Files
Unraid-AppOps-Gateway-Public/backend/tests/test_runner_dind_rollback_tail_recovery.py

279 lines
14 KiB
Python

from __future__ import annotations
import copy
import hashlib
import importlib.util
import json
import sys
from pathlib import Path
import pytest
BROKER = Path(__file__).resolve().parents[2] / "ops" / "gitea-deploy-broker"
def load(name, path):
spec = importlib.util.spec_from_file_location(name, path)
module = importlib.util.module_from_spec(spec)
sys.modules[name] = module
spec.loader.exec_module(module)
return module
@pytest.fixture
def recovery(monkeypatch):
monkeypatch.syspath_prepend(str(BROKER))
return load("dind_tail_test", BROKER / "runner_dind_rollback_tail_recovery.py")
@pytest.fixture
def owned(recovery, monkeypatch, tmp_path):
shared = load("tail_shared", Path(__file__).with_name("test_runner_failed_cutover_recovery.py"))
monkeypatch.setattr(recovery, "PROBE_ID", "a" * 64, raising=False)
values = shared.owned.__wrapped__(recovery, monkeypatch, tmp_path)
fleet, _, q, driver, marker, journal, _, rows, dind, events, pending = values
journal.update(stage="registrations_created", cgroupCreated=True, cgroupEnsurePending=False,
newRegistrationIds=dict(recovery.REGISTRATIONS), backupDirectory=recovery.BACKUP,
cgroupRootSubtree=list(recovery.CONTROLLERS), originalContainers=[])
for row in rows:
row.update(displayName=None, status="online")
rows.extend([{"id": rid, "name": name, "labels": [label], "disabled": False, "busy": False,
"displayName": None, "status": "online"}
for rid, name, label in ((13, "itworx-unraid-deploy", "unraid-deploy"),
(14, "itworx-windows11-native", "windows-native"))])
fleet.TASK_IMAGE_MANIFEST_PATH.unlink()
fleet.TASK_IMAGE_ROOT.rmdir()
monkeypatch.setattr(fleet, "ADMISSION_BOOTSTRAP_RECEIPT_PATH", tmp_path / "absent")
dind["Id"] = recovery.RUNTIME_IDS[fleet.DIND_NAME]
legacy = {name: {**copy.deepcopy(dind), "Id": cid} for name, cid in recovery.RUNTIME_IDS.items()
if name != fleet.DIND_NAME}
monkeypatch.setattr(fleet, "_validate_legacy_containers", lambda _: (copy.deepcopy(legacy), []))
monkeypatch.setattr(fleet, "_inspect", lambda name: copy.deepcopy(dind) if name == fleet.DIND_NAME else copy.deepcopy(legacy.get(name)))
family = {"exists": True, "controllers": list(recovery.CONTROLLERS),
"rootSubtreeControllers": list(recovery.CONTROLLERS), "familySubtreeControllers": list(recovery.CONTROLLERS),
"groupCount": 1, "hasProcesses": False, "familyCpuMax": "max 100000", "familyMemoryMax": "max"}
monkeypatch.setattr(fleet, "_cgroup_family_snapshot", lambda: copy.deepcopy(family))
monkeypatch.setattr(recovery, "_probe_absent", lambda: True)
monkeypatch.setattr(recovery, "_prior_receipt", lambda: b"pinned")
monkeypatch.setattr(recovery, "MARKER_SHA", hashlib.sha256(json.dumps(marker).encode()).hexdigest())
monkeypatch.setattr(recovery, "JOURNAL_SHA", hashlib.sha256(json.dumps(journal).encode()).hexdigest())
monkeypatch.setattr(recovery, "__file__", str(recovery.BUNDLE / "runner_dind_rollback_tail_recovery.py"))
def legacy_private_inventory(self, args, **kwargs):
assert args == ["docker", "exec", "-e", "DOCKER_HOST=tcp://127.0.0.1:2375",
fleet.DIND_NAME, "docker", "ps", "-a", "--no-trunc", "-q"]
assert kwargs == {"timeout": 20, "max_output": 128 * 1024}
return ""
monkeypatch.setattr(driver, "_run", legacy_private_inventory)
monkeypatch.setattr(recovery, "_load", lambda: (fleet, fleet, q, {}))
def ensure(*, allow_existing):
assert allow_existing is True
events.append("ensure")
family.update(familyCpuMax="800000 100000", familyMemoryMax=str(16 * 1024**3))
return False
monkeypatch.setattr(fleet, "_ensure_cgroup_family", ensure)
return values, family, legacy
def test_readonly_discloses_probe_and_never_claims_or_ensures(recovery, owned, monkeypatch):
values, _, _ = owned
monkeypatch.setattr(recovery, "_claim", lambda _: pytest.fail("claim"))
monkeypatch.setattr(values[0], "_ensure_cgroup_family", lambda **kw: pytest.fail("ensure"))
result = recovery.run()
assert result["status"] == "ready"
assert result["diagnosticProbeLifecycle"] is True
assert result["recoveryEffectsPerformed"] is False
assert values[9].count("guard-cas-checked") == 2
def test_claim_ensure_native_rollback_all_under_both_locks(recovery, owned, monkeypatch):
values, family, _ = owned
fleet, _, _, _, _, _, _, _, _, events, _ = values
monkeypatch.setattr(recovery, "_claim", lambda _: events.append("claim"))
def tail(journal, barrier):
barrier()
assert events.index("claim") < events.index("ensure")
assert "maintenance-exit" not in events and "migration-exit" not in events
assert family["familyCpuMax"] == "800000 100000"
events.append("native-tail")
