Files
Unraid-AppOps-Gateway-Public/backend/app/services/privatemuse_deployment_enrollment.py
T

547 lines
21 KiB
Python

from __future__ import annotations
import json
import time
from typing import Any
from urllib.parse import quote
from ..connectors.portainer import PortainerConnector
from ..core.redaction import redact_message
_CONTROLLER = "gitea-deploy-control"
_REPOSITORY = "jens/privatemuse"
_LEGACY_REVISION = "715f7d152d35745521a2e014e334a1509bdce6a2"
_PROJECT: dict[str, object] = {
"branch": "main",
"slug": "privatemuse",
"source_dir": "/mnt/cache/appdata/gitea-deploy-runners/sources/privatemuse",
"exact_source": True,
"env_source": "/srv/example/appdata/privatemuse/.env",
"compose_files": ["docker-compose.yml", "docker-compose.unraid.yml"],
"compose_project": "privatemuse",
"containers": [
"privatemuse-backend-1",
"privatemuse-frontend-1",
"privatemuse-audio-enhancement-1",
],
"verify_oci_revision": False,
"health_timeout": 180,
"environment": {"PRIVATEMUSE_RELEASE": "{short_revision}"},
}
# This is a fixed, application-owned one-time migration. Callers cannot supply
# commands, paths, container names, Compose files or deployment JSON. It uses
# the same controller-owned registry, global production lock and official
# deploy.py bootstrap-legacy path as the deployment broker itself.
_ENROLL_SCRIPT = r'''
import fcntl
import importlib.util
import json
import os
from pathlib import Path
import shutil
import stat
import subprocess
import sys
import time
registry_path = Path("/controller-config/projects.json")
host_registry_path = Path("/controller-hub-releases/controller-config/projects.json")
state_root = Path("/mnt/cache/appdata/gitea-deploy-runners/state")
state_path = state_root / "privatemuse.json"
project_lock_path = state_root / "locks" / "privatemuse.lock"
lock_path = Path("/mnt/cache/appdata/gitea-deploy-runners/deploy.lock")
repository, legacy_revision, raw_project = sys.argv[1:4]
project = json.loads(raw_project)
expected_project = {
"branch": "main",
"slug": "privatemuse",
"source_dir": "/mnt/cache/appdata/gitea-deploy-runners/sources/privatemuse",
"exact_source": True,
"env_source": "/srv/example/appdata/privatemuse/.env",
"compose_files": ["docker-compose.yml", "docker-compose.unraid.yml"],
"compose_project": "privatemuse",
"containers": [
"privatemuse-backend-1",
"privatemuse-frontend-1",
"privatemuse-audio-enhancement-1",
],
"verify_oci_revision": False,
"health_timeout": 180,
"environment": {"PRIVATEMUSE_RELEASE": "{short_revision}"},
}
expected_legacy = "715f7d152d35745521a2e014e334a1509bdce6a2"
stale_audio = {
"privatemuse-audio-runtime-base-1",
"privatemuse-audio-speech-1",
"privatemuse-audio-music-1",
"privatemuse-audio-sound-1",
"privatemuse-audio-transcription-1",
"privatemuse-audio-lipsync-1",
}
if repository != "jens/privatemuse" or legacy_revision != expected_legacy or project != expected_project:
raise SystemExit(40)
if registry_path.is_symlink() or not registry_path.is_file():
raise SystemExit(20)
registry_stat = registry_path.stat()
if not stat.S_ISREG(registry_stat.st_mode):
raise SystemExit(21)
try:
if (
host_registry_path.is_symlink()
or not host_registry_path.is_file()
or not os.path.samefile(registry_path, host_registry_path)
):
raise SystemExit(36)
except OSError:
raise SystemExit(36)
env_source = Path(str(project["env_source"]))
if env_source.is_symlink() or not env_source.is_file():
raise SystemExit(41)
owner, name = repository.split("/", 1)
bare_repo = Path("/gitea-repositories") / owner / f"{name}.git"
if not bare_repo.is_dir():
raise SystemExit(24)
def git_run(*args):
return subprocess.run(
["git", f"--git-dir={bare_repo}", *args],
text=True,
stdout=subprocess.PIPE,
stderr=subprocess.DEVNULL,
timeout=10,
check=False,
)
legacy = git_run("rev-parse", f"{legacy_revision}^{{commit}}")
if legacy.returncode != 0 or legacy.stdout.strip() != legacy_revision:
raise SystemExit(25)
main = git_run("rev-parse", "refs/heads/main")
main_revision = main.stdout.strip()
if main.returncode != 0 or len(main_revision) != 40:
raise SystemExit(26)
if git_run("merge-base", "--is-ancestor", legacy_revision, main_revision).returncode != 0:
raise SystemExit(27)
