Keep the gateway connection when a tunnel fails to open

Dockerfile

@@ -31,6 +31,10 @@ RUN useradd --system --no-create-home --uid 1001 appuser
 
 USER appuser
 
+# A deployment can run the image as a uid with no passwd entry. getpass,
+# and so asyncssh, reads the login name from LOGNAME first.
+ENV LOGNAME=appuser
+
 EXPOSE 8780
 
 HEALTHCHECK --interval=60s --timeout=5s --start-period=30s --retries=3 \

src/gateway/tests/test_server.py

@@ -12,6 +12,7 @@
 from core.database import async_session_factory
 from gateway import server as gateway_server
 from gateway.settings import GatewaySettings
+from gateway.tests.localprocess import LocalProcess
 from hosts.models import Host, HostStatus
 from providers.base import SandboxProcess, TerminalSize
 from providers.exceptions import ProviderTransportError
@@ -169,8 +170,6 @@ async def answer(reader, writer):
 
 @pytest.fixture
 def local_provider(monkeypatch):
-    from gateway.tests.localprocess import LocalProcess
-
     provider = SimpleNamespace(gateway_process_class=LocalProcess)
     monkeypatch.setattr(gateway_server, "get_vm_provider", lambda name: provider)
     return LocalProcess
@@ -418,7 +417,9 @@ async def test_gateway_shares_one_tunnel_and_closes_it_with_the_caller(
     await asyncio.wait_for(tunnel.wait_closed(), 5)
 
 
-async def test_gateway_reports_a_tunnel_that_fails_to_open(forwarding_caller, sandbox_provider):
+async def test_gateway_fails_only_the_channel_when_the_tunnel_fails_to_open(
+    forwarding_caller, sandbox_provider
+):
     async def refuse(name):
         raise ProviderTransportError("sandboxd is not running")
 
@@ -427,3 +428,4 @@ async def refuse(name):
         await forwarding_caller.open_connection("127.0.0.1", 80)
 
     assert refusal.value.code == OPEN_CONNECT_FAILED
+    assert (await forwarding_caller.run("true", input="\n")).exit_status == 7

src/providers/docker_sbx/provider.py

@@ -156,7 +156,8 @@ async def delete_vm(self, name: str) -> None:
 
     async def open_gateway_tunnel(self, name: str) -> asyncssh.SSHClientConnection:
         # sandboxd authenticates the OS user on its local socket, and sbx itself
-        # trusts the host key on first use. No key crosses a network.
+        # trusts the host key on first use. No key crosses a network. asyncssh
+        # raises ValueError when the local login name is unknown.
         try:
             return await asyncssh.connect(
                 f"{name}.sbx",
@@ -168,7 +169,7 @@ async def open_gateway_tunnel(self, name: str) -> asyncssh.SSHClientConnection:
                 agent_path=None,
                 preferred_auth="none",
             )
-        except (OSError, asyncssh.Error) as exc:
+        except (OSError, ValueError, asyncssh.Error) as exc:
             raise ProviderTransportError(f"sbx could not open a tunnel: {exc}") from exc
 
     async def build_template_image(

src/providers/docker_sbx/tests/test_provider.py

@@ -1,3 +1,4 @@
+import getpass
 from pathlib import Path
 from typing import Any
 from unittest.mock import AsyncMock, MagicMock
@@ -278,3 +279,15 @@ async def test_delete_template_image_translates_a_missing_template(tmp_path):
 
     with pytest.raises(ProviderNotFoundError, match="was not found"):
         await provider.delete_template_image("drukbox-template:missing")
+
+
+@pytest.mark.asyncio
+async def test_open_gateway_tunnel_translates_an_unknown_local_login_name(tmp_path, monkeypatch):
+    def unknown_uid():
+        raise KeyError("getpwuid(): uid not found: 1000")
+
+    monkeypatch.setattr(getpass, "getuser", unknown_uid)
+    provider = _provider(_api_mock(), _settings(tmp_path))
+
+    with pytest.raises(ProviderTransportError):
+        await provider.open_gateway_tunnel("sb-test")