Files
DECNET/decnet/agent/heartbeat.py
anti d1ca96b2f4 feat(agent): /deploy and /mutate become 202 fire-and-forget
The wizard API used to hang because /deckies/deploy ran docker compose
build && up -d synchronously, holding the request thread for minutes.
The worker side of that pipeline now returns 202 Accepted immediately
and runs the deploy in an asyncio.create_task.

On task completion (success or failure) the worker pushes a one-off
heartbeat carrying a lifecycle delta per decky:
  {decky_name, operation, status: succeeded|failed, error?, completed_at}

Master pivots these onto open DeckyLifecycle rows in the heartbeat
handler (next commit).  The scheduled 30s heartbeat tick is the
fallback if the immediate push drops.

- decnet/agent/app.py: /deploy and /mutate return 202; dry_run mutate
  still validates synchronously and returns 200.
- decnet/agent/executor.py: deploy_async + mutate_async wrap the work
  and push the completion delta.
- decnet/agent/heartbeat.py: push_lifecycle_delta() helper builds a
  one-off body and POSTs with the same mTLS context as the loop.
- decnet/swarm/client.py: revert deploy/mutate to control timeout
  (master no longer holds the HTTP request open for compose work).

Worker state.json gains no lifecycle field -- master DeckyLifecycle is
the source of truth; the master sweep handles crashed-mid-deploy
recovery.
2026-05-22 16:31:23 -04:00

