388 lines
17 KiB
Python
388 lines
17 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
|
|
from .portfolio_deployment_enrollment import PortfolioDeploymentEnrollment
|
|
|
|
_CONTROLLER = "gitea-deploy-control"
|
|
_REPOSITORY = "jens/portfolio"
|
|
_LIVE_REVISION = "52c6f2fc74a1e39ad288f1c5e987a9cd5b072d3e"
|
|
_FAILED_REVISION = "b7955d90b63860ffff8e0bd84be123aaba632043"
|
|
_FAILURE_MESSAGE = "Existing production containers require a valid healthy live deployment state"
|
|
|
|
# One-time, fail-closed re-attestation for the exact observed Portfolio state.
|
|
# The recovery never builds, restarts, recreates or stops application containers.
|
|
# It uses the controller's own receipt/runtime verifier and accepts only a change
|
|
# in ephemeral container IDs while immutable image/provenance evidence remains
|
|
# exactly equal to the previously recorded verified receipt.
|
|
_REATTEST_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
|
|
|
|
repository, live_revision, failed_revision, failure_message = sys.argv[1:5]
|
|
expected_repository = "jens/portfolio"
|
|
expected_live_revision = "52c6f2fc74a1e39ad288f1c5e987a9cd5b072d3e"
|
|
expected_failed_revision = "b7955d90b63860ffff8e0bd84be123aaba632043"
|
|
expected_failure_message = "Existing production containers require a valid healthy live deployment state"
|
|
if (
|
|
repository != expected_repository
|
|
or live_revision != expected_live_revision
|
|
or failed_revision != expected_failed_revision
|
|
or failure_message != expected_failure_message
|
|
):
|
|
raise SystemExit(40)
|
|
|
|
controller_path = Path("/opt/gitea-deploy/deploy.py")
|
|
registry_path = Path("/opt/gitea-deploy/projects.json")
|
|
state_root = Path("/mnt/cache/appdata/gitea-deploy-runners/state")
|
|
state_path = state_root / "portfolio.json"
|
|
project_lock_path = state_root / "locks" / "portfolio.lock"
|
|
global_lock_path = Path("/mnt/cache/appdata/gitea-deploy-runners/deploy.lock")
|
|
|
|
if controller_path.is_symlink() or not controller_path.is_file():
|
|
raise SystemExit(20)
|
|
if registry_path.is_symlink() or not registry_path.is_file():
|
|
raise SystemExit(21)
|
|
try:
|
|
projects = json.loads(registry_path.read_text(encoding="utf-8"))
|
|
except Exception:
|
|
raise SystemExit(22)
|
|
project = projects.get(repository) if isinstance(projects, dict) else None
|
|
if not isinstance(project, dict):
|
|
raise SystemExit(23)
|
|
if (
|
|
project.get("branch") != "main"
|
|
or project.get("slug") != "portfolio"
|
|
or project.get("compose_project") != "portfolio"
|
|
or project.get("containers") != ["Portfolio", "Portfolio-api"]
|
|
or project.get("compose_files") != ["docker-compose.yml"]
|
|
or project.get("exact_source") is not True
|
|
or project.get("verify_oci_revision") is not False
|
|
):
|
|
raise SystemExit(24)
|
|
|
|
owner, name = repository.split("/", 1)
|
|
bare_repo = Path("/gitea-repositories") / owner / f"{name}.git"
|
|
if not bare_repo.is_dir():
|
|
raise SystemExit(25)
|
|
|
|
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,
|
|
)
|
|
|
|
for revision in (live_revision, failed_revision):
|
|
resolved = git_run("rev-parse", f"{revision}^{{commit}}")
|
|
if resolved.returncode != 0 or resolved.stdout.strip() != revision:
|
|
raise SystemExit(26)
|
|
main = git_run("rev-parse", "refs/heads/main")
|
|
if main.returncode != 0 or main.stdout.strip() != failed_revision:
|
|
raise SystemExit(27)
|
|
if git_run("merge-base", "--is-ancestor", live_revision, failed_revision).returncode != 0:
