Spaces:
Paused
Paused
| """Phase 4/5: boot a real Linux VM under QEMU-TCG inside the HF Space container, | |
| then SSH from the container into the VM. Progress -> stdout (hf spaces logs).""" | |
| import os, subprocess, sys, threading, time, socket, textwrap, shutil | |
| T0 = time.time() | |
| LOG = [] | |
| IMG_URL = ("https://dl-cdn.alpinelinux.org/alpine/v3.22/releases/cloud/" | |
| "nocloud_alpine-3.22.4-x86_64-bios-cloudinit-r0.qcow2") | |
| WORK = "/tmp/vm" | |
| QCOW = WORK + "/alpine.qcow2" | |
| SEED = WORK + "/seed.iso" | |
| KEY = WORK + "/id_ed25519" | |
| SERIAL = WORK + "/serial.log" | |
| def log(msg): | |
| line = "[t+%6.1fs] %s" % (time.time() - T0, msg) | |
| LOG.append(line) | |
| print(line, flush=True) | |
| def sh(cmd, timeout=1800, quiet=False): | |
| log("$ " + cmd) | |
| r = subprocess.run(["bash", "-lc", cmd], capture_output=True, text=True, timeout=timeout) | |
| t = (r.stdout + r.stderr).strip() | |
| if t and not quiet: | |
| tail = "\n".join(t.splitlines()[-25:]) | |
| LOG.append(tail); print(tail, flush=True) | |
| if r.returncode: | |
| log("[exit %d]" % r.returncode) | |
| return r.returncode, t | |
| def port_open(port, host="127.0.0.1"): | |
| s = socket.socket(); s.settimeout(2) | |
| try: | |
| return s.connect_ex((host, port)) == 0 | |
| finally: | |
| s.close() | |
| def run(): | |
| os.makedirs(WORK, exist_ok=True) | |
| log("### PHASE 4a: install QEMU ###") | |
| sh("apt-get update -qq", timeout=900, quiet=True) | |
| rc, _ = sh("DEBIAN_FRONTEND=noninteractive apt-get install -y -qq --no-install-recommends " | |
| "qemu-system-x86 qemu-utils cloud-image-utils openssh-client", timeout=1800, quiet=True) | |
| if rc: | |
| log("!!! apt install FAILED — Phase 4 BLOCKED"); return | |
| sh("qemu-system-x86_64 --version | head -2") | |
| sh("qemu-system-x86_64 -accel help") | |
| log("### PHASE 4b: confirm KVM accel is refused, TCG is available ###") | |
| sh("qemu-system-x86_64 -accel kvm -m 128 -display none -no-reboot -kernel /dev/null 2>&1 | head -5 || true") | |
| log("### PHASE 4c: fetch Alpine nocloud image ###") | |
| if not os.path.exists(QCOW): | |
| rc, _ = sh("curl -fsSL -o %s '%s'" % (QCOW, IMG_URL), timeout=1800, quiet=True) | |
| if rc: | |
| log("!!! image download FAILED"); return | |
| sh("ls -lh %s && qemu-img info %s" % (QCOW, QCOW)) | |
| sh("qemu-img resize %s 4G" % QCOW) | |
| log("### PHASE 5a: cloud-init seed (ssh key injection) ###") | |
| sh("rm -f %s %s.pub && ssh-keygen -t ed25519 -N '' -f %s -q" % (KEY, KEY, KEY)) | |
| pub = open(KEY + ".pub").read().strip() | |
| open(WORK + "/user-data", "w").write(textwrap.dedent("""\ | |
| #cloud-config | |
| disable_root: false | |
| ssh_pwauth: false | |
| users: | |
| - default | |
| - name: root | |
| ssh_authorized_keys: | |
| - %s | |
| ssh_authorized_keys: | |
| - %s | |
| runcmd: | |
| - [ sh, -c, "echo '@@@ CLOUDINIT_DONE @@@' > /dev/console" ] | |
| """) % (pub, pub)) | |
| open(WORK + "/meta-data", "w").write("instance-id: hfvm-001\nlocal-hostname: hfvm\n") | |
| sh("cloud-localds %s %s/user-data %s/meta-data && ls -lh %s" % (SEED, WORK, WORK, SEED)) | |
| log("### PHASE 4d: BOOT VM (QEMU TCG, 2 vCPU / 1536 MB) ###") | |
| qemu = ("qemu-system-x86_64 -accel tcg,thread=multi -smp 2 -m 1536 " | |
| "-drive file=%s,if=virtio,format=qcow2 " | |
| "-drive file=%s,if=virtio,format=raw,readonly=on " | |
| "-netdev user,id=n0,hostfwd=tcp:127.0.0.1:2222-:22 " | |
| "-device virtio-net-pci,netdev=n0 " | |
| "-nographic -no-reboot" % (QCOW, SEED)) | |
| log("$ " + qemu) | |
| sf = open(SERIAL, "wb") | |
| proc = subprocess.Popen(["bash", "-lc", qemu], stdin=subprocess.DEVNULL, | |
| stdout=sf, stderr=subprocess.STDOUT) | |
| boot_t0 = time.time() | |
| log("### PHASE 5b: wait for guest sshd on 127.0.0.1:2222 ###") | |
| up = False | |
| for i in range(120): # up to 20 min | |
| if proc.poll() is not None: | |
| log("!!! QEMU exited early rc=%s" % proc.returncode); break | |
| if port_open(2222): | |
| up = True | |
| log(">>> GUEST SSH PORT OPEN after %.1fs of TCG boot" % (time.time() - boot_t0)) | |
| break | |
| if i % 6 == 0: | |
| try: | |
| tail = subprocess.run(["tail", "-c", "1200", SERIAL], | |
| capture_output=True, text=True).stdout | |
| log("serial tail: " + tail.replace("\r", "")[-700:]) | |
| except Exception: | |
| pass | |
| time.sleep(10) | |
| if not up: | |
| log("!!! guest sshd never came up") | |
| else: | |
| log("### PHASE 5c: SSH container -> VM ###") | |
| ssh = ("ssh -i %s -p 2222 -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null " | |
| "-o ConnectTimeout=15 -o BatchMode=yes" % KEY) | |
| for user in ("root", "alpine"): | |
| rc, out = sh("%s %s@127.0.0.1 'echo ===IN-VM===; uname -a; id; " | |
| "cat /etc/os-release | head -2; free -m | head -2; " | |
| "df -h / | tail -1; ip -4 addr show eth0 | grep inet' 2>&1" % (ssh, user), | |
| timeout=120) | |
| if rc == 0: | |
| log(">>> PHASE 5 PASS: SSH into VM succeeded as %s" % user) | |
| break | |
| else: | |
| log("!!! SSH into VM failed for all users") | |
| log("### benchmark: TCG throughput inside VM ###") | |
| sh("%s root@127.0.0.1 'time dd if=/dev/zero bs=1M count=64 2>&1 | tail -3' 2>&1" % ssh, timeout=180) | |
| log("### serial console (last 4000 bytes) ###") | |
| sh("tail -c 4000 %s | tr -d '\\r'" % SERIAL) | |
| log("### DONE ###") | |
| try: | |
| proc.terminate() | |
| except Exception: | |
| pass | |
| threading.Thread(target=lambda: (run() if True else None), daemon=True).start() | |
| import gradio as gr | |
| gr.Interface(lambda: "\n".join(LOG[-400:]), None, | |
| gr.Textbox(lines=40, label="phase 4/5 log"), | |
| title="QEMU-in-HF-Space experiment").launch(server_name="0.0.0.0", server_port=7860) | |