Enhance error handling and state management in access card and session modules

- Updated the `render_access_panel` function to conditionally hide LLM errors when the Ollama model is operational, improving user experience by reducing unnecessary error visibility.
- Introduced a new `_notes_from_disk` function to streamline the retrieval of notes from disk, enhancing state management during GPU environment checks.
- Refactored error handling in the `_bind_access` function to ensure that stack and GPU environment errors are accurately recorded and managed, improving robustness in session state updates.
- Added tests to validate the new behavior of error handling and state management, ensuring that LLM errors are appropriately suppressed when conditions are met.

tests/test_access_card.py

@@ -42,3 +42,27 @@ def test_access_panel_red_when_killer_failed(monkeypatch):
     )
     panel = render_access_panel(_Cfg(), tunneled=True)
     assert panel.border_style == "red"
+
+
+def test_access_panel_hides_stale_llm_error_when_ollama_ok(monkeypatch):
+    monkeypatch.setattr(
+        "gpu_rent.access_card.load_state",
+        lambda: type(
+            "S",
+            (),
+            {
+                "notes": {
+                    "llm_error": "Ollama /api/tags без моделей",
+                    "stack_vm": [
+                        {
+                            "name": "ollama",
+                            "ok": True,
+                            "detail": "1 models (huihui_ai/qwen2.5-vl-abliterated:7b)",
+                        }
+                    ],
+                }
+            },
+        )(),
+    )
+    panel = render_access_panel(_Cfg(), tunneled=True)
+    assert panel.border_style != "red"

tests/test_session.py

@@ -181,3 +181,51 @@ def test_lock_busy(monkeypatch):
     monkeypatch.setattr("gpu_rent.lock.os.getpid", lambda: 99)
     with pytest.raises(GpuRentError, match="уже работает"):
         SessionLock().__enter__()
+
+
+def test_bind_access_keeps_provision_llm_notes(monkeypatch):
+    """In-memory SessionState must not resurrect llm_error after provision_llm saved."""
+    from gpu_rent.ready import ServiceCheck
+    from gpu_rent.state import load_state
+
+    def fake_provision(cfg, host, log, **kw):
+        st = load_state()
+        st.notes = dict(st.notes or {})
+        st.notes["llm_runtime"] = "ollama"
+        st.notes.pop("llm_error", None)
+        save_state(st)
+
+    monkeypatch.setattr("gpu_rent.session.connect", lambda cfg: object())
+    monkeypatch.setattr("gpu_rent.session.compute_quotas", lambda conn: {"gpu": 1})
+    monkeypatch.setattr("gpu_rent.session.pick_existing_server", lambda conn: Server())
+    _mock_bind(monkeypatch)
+    monkeypatch.setattr("gpu_rent.session.provision_vm", fake_provision)
+    monkeypatch.setattr(
+        "gpu_rent.session.verify_stack_on_vm",
+        lambda cfg, host, log, **kw: [
+            ServiceCheck("swarmui", True, "HTTP 200"),
+            ServiceCheck("ollama", True, "1 models (qwen)"),
+        ],
+    )
+    monkeypatch.setattr(
+        "gpu_rent.session.create_gpu_server",
+        lambda *a, **k: (_ for _ in ()).throw(AssertionError("must not create")),
+    )
+    save_state(
+        SessionState(
+            server_id="s1",
+            floating_ip="203.0.113.9",
+            bootstrapped=True,
+            phase="ready_cloud",
+            notes={
+                "llm_error": "Ollama /api/tags без моделей из ollama-models.yaml",
+                "llm_runtime": "none",
+            },
+        )
+    )
+    state = cmd_up(_cfg(monkeypatch), yes=True)
+    assert state.notes.get("llm_error") is None
+    assert state.notes.get("llm_runtime") == "ollama"
+    disk = load_state()
+    assert disk.notes.get("llm_error") is None
+    assert disk.notes.get("llm_runtime") == "ollama"

tests/test_tune_swarm_perf.py

@@ -259,3 +259,21 @@ def test_jit_fail_strips_sage_extra_args(tmp_path, monkeypatch):
     assert marker["pip_ok"] is True
     assert marker["jit_ok"] is False
     assert "--use-sage-attention" not in backends.read_text(encoding="utf-8")
+
+
+def test_triton_jit_ok_runs_a_py_file(tmp_path, monkeypatch):
+    mod = _load()
+    py = tmp_path / "python"
+    py.write_text("", encoding="utf-8")
+    seen: list[list[str]] = []
+
+    def fake_output(cmd, **kw):
+        seen.append(list(cmd))
+        return "triton JIT ok\n"
+
+    monkeypatch.setattr(mod.subprocess, "check_output", fake_output)
+    assert mod.triton_jit_ok(py) is True
+    assert seen
+    assert seen[0][0] == str(py)
+    assert seen[0][1].endswith(".py")
+    assert "-c" not in seen[0]