feat(orchestrator): durable launch-broker secret via TrustDomain (#468)
prd-number-check / require-numbered-prds (pull_request) Failing after 12s
tracker-policy-pr / check-pr (pull_request) Successful in 13s
test / integration-docker (pull_request) Successful in 20s
lint / lint (push) Failing after 56s
test / unit (pull_request) Failing after 4m4s
test / coverage (pull_request) Has been skipped

Chunk 2 of the host-control-server stack: close the PRD's **durable
secret** gap and replace chunk 1's BOT_BOTTLE_BROKER_SECRET stopgap.

- trust_domain.py: two new domains. LAUNCH_BROKER holds the durable
  HS256 key both the orchestrator (signer) and the host control server
  (verifier) share for the broker's launch JWT — a host-canonical key
  file minted 0600 on first use, so a restarted orchestrator re-verifies
  against the same key (re-adoption). HOST_CONTROLLER is the separate
  domain for the controller's own lifecycle endpoints, keyed by a key the
  orchestrator never holds (its lifecycle role is `host`, deliberately
  outside the control-plane ROLES). LaunchBrokerProvisioning is the
  fail-closed seam, mirroring ControlPlaneProvisioning.
- orchestrator_auth.py: ROLE_HOST, outside ROLES so the orchestrator's
  control-plane key can neither mint nor accept it.
- paths.py: key-file + env-var constants for both domains.
- host_server.py / __main__.py: broker_secret() now resolves the durable
  LAUNCH_BROKER key — env-injected for a containerized launcher, else the
  host key file for a host-side dev-harness process — so `--broker http`
  and the host controller "just work" on one host without exporting a
  secret, and stay fail-closed when the root is unwritable.

Tested: domain boundary (host role unmintable by the control plane,
cross-domain tokens don't verify), distinct keys/env vars per domain,
provisioning fail-closed, and broker_secret env/file resolution +
durability. Full unit suite green; pyright clean; pylint 9.90.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-07-26 07:34:36 +00:00
parent 79dd4926bb
commit 1a25243505
7 changed files with 253 additions and 51 deletions
+9 -8
View File
@@ -17,9 +17,10 @@ from pathlib import Path
from .. import log
from .store.store_manager import StoreManager
from ..paths import LAUNCH_BROKER_KEY_ENV
from .broker import StubBroker, SubmitBroker
from .broker_client import BrokerClient
from .host_server import BROKER_SECRET_ENV, DEFAULT_PORT, broker_secret_from_env
from .host_server import DEFAULT_PORT, broker_secret
from .server import make_server
from .docker_broker import DockerBroker
from .store.registry_store import RegistryStore, default_db_path
@@ -57,17 +58,17 @@ def main(argv: list[str] | None = None) -> int:
# A signing secret ties the orchestrator (signer) to its broker (verifier).
# 'stub' records launches instead of starting anything; 'docker' runs real
# containers in-process; 'http' relays signed requests to a separate host
# control server, which verifies and launches. For 'stub'/'docker' the
# secret is ephemeral (signer and verifier share this process); for 'http'
# it must be the SAME secret the host controller holds, so it is read from
# the shared env var (the chunk-1 stand-in for out-of-band provisioning).
# control server, which verifies and launches. For 'stub'/'docker' the secret
# is ephemeral (signer and verifier share this process); for 'http' it must be
# the SAME key the host controller holds, so both resolve the durable
# launch-broker TrustDomain key (env-injected, or the host key file).
broker: SubmitBroker
if args.broker == "http":
secret = broker_secret_from_env()
secret = broker_secret()
if secret is None:
parser.error(
f"--broker http requires a shared signing secret in "
f"${BROKER_SECRET_ENV} (hex), matching the host control server"
f"--broker http requires the launch-broker key (${LAUNCH_BROKER_KEY_ENV}, "
"or a writable host root to mint it), matching the host control server"
)
broker = BrokerClient(args.host_controller_url)
else: