fix: isolate self-hosted e2e graph runs

Author: apstenku123

.github/workflows/e2e-matrix.yml

@@ -55,6 +55,12 @@ jobs:
           echo "VBGUI_E2E_VENV=$job_venv" >> "$GITHUB_ENV"
           echo "PYTHONPATH=" >> "$GITHUB_ENV"
           echo "PYTHONNOUSERSITE=1" >> "$GITHUB_ENV"
+      - name: Allocate isolated E2E ports
+        run: |
+          port_base=$((22000 + (GITHUB_RUN_ID % 1000) * 20))
+          port_offset=${{ matrix.shard }}
+          echo "VBGUI_E2E_BACKEND_PORT=$((port_base + port_offset))" >> "$GITHUB_ENV"
+          echo "VBGUI_E2E_FRONTEND_PORT=$((port_base + 10 + port_offset))" >> "$GITHUB_ENV"
       - name: Python deps
         run: |
           "$VBGUI_E2E_PYTHON" -m pip install --disable-pip-version-check \
@@ -122,6 +128,12 @@ jobs:
           echo "VBGUI_E2E_VENV=$job_venv" >> "$GITHUB_ENV"
           echo "PYTHONPATH=" >> "$GITHUB_ENV"
           echo "PYTHONNOUSERSITE=1" >> "$GITHUB_ENV"
+      - name: Allocate isolated E2E ports
+        run: |
+          port_base=$((22000 + (GITHUB_RUN_ID % 1000) * 20))
+          port_offset=8
+          echo "VBGUI_E2E_BACKEND_PORT=$((port_base + port_offset))" >> "$GITHUB_ENV"
+          echo "VBGUI_E2E_FRONTEND_PORT=$((port_base + 10 + port_offset))" >> "$GITHUB_ENV"
       - run: |
           "$VBGUI_E2E_PYTHON" -m pip install --disable-pip-version-check \
             -e ".[gui,parquet,widget]"
@@ -183,6 +195,12 @@ jobs:
           echo "VBGUI_E2E_VENV=$job_venv" >> "$GITHUB_ENV"
           echo "PYTHONPATH=" >> "$GITHUB_ENV"
           echo "PYTHONNOUSERSITE=1" >> "$GITHUB_ENV"
+      - name: Allocate isolated E2E ports
+        run: |
+          port_base=$((22000 + (GITHUB_RUN_ID % 1000) * 20))
+          port_offset=9
+          echo "VBGUI_E2E_BACKEND_PORT=$((port_base + port_offset))" >> "$GITHUB_ENV"
+          echo "VBGUI_E2E_FRONTEND_PORT=$((port_base + 10 + port_offset))" >> "$GITHUB_ENV"
       - run: |
           "$VBGUI_E2E_PYTHON" -m pip install --disable-pip-version-check \
             -e ".[gui,parquet,widget]"

cppmega_v4/jsonrpc/methods.py

@@ -111,8 +111,18 @@ def _graph_to_specs(graph: GraphSpec) -> list[dict[str, Any]]:
     error; the caller should pre-resolve parallel-block emit on the
     frontend.
     """
+    # ``residual_add`` was emitted by older VBGui canvases as the wire kind
+    # for an explicit residual join.  The model/fusion contract calls that
+    # executable identity brick ``residual``; the fan-in itself is represented
+    # by graph edges.  Accept the legacy wire spelling so saved canvases remain
+    # runnable, but materialise only the canonical kind downstream.
+    kind_aliases = {"residual_add": "residual"}
     return [
-        {"kind": n.kind, "name": n.id, "params": dict(n.params)}
+        {
+            "kind": kind_aliases.get(n.kind, n.kind),
+            "name": n.id,
+            "params": dict(n.params),
+        }
         for n in graph.nodes
     ]
 
@@ -747,4 +757,3 @@ def platform_get_info(
         available_topologies=info["available_topologies"],
         available_comm_backends=info["available_comm_backends"],
     )
-

tests/test_workflow_runner_policy.py

@@ -94,6 +94,17 @@ def test_macos_e2e_uses_an_isolated_job_venv() -> None:
     assert workflow.count("native optimizer extension unavailable") == 3
 