for compose_file in project["compose_files"]:
if git_run("cat-file", "-e", f"{main_revision}:{compose_file}").returncode != 0:
raise SystemExit(28)
try:
if state_root.is_symlink() or not state_root.is_dir():
raise SystemExit(50)
project_lock_path.parent.mkdir(parents=True, exist_ok=True)
if project_lock_path.parent.is_symlink() or project_lock_path.is_symlink():
raise SystemExit(29)
project_lock_handle = project_lock_path.open("a+", encoding="utf-8")
lock_handle = lock_path.open("a+", encoding="utf-8")
except Exception:
raise SystemExit(29)
with project_lock_handle as project_lock, lock_handle as lock:
try:
fcntl.flock(project_lock, fcntl.LOCK_EX)
fcntl.flock(lock, fcntl.LOCK_EX)
except Exception:
raise SystemExit(30)
try:
projects = json.loads(host_registry_path.read_text(encoding="utf-8"))
except Exception:
raise SystemExit(31)
if not isinstance(projects, dict):
raise SystemExit(22)
current = projects.get(repository)
canonical_target_match = current == project
if current is not None and current != project:
if not isinstance(current, dict):
raise SystemExit(54)
if current.get("appops_managed_target") is True:
raise SystemExit(55)
target_fields = [
"branch",
"slug",
"source_dir",
"exact_source",
"env_source",
"compose_files",
"compose_project",
"containers",
"verify_oci_revision",
"health_timeout",
"environment",
]
for index, key in enumerate(target_fields, start=60):
if key not in current or current.get(key) != project[key]:
raise SystemExit(index)
extra_keys = set(current) - set(project)
allowed_metadata_keys = {
"gitea_repository_id",
"gitea_webhook_secret_file",
"validation",
}
expected_validation = {
"validation_gate_version": 1,
"required": True,
"contexts": ["Managed validation / full (push)"],
"timeout_seconds": 1800,
"poll_interval_seconds": 5,
"max_poll_interval_seconds": 30,
}
repository_id = current.get("gitea_repository_id")
if (
extra_keys != allowed_metadata_keys
or not isinstance(repository_id, int)
or isinstance(repository_id, bool)
or repository_id <= 0
or current.get("gitea_webhook_secret_file") != "jens__privatemuse"
or current.get("validation") != expected_validation
):
raise SystemExit(71)
canonical_target_match = True
existing_state = None
if state_path.exists():
if state_path.is_symlink() or not state_path.is_file():
raise SystemExit(42)
try:
existing_state = json.loads(state_path.read_text(encoding="utf-8"))
except Exception:
raise SystemExit(42)
if not isinstance(existing_state, dict):
raise SystemExit(42)
if existing_state.get("schema_version") == 3:
if existing_state.get("repository") != repository:
raise SystemExit(43)
if not canonical_target_match:
raise SystemExit(23)
print("already-enrolled", flush=True)
raise SystemExit(0)
if not (
existing_state.get("repository") == repository
and existing_state.get("status") == "success"
and existing_state.get("revision") == legacy_revision
and "schema_version" not in existing_state
and "live_receipt" not in existing_state
and "live_legacy_provenance" not in existing_state
):
raise SystemExit(43)
expected_names = set(project["containers"])
observed_result = subprocess.run(
[
"docker",
"ps",
"--all",
"--filter",
f"label=com.docker.compose.project={project['compose_project']}",
"--format",
"{{.Names}}",
],
text=True,
stdout=subprocess.PIPE,
stderr=subprocess.DEVNULL,
timeout=15,
check=False,
)
if observed_result.returncode != 0:
raise SystemExit(44)
observed_names = {line.strip() for line in observed_result.stdout.splitlines() if line.strip()}
unexpected = observed_names - expected_names - stale_audio
if unexpected:
raise SystemExit(45)
for required in sorted(expected_names):
inspected = subprocess.run(
[
"docker",
"inspect",
"--format",
"{{.State.Running}}|{{if .State.Health}}{{.State.Health.Status}}{{else}}none{{end}}|{{index .Config.Labels \"com.docker.compose.project\"}}",
required,
],
text=True,
stdout=subprocess.PIPE,
stderr=subprocess.DEVNULL,
timeout=10,
check=False,
)
if inspected.returncode != 0:
raise SystemExit(46)
try:
running, health, compose_project = inspected.stdout.strip().split("|", 2)
except ValueError:
raise SystemExit(46)
if running != "true" or health in {"starting", "unhealthy"} or compose_project != "privatemuse":
raise SystemExit(46)