210 lines
7.0 KiB
Python

"""Agent → master liveness heartbeat loop.
Every ``INTERVAL_S`` seconds the worker posts ``executor.status()`` to
``POST <master>/swarm/heartbeat`` over mTLS. The master pins the
presented client cert's SHA-256 against the ``SwarmHost`` row for the
claimed ``host_uuid``; a match refreshes ``last_heartbeat`` + each
``DeckyShard``'s snapshot + runtime state.
Identity comes from ``/etc/decnet/decnet.ini`` (seeded by the enroll
bundle) — specifically ``DECNET_HOST_UUID`` and ``DECNET_MASTER_HOST``.
The worker's existing ``~/.decnet/agent/`` bundle (or
``/etc/decnet/agent/``) provides the mTLS client cert.
Started/stopped via the agent FastAPI app's lifespan. If identity
plumbing is missing (pre-enrollment dev runs) the loop logs at DEBUG and
declines to start — callers don't have to guard it.
"""
from __future__ import annotations
import asyncio
import pathlib
from typing import Optional
import httpx
from decnet.agent import executor as _exec
from decnet.logging import get_logger
from decnet.swarm import pki
from decnet.swarm.log_forwarder import build_worker_ssl_context
log = get_logger("agent.heartbeat")
INTERVAL_S = 30.0
_TIMEOUT = httpx.Timeout(connect=5.0, read=10.0, write=5.0, pool=5.0)
_task: Optional[asyncio.Task] = None
def _resolve_agent_dir() -> pathlib.Path:
"""Match the agent-dir resolution order used by the agent server:
DECNET_AGENT_DIR env, else /etc/decnet/agent (production install),
else ~/.decnet/agent (dev)."""
import os
env = os.environ.get("DECNET_AGENT_DIR")
if env:
return pathlib.Path(env)
system = pathlib.Path("/etc/decnet/agent")
if system.exists():
return system
return pki.DEFAULT_AGENT_DIR
async def _build_body(
host_uuid: str,
agent_version: str,
lifecycle: Optional[list[dict]] = None,
) -> dict:
snap = await _exec.status()
body: dict = {
"host_uuid": host_uuid,
"agent_version": agent_version,
"status": snap,
}
# Best-effort: fold in applied-topology snapshot. Failures must never
# wedge the heartbeat loop — master will fall back to "no topology
# reported" which triggers a resync if it expected one.
try:
from decnet.agent import topology_ops as _topo_ops
from decnet.agent.topology_store import TopologyStore
store = TopologyStore(_resolve_agent_dir() / "topology.db")
try:
body["topology"] = _topo_ops.state(store)
finally:
store.close()
except Exception:
log.debug("heartbeat: topology state unavailable", exc_info=True)
if lifecycle:
body["lifecycle"] = lifecycle
return body
async def _tick(client: httpx.AsyncClient, url: str, host_uuid: str, agent_version: str) -> None:
body = await _build_body(host_uuid, agent_version)
resp = await client.post(url, json=body)
# 403 / 404 are terminal-ish — we still keep looping because an
# operator may re-enrol the host mid-session, but we log loudly so
# prod ops can spot cert-pinning drift.
if resp.status_code == 204:
return
log.warning(
"heartbeat rejected status=%d body=%s",
resp.status_code, resp.text[:200],
)
async def _loop(url: str, host_uuid: str, agent_version: str, ssl_ctx) -> None:
log.info("heartbeat loop starting url=%s host_uuid=%s interval=%ss",
url, host_uuid, INTERVAL_S)
async with httpx.AsyncClient(verify=ssl_ctx, timeout=_TIMEOUT) as client:
while True:
try:
await _tick(client, url, host_uuid, agent_version)
except asyncio.CancelledError:
raise
except Exception:
log.exception("heartbeat tick failed — will retry in %ss", INTERVAL_S)
await asyncio.sleep(INTERVAL_S)
def start() -> Optional[asyncio.Task]:
"""Kick off the background heartbeat task. No-op if identity is
unconfigured (dev mode) — the caller doesn't need to check."""
global _task
from decnet.env import (
DECNET_HOST_UUID,
DECNET_MASTER_HOST,
DECNET_SWARMCTL_PORT,
)
if _task is not None and not _task.done():
return _task
if not DECNET_HOST_UUID or not DECNET_MASTER_HOST:
log.debug("heartbeat not starting — DECNET_HOST_UUID or DECNET_MASTER_HOST unset")
return None
agent_dir = _resolve_agent_dir()
try:
ssl_ctx = build_worker_ssl_context(agent_dir)
except Exception:
log.exception("heartbeat not starting — worker SSL context unavailable at %s", agent_dir)
return None
try:
from decnet import __version__ as _v # type: ignore[attr-defined]
agent_version = _v
except Exception:
agent_version = "unknown"
url = f"https://{DECNET_MASTER_HOST}:{DECNET_SWARMCTL_PORT}/swarm/heartbeat"
_task = asyncio.create_task(
_loop(url, DECNET_HOST_UUID, agent_version, ssl_ctx),
name="agent-heartbeat",
)
return _task
async def push_lifecycle_delta(deltas: list[dict]) -> None:
"""Fire a one-off heartbeat POST carrying *deltas* in the
``lifecycle`` field. Each delta: ``{decky_name, operation, status,
error?, completed_at?}``.
Called by the agent executor on /deploy and /mutate completion so
the master observes the terminal transition immediately rather than
waiting up to ``INTERVAL_S`` for the next scheduled tick. Failures
are logged and swallowed; the next scheduled heartbeat carries the
same deltas via DB-side reconciliation, since the worker has no
durable per-row state to lose.
"""
from decnet.env import (
DECNET_HOST_UUID,
DECNET_MASTER_HOST,
DECNET_SWARMCTL_PORT,
)
if not deltas:
return
if not DECNET_HOST_UUID or not DECNET_MASTER_HOST:
log.debug("push_lifecycle_delta: identity unconfigured — skipping")
return
agent_dir = _resolve_agent_dir()
try:
ssl_ctx = build_worker_ssl_context(agent_dir)
except Exception:
log.exception("push_lifecycle_delta: SSL context unavailable")
return
try:
from decnet import __version__ as _v # type: ignore[attr-defined]
agent_version = _v
except Exception:
agent_version = "unknown"
url = f"https://{DECNET_MASTER_HOST}:{DECNET_SWARMCTL_PORT}/swarm/heartbeat"
try:
async with httpx.AsyncClient(verify=ssl_ctx, timeout=_TIMEOUT) as client:
body = await _build_body(
DECNET_HOST_UUID, agent_version, lifecycle=deltas,
)
resp = await client.post(url, json=body)
if resp.status_code not in (200, 204):
log.warning(
"lifecycle delta push rejected status=%d body=%s",
resp.status_code, resp.text[:200],
)
except Exception:
log.exception("push_lifecycle_delta failed — next scheduled tick will retry")
async def stop() -> None:
global _task
if _task is None:
return
_task.cancel()
try:
await _task
except (asyncio.CancelledError, Exception):
pass
_task = None