Files
Unraid-AppOps-Gateway-Public/backend/tests/test_appops_live_state_recovery.py
T

263 lines
12 KiB
Python

from __future__ import annotations
import copy
import hashlib
import importlib.util
import json
import os
import sys
import types
from pathlib import Path
import pytest
ROOT = Path(__file__).resolve().parents[2]
REPOSITORY = "jens/unraid-appops-gateway"
CONTAINER = "Unraid-AppOps-Gateway"
OBSERVED = "5" * 40
RECORDED = "6" * 40
ATTEMPT = "a" * 40
IMAGE = "sha256:" + "5" * 64
OLD_IMAGE = "sha256:" + "6" * 64
def receipt(revision=RECORDED, image=OLD_IMAGE):
tree = "b" * 40
digest = hashlib.sha256(f"git-tree:{tree}".encode()).hexdigest()
built = "2026-10-02T00:00:00.000Z"
return {
"schema": "gitea-deploy-receipt.v1", "repository": REPOSITORY,
"revision": revision, "tree": tree, "treeDigest": digest,
"build": {"id": revision[:12], "repository": REPOSITORY,
"revision": revision, "tree": tree, "treeDigest": digest,
"builtAt": built},
"containers": {CONTAINER: {"image_id": image,
"image_revision": revision, "image_source": REPOSITORY,
"image_tree": tree, "image_version": revision[:12],
"image_created": built}}, "live": None,
}
@pytest.fixture
def recovery(monkeypatch, tmp_path):
if "fcntl" not in sys.modules:
shim = types.ModuleType("fcntl")
shim.LOCK_EX, shim.LOCK_UN = 2, 8
shim.flock = lambda *_args: None
monkeypatch.setitem(sys.modules, "fcntl", shim)
spec = importlib.util.spec_from_file_location(
"appops_recovery_test", ROOT / "ops/gitea-deploy-broker/deploy.py")
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
state_root = tmp_path / "state"
state_root.mkdir(mode=0o700)
monkeypatch.setattr(module, "STATE_ROOT", state_root)
monkeypatch.setattr(module, "require_host_root", lambda: None)
monkeypatch.setattr(module, "_root_owned", lambda _path: True)
if os.name == "nt":
# Windows does not expose POSIX write bits; keep real inode/type checks.
real_lstat = Path.lstat
def posix_permissions(path):
observed = real_lstat(path)
return types.SimpleNamespace(st_mode=observed.st_mode & ~0o022,
st_dev=observed.st_dev, st_ino=observed.st_ino)
monkeypatch.setattr(Path, "lstat", posix_permissions)
fsyncs = []
monkeypatch.setattr(module, "_fsync_directory", lambda path: fsyncs.append(path))
project = {"slug": "unraid-appops-gateway", "branch": "main",
"gitea_repository_id": 9, "containers": [CONTAINER],
"compose_project": "unraid-appops-gateway",
"compose_files": ["docker-compose.yml"], "verify_oci_revision": True,
"environment": {"APPOPS_DATA_PATH": str(module.APPOPS_DATA_ROOT)},
"source_dir": str(tmp_path), "_repository_name": REPOSITORY}
(tmp_path / "docker-compose.yml").write_text("services: {}\n")
public = {key: value for key, value in project.items() if not key.startswith("_")}
monkeypatch.setattr(module, "load_projects", lambda: {REPOSITORY: copy.deepcopy(public)})
def validate(_repo, branch, revision):
assert branch == "main"
if revision not in {OBSERVED, ATTEMPT}:
raise module.DeployError("revision is not a protected ancestor")
return revision
monkeypatch.setattr(module, "validate_commit", validate)
source = "https://gitea.example.com/Jens/Unraid-AppOps-Gateway"
labels = {"org.opencontainers.image.revision": OBSERVED,
"org.opencontainers.image.source": source,
"com.docker.compose.project": "unraid-appops-gateway"}
container = {"Id": "c" * 64, "Image": IMAGE,
"State": {"Running": True, "Health": {"Status": "healthy"}},
"Config": {"Image": "appops:observed", "Labels": labels},
"NetworkSettings": {"Ports": {"1216/tcp": [{"HostPort": "1216"}]}},
"Mounts": [{"Destination": "/data", "Type": "bind", "RW": True,
"Source": str(module.APPOPS_DATA_ROOT)}]}
image = {"Id": IMAGE, "Config": {"Labels": copy.deepcopy(labels)}}
