feat: pull private images without a hand docker login; surface the pull error

Design merged first: monky-design-docs !225 (doc 24 §4a). Pairs with monky-tenancy!40, which
copies the estate-wide read-only Harbor robot into each environment's own prefix and marks the
manifest entry `use: registry-auth`.

- That entry is not an env var (it would otherwise land in .env and therefore in every container's
  environment). The agent parses it — JSON, or `username:password` with the new `registry_host` —
  and writes `<state_dir>/docker/config.json` 0600 in a directory it owns, with an explicit
  DOCKER_CONFIG pointing the docker CLI at it. The unit runs as monky-deployd, so a `docker login`
  by a human or by root is invisible to the agent: that is what made env-dev-01 look like it had no
  credential at all after the operator had just logged in.
- `compose pull` failures now carry the registry's own message ("no basic auth credentials",
  "manifest unknown", DNS) into the journal and the report instead of `rc=1`.
- Tests: both credential shapes, the refusals, 0600/0700 modes, idempotent rewrite, and that the
  runner never silently falls back to a human's $HOME.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KLB7jieMNRkTsJ2epr4Ds1
This commit is contained in:
2026-09-08 15:45:22 +00:00
parent 35949a9246
commit b25c6b3b8b
9 changed files with 205 additions and 9 deletions
+80
View File
@@ -0,0 +1,80 @@
"""The registry credential (monky-design-docs doc 24 §4a).
The bundle's manifest carries one entry marked `use: registry-auth` — the estate-wide read-only
Harbor robot, copied by monky-tenancy into this environment's own prefix so the agent can read it
with the OpenBao policy it already has. It is NOT an env var (an entry in `.env` would put the
registry password into every container's environment), so it never reaches the compose file: it is
written to a Docker config **the agent owns**, and `DOCKER_CONFIG` points the docker CLI at it.
That last part is the whole point. The unit runs as `monky-deployd`, whose home is the state dir,
so a `docker login` performed by a human or by root is invisible to it — the failure looks exactly
like "no credentials at all" (env-dev-01, 2026-09-08).
Accepted shapes for the secret's value, because the seeded robot has been written both ways:
* a JSON object: `{"registry": …, "username": …, "password": …}`
* a `username:password` string, with the registry taken from `registry_host` config
"""
from __future__ import annotations
import base64
import json
import logging
from dataclasses import dataclass
from pathlib import Path
log = logging.getLogger("monky-deployd.registry")
USE = "registry-auth"
class RegistryAuthError(Exception):
pass
@dataclass
class RegistryAuth:
registry: str
username: str
password: str
def docker_config(self) -> dict:
token = base64.b64encode(f"{self.username}:{self.password}".encode()).decode()
return {"auths": {self.registry: {"auth": token}}}
def parse(value: str, *, default_registry: str) -> RegistryAuth:
"""`value` is whatever the KV entry held; never logged, never echoed."""
text = (value or "").strip()
if not text:
raise RegistryAuthError("empty registry credential")
if text.startswith("{"):
try:
data = json.loads(text)
except json.JSONDecodeError as exc:
raise RegistryAuthError("registry credential is not valid JSON") from exc
user, pw = data.get("username"), data.get("password")
registry = data.get("registry") or default_registry
if not user or not pw:
raise RegistryAuthError("registry credential JSON needs username + password")
return RegistryAuth(registry=str(registry), username=str(user), password=str(pw))
if ":" not in text:
raise RegistryAuthError("registry credential is neither JSON nor username:password")
user, _, pw = text.partition(":")
if not default_registry:
raise RegistryAuthError("username:password credential needs a configured registry host")
return RegistryAuth(registry=default_registry, username=user, password=pw)
def write_docker_config(dir_path: Path, auth: RegistryAuth) -> Path:
"""0600 `config.json` in a directory the agent owns; DOCKER_CONFIG points the CLI at it."""
dir_path = Path(dir_path)
dir_path.mkdir(parents=True, exist_ok=True)
dir_path.chmod(0o700)
target = dir_path / "config.json"
tmp = dir_path / "config.json.tmp"
tmp.write_text(json.dumps(auth.docker_config(), indent=2) + "\n")
tmp.chmod(0o600)
tmp.replace(target)
log.info("registry credential in place for %s (%s)", auth.registry, auth.username)
return target