|
|
raise SystemExit(28)
|
|
|
|
try:
|
|
spec = importlib.util.spec_from_file_location("itworx_deploy_controller", controller_path)
|
|
controller = importlib.util.module_from_spec(spec) if spec and spec.loader else None
|
|
if controller is None or spec is None or spec.loader is None:
|
|
raise RuntimeError("controller import unavailable")
|
|
spec.loader.exec_module(controller)
|
|
except Exception:
|
|
raise SystemExit(29)
|
|
|
|
state_root.mkdir(parents=True, exist_ok=True)
|
|
project_lock_path.parent.mkdir(parents=True, exist_ok=True)
|
|
try:
|
|
project_lock = project_lock_path.open("a+", encoding="utf-8")
|
|
global_lock = global_lock_path.open("a+", encoding="utf-8")
|
|
except Exception:
|
|
raise SystemExit(30)
|
|
|
|
with project_lock:
|
|
try:
|
|
fcntl.flock(project_lock, fcntl.LOCK_EX)
|
|
except Exception:
|
|
raise SystemExit(31)
|
|
with global_lock:
|
|
try:
|
|
fcntl.flock(global_lock, fcntl.LOCK_EX)
|
|
except Exception:
|
|
raise SystemExit(31)
|
|
|
|
try:
|
|
current_projects = json.loads(registry_path.read_text(encoding="utf-8"))
|
|
except Exception:
|
|
raise SystemExit(32)
|
|
if current_projects.get(repository) != project:
|
|
raise SystemExit(33)
|
|
|
|
if state_path.is_symlink() or not state_path.is_file():
|
|
raise SystemExit(34)
|
|
try:
|
|
state = json.loads(state_path.read_text(encoding="utf-8"))
|
|
except Exception:
|
|
raise SystemExit(34)
|
|
if not isinstance(state, dict):
|
|
raise SystemExit(34)
|
|
attempt = state.get("last_attempt")
|
|
receipt = state.get("live_receipt")
|
|
artifacts = state.get("live_artifacts")
|
|
if not (
|
|
state.get("schema_version") == 3
|
|
and state.get("repository") == repository
|
|
and state.get("revision") == live_revision
|
|
and state.get("live_revision") == live_revision
|
|
and state.get("live_status") == "unknown"
|
|
and state.get("live_legacy_provenance") is False
|
|
and state.get("status") == "failed"
|
|
and isinstance(attempt, dict)
|
|
and attempt.get("revision") == failed_revision
|
|
and attempt.get("status") == "failed"
|
|
and attempt.get("message") == failure_message
|
|
and isinstance(receipt, dict)
|
|
and bool(receipt)
|
|
and isinstance(artifacts, dict)
|
|
):
|
|
raise SystemExit(35)
|
|
|
|
project = dict(project)
|
|
project["_repository_name"] = repository
|
|
project["_repository_path"] = str(bare_repo)
|
|
|
|
try:
|
|
build = controller.validate_recorded_receipt(project, receipt, repository, live_revision)
|
|
recorded_receipt_containers = receipt.get("containers")
|
|
if not isinstance(recorded_receipt_containers, dict):
|
|
raise RuntimeError("receipt containers missing")
|
|
expected_image_ids = {
|
|
str(container_name): str(dict(raw).get("image_id") or "")
|
|
for container_name, raw in recorded_receipt_containers.items()
|
|
if isinstance(raw, dict)
|
|
}
|
|
if set(expected_image_ids) != {"Portfolio", "Portfolio-api"} or not all(expected_image_ids.values()):
|
|
raise RuntimeError("receipt image IDs missing")
|
|
observed_receipt = controller.verify_deployment(
|
|
project,
|
|
build,
|
|
expected_image_ids=expected_image_ids,
|
|
)
|
|
observed_artifacts = controller.deployment_artifacts(project)
|
|
except Exception:
|
|
raise SystemExit(36)
|
|
|
|
immutable_receipt_keys = ("schema", "repository", "revision", "tree", "treeDigest", "build", "live")
|
|
if any(observed_receipt.get(key) != receipt.get(key) for key in immutable_receipt_keys):
|
|
raise SystemExit(37)
|
|
observed_receipt_containers = observed_receipt.get("containers")