 
+def test_macos_e2e_jobs_use_run_scoped_ports() -> None:
+    workflow = (REPO_ROOT / ".github" / "workflows" / "e2e-matrix.yml").read_text(
+        encoding="utf-8"
+    )
+
+    assert workflow.count("Allocate isolated E2E ports") == 3
+    assert workflow.count("GITHUB_RUN_ID % 1000") == 3
+    assert workflow.count('echo "VBGUI_E2E_BACKEND_PORT=') == 3
+    assert workflow.count('echo "VBGUI_E2E_FRONTEND_PORT=') == 3
+
+
 def test_build_backend_declares_mlx_imported_by_setup_py() -> None:
     config = tomllib.loads((REPO_ROOT / "pyproject.toml").read_text(encoding="utf-8"))
 

tests/v4/test_jsonrpc_methods.py

@@ -12,6 +12,7 @@
 
 from cppmega_v4.jsonrpc import LRUCache
 from cppmega_v4.jsonrpc.methods import (
+    _graph_to_specs,
     _make_optim,
     build_preset_specs,
     probe_run,
@@ -50,6 +51,22 @@ def _simple_verify_params(**extra) -> VerifyParams:
     return VerifyParams.model_validate(payload)
 
 
+def test_graph_to_specs_canonicalizes_legacy_residual_add() -> None:
+    params = _simple_verify_params(
+        graph={
+            "nodes": [
+                {"id": "branch", "kind": "mlp"},
+                {"id": "join", "kind": "residual_add"},
+            ],
+            "edges": [{"src": "branch", "dst": "join"}],
+        }
+    )
+
+    specs = _graph_to_specs(params.graph)
+
+    assert [spec["kind"] for spec in specs] == ["mlp", "residual"]
+
+
 def test_make_optim_threads_mixed_precision_flag():
     optim = _make_optim(OptimSpecPayload(
         kind="adamw",

vbgui/e2e/globalSetup.ts

@@ -43,13 +43,32 @@ function spawnLogged(cmd: string, args: string[],
   return child;
 }
 
-async function waitFor(url: string, label: string, timeoutMs: number) {
+function assertRunning(child: ChildProcess, label: string): void {
+  if (child.exitCode !== null || child.signalCode !== null || child.pid == null) {
+    throw new Error(
+      `${label} exited before readiness ` +
+      `(code=${child.exitCode}, signal=${child.signalCode})`,
+    );
+  }
+  try {
+    process.kill(child.pid, 0);
+  } catch {
+    throw new Error(`${label} process ${child.pid} is not running`);
+  }
+}
+
+async function waitFor(url: string, label: string, timeoutMs: number,
+                       child: ChildProcess) {
   const deadline = Date.now() + timeoutMs;
   let lastErr: string = "no attempts";
   while (Date.now() < deadline) {
+    assertRunning(child, label);
     try {
       const res = await fetch(url);
-      if (res.ok) return;
+      if (res.ok) {
+        assertRunning(child, label);
+        return;
+      }
       lastErr = `HTTP ${res.status}`;
     } catch (e) {
       lastErr = String(e);
@@ -99,7 +118,7 @@ export default async function globalSetup(): Promise<void> {
   }, null, 2));
 
   await Promise.all([
-    waitFor(`${BACKEND_URL}/health`, "backend", 30_000),
-    waitFor(FRONTEND_URL, "frontend", 30_000),
+    waitFor(`${BACKEND_URL}/health`, "backend", 30_000, handles.backend),
+    waitFor(FRONTEND_URL, "frontend", 30_000, handles.frontend),
   ]);
 }

vbgui/src/App.tsx

@@ -580,7 +580,7 @@ function presetSpecsToNodes(specs: BrickSpec[], canvasWidth?: number): { nodes:
         id: addNodeName,
         type: "residual_add",
         position: { x: 0, y: 0 },
-        data: { kind: "residual_add", params: {} } as any,
+        data: { kind: "residual", params: {} } as any,
       });
 
       branchNames.forEach((name) => {
@@ -618,7 +618,7 @@ function presetSpecsToNodes(specs: BrickSpec[], canvasWidth?: number): { nodes:
           id: addNodeName,
           type: "residual_add",
           position: { x: 0, y: 0 },
-          data: { kind: "residual_add", params: {} } as any,
+          data: { kind: "residual", params: {} } as any,
         });
 
         edges.push({
@@ -1658,7 +1658,7 @@ export function App(): JSX.Element {
             id: addNodeName,
             type: "residual_add",
             position: { x, y },
-            data: { kind: "residual_add", params: {} } as any,
+            data: { kind: "residual", params: {} } as any,
           });
 
           newEdges.push({