# Remove only known, stopped provider containers. Images, named volumes,
# model directories and persistent media are deliberately left untouched.
for stale in sorted(observed_names & stale_audio):
inspected = subprocess.run(
[
"docker",
"inspect",
"--format",
"{{.State.Running}}|{{index .Config.Labels \"com.docker.compose.project\"}}",
stale,
],
text=True,
stdout=subprocess.PIPE,
stderr=subprocess.DEVNULL,
timeout=10,
check=False,
)
if inspected.returncode != 0:
raise SystemExit(47)
try:
running, compose_project = inspected.stdout.strip().split("|", 1)
except ValueError:
raise SystemExit(47)
if running == "true" or compose_project != "privatemuse":
raise SystemExit(48)
removed = subprocess.run(
["docker", "rm", stale],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
timeout=20,
check=False,
)
if removed.returncode != 0:
raise SystemExit(49)
if current is None:
backup = host_registry_path.with_name(
f"projects.json.privatemuse-backup-{int(time.time())}-{os.getpid()}"
)
try:
shutil.copy2(host_registry_path, backup)
except Exception:
raise SystemExit(32)
updated = dict(projects)
updated[repository] = project
temporary = host_registry_path.with_name(f".{host_registry_path.name}.{os.getpid()}.tmp")
payload = json.dumps(updated, indent=2, sort_keys=True) + "\n"
try:
with temporary.open("w", encoding="utf-8") as handle:
handle.write(payload)
handle.flush()
os.fsync(handle.fileno())
os.chmod(temporary, stat.S_IMODE(registry_stat.st_mode))
os.replace(temporary, host_registry_path)
directory_fd = os.open(str(host_registry_path.parent), os.O_RDONLY)
try:
os.fsync(directory_fd)
finally:
os.close(directory_fd)
except Exception:
try:
temporary.unlink(missing_ok=True)
except Exception:
pass
raise SystemExit(34)
if existing_state is None:
state_root.mkdir(parents=True, exist_ok=True)
if state_root.is_symlink() or not state_root.is_dir():
raise SystemExit(50)
temporary_state = state_path.with_name(f".{state_path.name}.{os.getpid()}.tmp")
payload = {
"repository": repository,
"revision": legacy_revision,
"status": "success",
}
try:
with temporary_state.open("x", encoding="utf-8") as handle:
handle.write(json.dumps(payload, indent=2, sort_keys=True) + "\n")
handle.flush()
os.fsync(handle.fileno())
os.chmod(temporary_state, 0o600)
os.replace(temporary_state, state_path)
directory_fd = os.open(str(state_root), os.O_RDONLY)
try:
os.fsync(directory_fd)
finally:
os.close(directory_fd)
except Exception:
try:
temporary_state.unlink(missing_ok=True)
except Exception:
pass
raise SystemExit(51)
try:
controller_path = Path("/opt/gitea-deploy/deploy.py")
if controller_path.is_symlink() or not controller_path.is_file():
raise RuntimeError("controller module is unavailable")
controller_spec = importlib.util.spec_from_file_location(
"appops_locked_privatemuse_bootstrap", controller_path
)
if controller_spec is None or controller_spec.loader is None:
raise RuntimeError("controller module cannot be loaded")
controller = importlib.util.module_from_spec(controller_spec)
controller_spec.loader.exec_module(controller)
controller.bootstrap_legacy_state(repository, project, legacy_revision)
except Exception:
raise SystemExit(52)
try:
final_state = json.loads(state_path.read_text(encoding="utf-8"))
except Exception:
raise SystemExit(53)
if not (
isinstance(final_state, dict)
and final_state.get("schema_version") == 3
and final_state.get("repository") == repository
and final_state.get("live_revision") == legacy_revision
and final_state.get("live_status") == "healthy"
and final_state.get("live_legacy_provenance") is True
):
raise SystemExit(53)
print("enrolled", flush=True)
'''
class PrivateMuseDeploymentEnrollment:
"""One-time, fail-closed enrollment of the existing PrivateMuse production runtime."""
