Fix replication recovery test daemon restart race

    repair_env["GKFS_NUM_REPL"] = "1"

    file_path = d00.mountdir / "replicated_failure_recovery"

    ready_marker = Path(test_workspace.twd) / "replica_failure_test_ready"

    repair_process = None

    try:

        result = setup_client.script(

            f"dd if=/dev/zero of='{file_path}' bs=1M count=2 conv=fsync status=none",

        repair_script = f"""

            dd if=/dev/zero of='{file_path}' bs=1M count=2 oflag=append conv=notrunc status=none || true

            : > '{ready_marker}'

            sleep 4

            read -r restart_ready || exit 1

            sleep 3

        """

        # The degraded append can fail; the marker acknowledges its attempt.

        # Wait using a shell builtin, not a preloaded external sleep: removing

        # the dead registration temporarily leaves only one host, so a fresh

        # client with GKFS_NUM_REPL=1 cannot initialize until restart completes.

        repair_process = subprocess.Popen(

            ["bash", "-c", repair_script],

            env=repair_env,

            stdin=subprocess.PIPE,

            stdout=subprocess.PIPE,

            stderr=subprocess.PIPE,

        )

        deadline = time.monotonic() + 15

        while not ready_marker.exists() and time.monotonic() < deadline:

            time.sleep(0.1)

        assert ready_marker.exists(), "repair client did not complete initial write"

        assert ready_marker.exists(), "repair client did not finish initial write attempt"

        _remove_dead_daemon_registration(d01)

        d01.run()

        assert d01._proc.poll() is None

        stdout, stderr = repair_process.communicate(timeout=30)

        stdout, stderr = repair_process.communicate(input=b"ready\n", timeout=30)

        assert repair_process.returncode == 0, stderr.decode(errors="replace")

        result = setup_client.script(