|
|
if not isinstance(observed_receipt_containers, dict) or set(observed_receipt_containers) != set(recorded_receipt_containers):
|
|
raise SystemExit(37)
|
|
|
|
container_id_changed = False
|
|
for container_name in sorted(recorded_receipt_containers):
|
|
old = dict(recorded_receipt_containers[container_name])
|
|
new = dict(observed_receipt_containers[container_name])
|
|
if old.get("container_id") != new.get("container_id"):
|
|
container_id_changed = True
|
|
old_without_id = {key: value for key, value in old.items() if key != "container_id"}
|
|
new_without_id = {key: value for key, value in new.items() if key != "container_id"}
|
|
if old_without_id != new_without_id:
|
|
raise SystemExit(38)
|
|
if not container_id_changed:
|
|
raise SystemExit(39)
|
|
|
|
recorded_artifact_containers = artifacts.get("containers")
|
|
observed_artifact_containers = observed_artifacts.get("containers")
|
|
if (
|
|
not isinstance(recorded_artifact_containers, dict)
|
|
or not isinstance(observed_artifact_containers, dict)
|
|
or set(recorded_artifact_containers) != set(observed_artifact_containers)
|
|
or artifacts.get("compose_sha256") != observed_artifacts.get("compose_sha256")
|
|
):
|
|
raise SystemExit(41)
|
|
for container_name in sorted(recorded_artifact_containers):
|
|
old = dict(recorded_artifact_containers[container_name])
|
|
new = dict(observed_artifact_containers[container_name])
|
|
old_without_id = {key: value for key, value in old.items() if key != "container_id"}
|
|
new_without_id = {key: value for key, value in new.items() if key != "container_id"}
|
|
if old_without_id != new_without_id:
|
|
raise SystemExit(41)
|
|
|
|
now = int(time.time())
|
|
repaired = dict(state)
|
|
repaired["live_status"] = "healthy"
|
|
repaired["live_artifacts"] = observed_artifacts
|
|
repaired["artifacts"] = observed_artifacts
|
|
repaired["live_receipt"] = observed_receipt
|
|
repaired["status"] = "success"
|
|
repaired["message"] = "Verified Portfolio runtime re-attested after container recreation"
|
|
repaired["updated_at_epoch"] = now
|
|
repaired["last_attempt"] = {
|
|
"revision": live_revision,
|
|
"status": "success",
|
|
"message": "Verified live Portfolio runtime receipt re-attestation",
|
|
"updated_at_epoch": now,
|
|
"metadata": {
|
|
"reconciled_failed_revision": failed_revision,
|
|
"container_identity_only": True,
|
|
},
|
|
}
|
|
|
|
backup = state_path.with_name(f"portfolio.json.reattest-backup-{now}-{os.getpid()}")
|
|
temporary = state_path.with_name(f".{state_path.name}.{os.getpid()}.reattest.tmp")
|
|
try:
|
|
mode = stat.S_IMODE(state_path.stat().st_mode)
|
|
shutil.copy2(state_path, backup)
|
|
with temporary.open("x", encoding="utf-8") as handle:
|
|
handle.write(json.dumps(repaired, indent=2, sort_keys=True) + "\n")
|
|
handle.flush()
|
|
os.fsync(handle.fileno())
|
|
os.chmod(temporary, mode)
|
|
os.replace(temporary, 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.unlink(missing_ok=True)
|
|
except Exception:
|
|
pass
|
|
raise SystemExit(42)
|
|
|
|
try:
|
|
verified = controller.verified_current_receipt(
|
|
repository,
|
|
project,
|
|
live_revision,
|
|
str(receipt.get("tree") or ""),
|
|
)
|
|
except Exception:
|
|
verified = None
|
|
if not isinstance(verified, dict) or verified != observed_receipt:
|
|
raise SystemExit(43)
|
|
|
|
print("re-attested", flush=True)
|
|
'''
|
|
|
|
|
|
class PortfolioDeploymentRecovery:
|
|
"""Recover the exact known Portfolio state drift, otherwise use normal enrollment."""
