fix(k8s): replace blocking time.sleep with asyncio.sleep in _wait_for_sandbox_ready

[qingyuppp]

Problem

In _wait_for_sandbox_ready, the polling loop has two sleep branches with inconsistent behavior:

if not workload:
    time.sleep(poll_interval_seconds)    # blocks the event loop
    continue

# ...

await asyncio.sleep(poll_interval_seconds)  # correct async sleep

When a workload is not yet visible in the K8s API immediately after creation, the code falls into the time.sleep branch. Since _wait_for_sandbox_ready is an async method, this blocks the entire event loop for poll_interval_seconds (default: 1s) per iteration.

Impact

• FastAPI cannot handle any other requests during the blocked period, including /health
• Kubernetes liveness probe times out → pod is killed and restarted
• Affects every sandbox creation under K8s runtime, since the workload-not-found window occurs naturally right after creation

Fix

Replace time.sleep with await asyncio.sleep in the not workload branch, consistent with the sleep call later in the same loop.

File: server/opensandbox_server/services/k8s/kubernetes_service.py:199

# Before
time.sleep(poll_interval_seconds)

# After
await asyncio.sleep(poll_interval_seconds)

[CLAassistant]

All committers have signed the CLA.

[Copilot]

Pull request overview

This PR fixes an event-loop blocking bug in the Kubernetes sandbox readiness polling loop by replacing a synchronous sleep call with an async sleep inside the _wait_for_sandbox_ready coroutine. This prevents FastAPI/ASGI request handling (e.g., /health) from being stalled while waiting for the workload to appear in the Kubernetes API.

Changes:

• Replace time.sleep(...) with await asyncio.sleep(...) when the workload is not yet visible during polling.
• Keep polling behavior consistent across both sleep branches in the loop.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Pangjiping]

LGTM

[qingyuppp]

Thanks for merging.