test(tmux): cover stalled-client cleanup and queued-seat recovery

packages/daemon/test/tmux-exec-deadline.test.ts

@@ -0,0 +1,72 @@
+import { mkdtempSync, writeFileSync, readFileSync, existsSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { it, expect, vi } from "vitest";
+import { execArgvCommand } from "../src/adapters/tmux-exec.js";
+import { shellQuote } from "../src/adapters/shell-quote.js";
+import { SeatDeliveryGuard } from "../src/domain/seat-delivery-guard.js";
+
+// This owns only a fake executable and child in a private home; it never contacts a tmux server.
+it.skipIf(process.platform === "win32")("kills a stalled tmux client at the deadline and releases the next operation for that seat", async () => {
+  const home = mkdtempSync(join(tmpdir(), "openrig-tmux-deadline-"));
+  const pidFile = join(home, "child.json"), calls = join(home, "calls"), script = join(home, "client.cjs");
+  const seat = "deadline-worker@fixture";
+  writeFileSync(script, `const fs = require("node:fs");
+fs.appendFileSync(${JSON.stringify(calls)}, process.argv[2] + "\\n");
+if (process.argv[2] === "capture-pane") {
+  process.on("SIGTERM", () => {});
+  fs.writeFileSync(${JSON.stringify(pidFile)}, JSON.stringify({ pid: process.pid, owner: ${JSON.stringify(home)} }));
+  setInterval(() => {}, 1000);
+  setTimeout(() => process.exit(99), 60000); // Last-resort cleanup if the test process is interrupted.
+} else process.stdout.write("next operation completed\\n");
+`);
+  writeFileSync(join(home, "tmux"), `#!/bin/sh\nexec ${shellQuote(process.execPath)} ${shellQuote(script)} "$@"\n`, { mode: 0o755 });
+  vi.stubEnv("PATH", `${home}:${process.env.PATH ?? ""}`);
+  const warn = vi.spyOn(console, "warn").mockImplementation(() => {});
+  const target = { nodeId: "deadline-node", session: seat, occupant: "g1", pane: "%1" };
+  const db = { prepare: () => ({ get: () => undefined, all: () => [] }) };
+  const guard = new SeatDeliveryGuard(db as never, name => name === seat ? target : null);
+  let pid: number | undefined, settled = false, nextStarted = false;
+  let timer: ReturnType<typeof setTimeout> | undefined;
+  const first = guard.operation(seat, () => execArgvCommand(["tmux", "capture-pane", "-p", "-t", seat]))
+    .then(value => { settled = true; return { kind: "resolved" as const, value }; },
+      error => { settled = true; return { kind: "rejected" as const, error }; });
+  let next: Promise<unknown> | undefined;
+  const ownedPid = () => {
+    const record = JSON.parse(readFileSync(pidFile, "utf8"));
+    expect(record.owner).toBe(home);
+    expect(Number.isSafeInteger(record.pid) && record.pid > 0).toBe(true);
+    return record.pid as number;
+  };
+  try {
+    await expect.poll(() => existsSync(pidFile), { timeout: 10_000 }).toBe(true);
+    pid = ownedPid();
+    next = guard.operation(seat, async () => {
+      nextStarted = true;
+      return execArgvCommand(["tmux", "display-message", "-p", "-t", seat]);
+    });
+    expect(nextStarted).toBe(false);
+    const result = await Promise.race([first, new Promise<{ kind: "not-settled" }>(resolve => {
+      timer = setTimeout(() => resolve({ kind: "not-settled" }), 40_000);
+    })]);
+    expect(result).toMatchObject({ kind: "rejected", error: { killed: true, signal: "SIGKILL" } });
+    expect(() => process.kill(pid!, 0)).toThrow(/ESRCH/);
+    expect(await next).toBe("next operation completed\n");
+    expect(nextStarted).toBe(true);
+    expect(readFileSync(calls, "utf8").trim().split("\n")).toEqual(["capture-pane", "display-message"]);
+    const warnings = JSON.stringify(warn.mock.calls);
+    expect(warnings).toContain("capture-pane");
+    expect(warnings).toContain(seat);
+    expect(warnings).toContain("30000");
+  } finally {
+    if (timer) clearTimeout(timer);
+    if (!settled && existsSync(pidFile)) {
+      try { process.kill(pid ?? ownedPid(), "SIGKILL"); } catch (error) {
+        if ((error as NodeJS.ErrnoException).code !== "ESRCH") throw error;
+      }
+    }
+    await Promise.allSettled([first, ...(next ? [next] : [])]);
+    warn.mockRestore(); vi.unstubAllEnvs();
+    rmSync(home, { recursive: true, force: true });
+  }
+}, 55_000);