|
|
|
|
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)
|
|
controller = dict(status.get("container") or {}) if status.get("status") == "ok" else {}
|
|
container_id = str(controller.get("id") or "")
|
|
config = self.portainer._config()
|
|
if controller.get("name") != _CONTROLLER or controller.get("state") != "running" or not container_id or not config:
|
|
return PortfolioDeploymentEnrollment(portainer=self.portainer).run()
|
|
|
|
command = [
|
|
"python3",
|
|
"-c",
|
|
_REATTEST_SCRIPT,
|
|
_REPOSITORY,
|
|
_LIVE_REVISION,
|
|
_FAILED_REVISION,
|
|
_FAILURE_MESSAGE,
|
|
]
|
|
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": "Portfolio runtime re-attestation 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() + 45
|
|
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": "Portfolio runtime re-attestation timed out"}
|
|
if exit_code == 0:
|
|
return {
|
|
"status": "reconciled",
|
|
"repository": _REPOSITORY,
|
|
"live_revision": _LIVE_REVISION,
|
|
"failed_revision": _FAILED_REVISION,
|
|
"containers": ["Portfolio", "Portfolio-api"],
|
|
"compose_project": "portfolio",
|
|
}
|
|
|
|
# Exit 35 means the exact recoverable state is no longer present. In
|
|
# normal/healthy states the established enrollment routine remains the
|
|
# authoritative path. Every other re-attestation failure is fail-closed.
|
|
if exit_code == 35:
|
|
return PortfolioDeploymentEnrollment(portainer=self.portainer).run()
|
|
|
|
reasons = {
|
|
20: "deployment controller executable is unavailable or unsafe",
|
|
21: "deployment projects registry is unavailable or unsafe",
|
|
22: "deployment projects registry cannot be read",
|
|
23: "Portfolio deployment target is missing",
|
|
24: "Portfolio deployment target differs from the bounded paired contract",
|
|
25: "mounted Portfolio bare repository is unavailable",
|
|
26: "known Portfolio recovery revisions cannot be resolved",
|
|
27: "Portfolio main changed after the bounded recovery contract was recorded",
|
|
28: "known Portfolio live revision is not an ancestor of the failed revision",
|
|
29: "deployment controller verifier cannot be loaded",
|
|
30: "deployment lock files cannot be opened",
|
|
31: "deployment locks cannot be acquired",
|
|
32: "deployment projects registry changed or became unreadable while waiting",
|
|
33: "Portfolio deployment target changed while recovery was waiting",
|
|
34: "Portfolio deployment state is unavailable or invalid",
|
|
36: "controller could not re-attest the current Portfolio runtime against recorded immutable image IDs",
|
|
37: "Portfolio receipt identity changed during re-attestation",
|
|
38: "Portfolio immutable container/image provenance differs from the recorded receipt",
|
|
39: "Portfolio container identities did not change; the known recovery cause is absent",
|
|
40: "fixed Portfolio re-attestation contract was altered",
|
|
41: "Portfolio recorded live artifacts differ beyond ephemeral container IDs",
|
|
42: "re-attested Portfolio deployment state could not be written durably",
|
|
43: "controller rejected the repaired Portfolio receipt after write",
|
|
}
|
|
return {
|
|
"status": "validation_failed",
|
|
"reason": reasons.get(exit_code, "Portfolio runtime re-attestation failed closed"),
|
|
}
|