def __init__(self, portainer: PortainerConnector | None = None):
self.portainer = portainer or PortainerConnector()
def run(self) -> dict[str, Any]:
status = self.portainer.get_container_status(_CONTROLLER)
if status.get("status") != "ok":
return {
"status": "controller_unavailable",
"reason": redact_message(
str(status.get("reason") or status.get("status") or "unavailable")
),
}
controller = dict(status.get("container") or {})
if controller.get("name") != _CONTROLLER or controller.get("state") != "running":
return {
"status": "controller_unavailable",
"reason": "deployment controller is not running",
}
container_id = str(controller.get("id") or "")
config = self.portainer._config()
if not config or not container_id:
return {
"status": "connector_unavailable",
"reason": "Portainer deployment control is unavailable",
}
command = [
"python3",
"-c",
_ENROLL_SCRIPT,
_REPOSITORY,
_LEGACY_REVISION,
json.dumps(_PROJECT, separators=(",", ":"), sort_keys=True),
]
endpoint = quote(str(config.endpoint_id), safe="")
safe_container = quote(container_id, safe="")
try:
created = self.portainer._request(
"POST",
f"/api/endpoints/{endpoint}/docker/containers/{safe_container}/exec",
json={
"AttachStdout": False,
"AttachStderr": False,
"Tty": False,
"Cmd": command,
},
).json()
exec_id = str(created.get("Id") or "")
if not exec_id:
return {"status": "error", "reason": "PrivateMuse enrollment could not start"}
safe_exec = quote(exec_id, safe="")
self.portainer._request(
"POST",
f"/api/endpoints/{endpoint}/docker/exec/{safe_exec}/start",
json={"Detach": True, "Tty": False},
)
deadline = time.monotonic() + 120
exit_code: int | None = None
while time.monotonic() < deadline:
inspected = self.portainer._request(
"GET", f"/api/endpoints/{endpoint}/docker/exec/{safe_exec}/json"
).json()
if not bool(inspected.get("Running")):
raw_exit = inspected.get("ExitCode")
exit_code = int(raw_exit) if raw_exit is not None else -1
break
time.sleep(0.25)
except (RuntimeError, ValueError, TypeError) as exc:
return {"status": "error", "reason": redact_message(str(exc))}
if exit_code is None:
return {"status": "error", "reason": "PrivateMuse enrollment timed out"}
reasons = {
20: "deployment projects registry is unavailable or unsafe",
21: "deployment projects registry is not a regular file",
22: "deployment projects registry does not contain an object",
23: "PrivateMuse already has a different deployment target",
24: "mounted PrivateMuse bare repository is unavailable",
25: "verified PrivateMuse legacy revision cannot be resolved",
26: "PrivateMuse main cannot be resolved",
27: "verified PrivateMuse legacy revision is not an ancestor of main",
28: "required PrivateMuse production Compose file is absent from main",
29: "deployment registry lock file cannot be opened",
30: "deployment registry lock cannot be acquired",
31: "deployment projects registry cannot be read as JSON",
32: "deployment projects registry backup cannot be created",
34: "deployment projects registry replacement failed",
36: "deployment projects registry host alias is unavailable or mismatched",
40: "PrivateMuse fixed enrollment contract was altered",
41: "PrivateMuse production environment file is unavailable or unsafe",
42: "PrivateMuse deployment state file is unavailable or invalid",
43: "PrivateMuse deployment state conflicts with the verified legacy baseline",
44: "PrivateMuse Compose project inventory could not be read",
45: "PrivateMuse Compose project contains an unexpected container",
46: "required PrivateMuse production container is missing or unhealthy",
47: "stale PrivateMuse audio container could not be inspected",
48: "a stale PrivateMuse audio provider is still running and was not removed",
49: "a verified stopped PrivateMuse audio container could not be removed",
50: "deployment state directory is unavailable or unsafe",
51: "verified PrivateMuse legacy state could not be written durably",
52: "deployment controller rejected PrivateMuse legacy-state bootstrap",
53: "PrivateMuse legacy-state bootstrap did not produce the expected managed state",
54: "PrivateMuse deployment target is not an object",
55: "PrivateMuse deployment target is owned by the controller v2 managed-target contract",
60: "PrivateMuse deployment target differs in branch",
61: "PrivateMuse deployment target differs in slug",
62: "PrivateMuse deployment target differs in source_dir",
63: "PrivateMuse deployment target differs in exact_source",
64: "PrivateMuse deployment target differs in env_source",
65: "PrivateMuse deployment target differs in compose_files",
66: "PrivateMuse deployment target differs in compose_project",
67: "PrivateMuse deployment target differs in containers",
68: "PrivateMuse deployment target differs in verify_oci_revision",
69: "PrivateMuse deployment target differs in health_timeout",
70: "PrivateMuse deployment target differs in environment",
71: "PrivateMuse deployment target matches canonical fields but has additional or invalid controller metadata",
}
if exit_code != 0:
return {
"status": "validation_failed",
"reason": reasons.get(
exit_code,
"PrivateMuse enrollment failed closed before a production application deploy",
),
}
return {
"status": "enrolled",
"repository": _REPOSITORY,
"legacy_revision": _LEGACY_REVISION,
"compose_project": _PROJECT["compose_project"],
"containers": list(_PROJECT["containers"]),
}