monkeypatch.setattr(module, "inspect_container", lambda name: copy.deepcopy(container))
monkeypatch.setattr(module, "inspect_image", lambda name: copy.deepcopy(image))
monkeypatch.setattr(module, "compose_project_container_names", lambda name: {CONTAINER})
def output(command, **_kwargs):
assert command == ["docker", "ps", "-a", "--format", "{{.Names}}"]
return CONTAINER + "\n"
monkeypatch.setattr(module, "output", output)
def forbidden(*_args, **_kwargs):
pytest.fail("recovery attempted a host mutation")
monkeypatch.setattr(module, "run", forbidden)
prior = {"schema_version": 3, "repository": REPOSITORY,
"slug": project["slug"], "compose_project": project["compose_project"],
"revision": RECORDED, "live_revision": RECORDED,
"status": "failed", "live_status": "unknown", "live_legacy_provenance": False,
"live_artifacts": {"containers": {CONTAINER: {"image_id": OLD_IMAGE}}},
"live_receipt": receipt(), "rollback_snapshot": {"revision": "7" * 40},
"last_attempt": {"revision": ATTEMPT, "status": "failed",
"message": module._APPOPS_RECOVERY_FAILURE}}
path = state_root / "unraid-appops-gateway.json"
def save():
path.write_text(json.dumps(prior), encoding="utf-8")
path.chmod(0o600)
save()
return types.SimpleNamespace(m=module, project=project, repo=tmp_path,
path=path, prior=prior, container=container, image=image,
public=public, save=save, fsyncs=fsyncs)
def test_reattest_preserves_evidence_and_observed_legacy_identity(recovery):
r = recovery
before = r.path.read_bytes()
compose = (r.repo / "docker-compose.yml").read_bytes()
result = r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)
state = json.loads(r.path.read_bytes())
backup = r.path.parent / result["backupFile"]
assert backup.read_bytes() == before
assert state["recovery"]["prior_state_sha256"] == hashlib.sha256(before).hexdigest()
assert state["last_attempt"] == r.prior["last_attempt"]
assert state["live_revision"] == OBSERVED
assert state["live_legacy_provenance"] is True
assert state["live_status"] == "healthy"
assert state["live_receipt"] == state["rollback_snapshot"] == {}
assert state["live_artifacts"]["compose_sha256"] == ""
assert state["live_artifacts"]["containers"][CONTAINER]["image_id"] == IMAGE
assert state["recovery"]["retired_receipt_revision"] == RECORDED
assert (r.repo / "docker-compose.yml").read_bytes() == compose
assert r.fsyncs == [r.path.parent, r.path.parent]
@pytest.mark.parametrize("field,value", [
("schema_version", 2), ("status", "success"), ("live_status", "healthy"),
("live_legacy_provenance", None), ("live_legacy_provenance", 0),
("repository", "jens/other"), ("live_revision", OBSERVED),
("rollback_snapshot", None),
("live_receipt", None), ("live_receipt", {}), ("live_receipt", "invalid"),
("live_artifacts", None), ("live_artifacts", {}),
])
def test_rejects_ineligible_recorded_state(recovery, field, value):
r = recovery
r.prior[field] = value
r.save()
before = r.path.read_bytes()
with pytest.raises(r.m.DeployError):
r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)
assert r.path.read_bytes() == before
assert not list(r.path.parent.glob("*.bak"))
@pytest.mark.parametrize("failure", ["wrong-failure", "non-ancestor", "same-image", "receipt-image", "receipt-revision"])
def test_rejects_unproven_historical_drift(recovery, failure):
r = recovery
if failure == "wrong-failure":
r.prior["last_attempt"]["message"] = "Deployment failed"
elif failure == "non-ancestor":
r.prior["last_attempt"]["revision"] = "d" * 40
elif failure == "same-image":
r.prior["live_artifacts"]["containers"][CONTAINER]["image_id"] = IMAGE
elif failure == "receipt-image":
r.prior["live_receipt"]["containers"][CONTAINER]["image_id"] = IMAGE
else:
r.prior["live_receipt"]["revision"] = OBSERVED
