as exc:
         raise DockerError("docker_error", str(exc)) from exc
+    # docker-py lets transport errors from requests through unwrapped; without
+    # these a slow daemon surfaced as a bare 500 with no message.
+    except requests.exceptions.Timeout as exc:
+        raise DockerError(
+            "docker_timeout",
+            "Docker did not answer in time; the operation may still be running on the host",
+        ) from exc
+    except requests.exceptions.RequestException as exc:
+        raise DockerError("docker_unavailable", str(exc)) from exc

backend/services/image_service.py

@@ -1,7 +1,7 @@
 """Image listing for the images router."""
 from __future__ import annotations
 
-from docker_client import DockerError, get_client, safe_call
+from docker_client import DockerError, get_client, get_long_client, safe_call
 from services import update_service
 
 COMPOSE_PROJECT_LABEL = "com.docker.compose.project"
@@ -51,7 +51,7 @@ def prune_images(all_unused: bool = False) -> dict:
     """Remove unused images. By default only dangling (untagged) images are
     removed; ``all_unused=True`` removes every image not referenced by a
     container (``docker image prune -a``)."""
-    client = get_client()
+    client = get_long_client()
     # dangling=false tells the engine to also consider tagged-but-unused images.
     filters = {"dangling": False} if all_unused else {"dangling": True}
     result = safe_call(client.images.prune, filters=filters)

backend/services/network_service.py

@@ -6,7 +6,7 @@ from typing import Optional
 from docker.errors import APIError
 from docker.types import IPAMConfig, IPAMPool
 
-from docker_client import DockerError, get_client, safe_call
+from docker_client import DockerError, get_client, get_long_client, safe_call
 
 COMPOSE_PROJECT_LABEL = "com.docker.compose.project"
 DEFAULT_NETWORKS = {"bridge", "host", "none"}
@@ -133,5 +133,5 @@ def delete_network(network_id: str) -> None:
 
 
 def prune_networks() -> dict:
-    client = get_client()
+    client = get_long_client()
     return safe_call(client.networks.prune)

backend/services/volume_service.py

@@ -7,7 +7,7 @@ from typing import Literal, Optional
 
 import yaml
 
-from docker_client import get_client, safe_call
+from docker_client import get_client, get_long_client, safe_call
 
 COMPOSE_PROJECT_LABEL = "com.docker.compose.project"
 
@@ -86,7 +86,7 @@ def remove_volume(name: str, force: bool = False) -> None:
 
 
 def prune_volumes() -> dict:
-    client = get_client()
+    client = get_long_client()
     return safe_call(client.volumes.prune)
 
 

backend/tests/test_docker_timeouts.py

@@ -0,0 +1,81 @@
+"""Slow Docker calls must fail with a message, not a bare 500.
+
+A volume or image prune that deletes a lot of data easily runs past docker-py's
+60s read timeout. requests' ``ReadTimeout`` is not a ``DockerException``, so it
+slipped past ``safe_call`` and the UI only showed "Request failed with status
+code 500". Prunes now use a client with a long timeout, and any transport
+timeout that still happens is turned into a ``DockerError`` (-> 502 + detail).
+"""
+from __future__ import annotations
+
+import pytest
+import requests
+
+
+def test_safe_call_wraps_read_timeout():
+    from docker_client import DockerError, safe_call
+
+    def slow():
+        raise requests.exceptions.ReadTimeout("Read timed out. (read timeout=60)")
+
+    with pytest.raises(DockerError) as info:
+        safe_call(slow)
+    assert info.value.error == "docker_timeout"
+
+
+def test_safe_call_wraps_connection_error():
+    from docker_client import DockerError, safe_call
+
+    def gone():
+        raise requests.exceptions.ConnectionError("socket went away")
+
+    with pytest.raises(DockerError) as info:
+        safe_call(gone)
+    assert info.value.error == "docker_unavailable"
+
+
+def test_volume_prune_timeout_is_a_502_with_detail(as_admin, monkeypatch):
+    from services import volume_service
+
+    class Volumes:
+        def prune(self, filters=None):
+            raise requests.exceptions.ReadTimeout("Read timed out.")
+
+    class Client:
+        volumes = Volumes()
+
+    monkeypatch.setattr(volume_service, "get_long_client", lambda: Client())
+    r = as_admin.post("/api/volumes/prune")
+    assert r.status_code == 502
+    assert r.json()["error"] == "docker_timeout"
+
+
+@pytest.mark.parametrize(
+    "module, fn, args",
+    [
+        ("volume_service", "prune_volumes", ()),
+        ("image_service", "prune_images", (True,)),
+        ("network_service", "prune_networks", ()),
+    ],
+)
+def test_prunes_use_the_long_timeout_client(monkeypatch, module, fn, args):
+    import importlib
+
+    svc = importlib.import_module(f"services.{module}")
+    used = []
+
+    class Any_:
+        def prune(self, filters=None):
+            used.append("long")
+            return {}
+
+    class Client:
+        volumes = images = networks = Any_()
+
+    def short():
+        raise AssertionError("prune must not use the 60s client")
+
+    monkeypatch.setattr(svc, "get_client", short)
+    monkeypatch.setattr(svc, "get_long_client", lambda: Client())
+    getattr(svc, fn)(*args)
+    assert used == ["long"]

backend/version.py

@@ -1,3 +1,3 @@
 """Single source of truth for the StackPilot release version."""
 
-APP_VERSION = "0.60.0"
+APP_VERSION = "0.60.1"

frontend/package.json

@@ -1,7 +1,7 @@
 {
   "name": "stackpilot-frontend",
   "private": true,
-  "version": "0.60.0",
+  "version": "0.60.1",
   "type": "module",
   "scripts": {
     "dev": "vite",