r.save()
before = r.path.read_bytes()
with pytest.raises(r.m.DeployError):
r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)
assert r.path.read_bytes() == before
@pytest.mark.parametrize("failure", ["unhealthy", "volume", "readonly", "mount-path", "port", "image", "source", "revision", "extra-container", "config", "owner"])
def test_rejects_wrong_actual_runtime_or_target(recovery, monkeypatch, failure):
r = recovery
if failure == "unhealthy":
r.container["State"]["Health"]["Status"] = "unhealthy"
elif failure in {"volume", "readonly", "mount-path"}:
key, value = {"volume": ("Type", "volume"), "readonly": ("RW", False),
"mount-path": ("Source", "/other")}[failure]
r.container["Mounts"][0][key] = value
elif failure == "port":
r.container["NetworkSettings"]["Ports"]["1216/tcp"][0]["HostPort"] = "9999"
elif failure == "image":
r.image["Id"] = OLD_IMAGE
elif failure in {"source", "revision"}:
r.image["Config"]["Labels"]["org.opencontainers.image." + failure] = "wrong"
elif failure == "extra-container":
monkeypatch.setattr(r.m, "compose_project_container_names", lambda name: {CONTAINER, "other"})
elif failure == "config":
r.public["branch"] = "other"
else:
monkeypatch.setattr(r.m, "_root_owned", lambda path: False)
before = r.path.read_bytes()
with pytest.raises(r.m.DeployError):
r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)
assert r.path.read_bytes() == before
@pytest.mark.parametrize("change", ["bytes", "inode", "runtime", "config"])
def test_rechecks_after_durable_backup_before_replace(recovery, monkeypatch, change):
r = recovery
def after_backup(_path):
if change == "bytes":
r.path.write_bytes(b'{"external": true}')
elif change == "inode":
replacement = r.path.with_suffix(".replacement")
replacement.write_bytes(r.path.read_bytes())
replacement.chmod(0o600)
replacement.replace(r.path)
elif change == "runtime":
r.container["Id"] = "d" * 64
else:
r.public["branch"] = "changed"
monkeypatch.setattr(r.m, "_fsync_directory", after_backup)
with pytest.raises(r.m.DeployError):
r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)
state = json.loads(r.path.read_bytes())
assert "recovery" not in state
assert len(list(r.path.parent.glob("*.bak"))) == 1
assert not list(r.path.parent.glob("*.tmp"))
def test_next_normal_delivery_keeps_recovery_audit(recovery):
r = recovery
result = r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)
audit = json.loads(r.path.read_bytes())["recovery"]
r.m.write_state(REPOSITORY, r.project, OBSERVED, "success", "Verified normal deployment",
live_receipt=receipt(OBSERVED, IMAGE), live_legacy_provenance=False)
state = json.loads(r.path.read_bytes())
assert state["recovery"] == audit
assert (r.path.parent / result["backupFile"]).exists()
assert state["live_receipt"]["revision"] == OBSERVED
assert state["live_legacy_provenance"] is False
assert state["last_attempt"]["message"] == "Verified normal deployment"
def test_post_replace_fsync_failure_reports_applied_unknown_and_retains_backup(recovery, monkeypatch):
r = recovery
calls = 0
def fail_after_replace(_path):
nonlocal calls
calls += 1
if calls == 2:
raise OSError("synthetic directory fsync failure")
monkeypatch.setattr(r.m, "_fsync_directory", fail_after_replace)
before = r.path.read_bytes()
with pytest.raises(r.m.DeployError, match="state was replaced but durability is unknown"):
r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)
state = json.loads(r.path.read_bytes())
assert state["live_revision"] == OBSERVED
assert (r.path.parent / state["recovery"]["backup_file"]).read_bytes() == before
with pytest.raises(r.m.DeployError, match="exact failed pre-mutation"):
r.m.reattest_appops_live_state(r.project, r.repo, OBSERVED)