return True
monkeypatch.setattr(fleet, "_rollback_before_activation_locked", tail)
result = recovery.run(execute=True, confirmation=recovery.CONFIRMATION)
assert result["status"] == "rolled_back" and result["invocationConsumed"] is True
assert events.index("native-tail") < events.index("maintenance-exit") < events.index("migration-exit")
@pytest.mark.parametrize("mutation", ["cid", "unhealthy", "busy", "extra-registration", "specialized-file", "guards", "private-work",
"family-process", "family-child", "family-cpu", "family-memory", "family-subtree", "journal", "claim"])
def test_mutations_rejected_before_claim_and_effect(recovery, owned, monkeypatch, mutation):
values, family, legacy = owned
fleet, _, _, _, _, journal, _, rows, dind, events, pending = values
if mutation == "cid":
dind["Id"] = "f" * 64
elif mutation == "unhealthy":
dind["State"]["Health"]["Status"] = "unhealthy"
elif mutation == "busy":
rows[0]["busy"] = True
elif mutation == "extra-registration":
rows.append({"id": 23, "busy": False})
elif mutation == "specialized-file":
fleet.RUNNER_CONFIG_ROOT.write_text("private-canary")
elif mutation == "guards":
monkeypatch.setattr(fleet, "_guard_state_unchanged", lambda _: False)
elif mutation == "private-work":
monkeypatch.setattr(values[3], "_run", lambda *a, **kw: "private-canary")
elif mutation == "family-process":
family["hasProcesses"] = True
elif mutation == "family-child":
family["groupCount"] = 2
elif mutation == "family-cpu":
family["familyCpuMax"] = "800000 100000"
elif mutation == "family-memory":
family["familyMemoryMax"] = "1"
elif mutation == "family-subtree":
family["familySubtreeControllers"].pop()
elif mutation == "journal":
journal["backupDirectory"] += "foreign"
elif mutation == "claim":
recovery.CLAIM.write_text("consumed")
monkeypatch.setattr(recovery, "_claim", lambda _: pytest.fail("claim"))
result = recovery.run(execute=True, confirmation=recovery.CONFIRMATION)
assert result["status"] == "blocked"
assert result["recoveryEffectsPerformed"] is False
assert "ensure" not in events and "private-canary" not in json.dumps(result)
def test_admission_drift_refuses_before_claim(recovery, owned, monkeypatch):
original = recovery._admit
calls = []
def drift(*args, **kwargs):
journal, proof = original(*args, **kwargs)
calls.append(1)
return journal, {**proof, "runtimeDigest": str(len(calls))}
monkeypatch.setattr(recovery, "_admit", drift)
monkeypatch.setattr(recovery, "_claim", lambda _: pytest.fail("claim"))
assert recovery.run(execute=True, confirmation=recovery.CONFIRMATION)["reason"] == "admission_changed"
def test_ensure_failure_consumes_and_reports_unknown_redacted(recovery, owned, monkeypatch):
monkeypatch.setattr(recovery, "_claim", lambda _: None)
def fail(**kwargs):
raise RuntimeError("token=private-canary")
monkeypatch.setattr(owned[0][0], "_ensure_cgroup_family", fail)
result = recovery.run(execute=True, confirmation=recovery.CONFIRMATION)
assert result["status"] == "unknown" and result["invocationConsumed"] is True
assert result["recoveryEffectsPerformed"] is None and "private-canary" not in json.dumps(result)
def test_prior_receipt_closed_schema_and_raw_hash(recovery, monkeypatch):
raw = json.dumps({"schema": "appops.failed-cutover-recovery/v1", "phase": "consumed",
"incidentJobId": recovery.INCIDENT_JOB, "incidentBackupDirectory": recovery.BACKUP,
"sourceRevision": recovery.ORIGIN, "sourceTree": recovery.TREE, "artifactSha256": recovery.ARTIFACT_SHA,
"candidateSha256": recovery.FLEET_SHA, "proof": recovery.PRIOR_PROOF}).encode()
monkeypatch.setattr(recovery, "_read", lambda *a, **kw: raw)
monkeypatch.setattr(recovery, "PRIOR_SHA", hashlib.sha256(raw).hexdigest())
assert recovery._prior_receipt() == raw
for changed in ({"extra": "private-canary"}, {"incidentJobId": "other"}, {"proof": {}}):
altered = json.dumps({**json.loads(raw), **changed}).encode()
monkeypatch.setattr(recovery, "_read", lambda *a, **kw: altered)
monkeypatch.setattr(recovery, "PRIOR_SHA", hashlib.sha256(altered).hexdigest())
with pytest.raises(recovery.RecoveryRejected, match="prior_receipt_changed"):
recovery._prior_receipt()
def test_postclaim_runtime_change_refuses_before_ensure(recovery, owned, monkeypatch):
values, _, _ = owned
def consume(_):
values[8]["Id"] = "f" * 64
monkeypatch.setattr(recovery, "_claim", consume)
result = recovery.run(execute=True, confirmation=recovery.CONFIRMATION)
assert result["status"] == "unknown" and result["invocationConsumed"] is True
assert "ensure" not in values[9]
def test_once_claim_cannot_be_replayed(recovery, owned, monkeypatch):
values, _, _ = owned
def consume(_):
recovery.CLAIM.write_text("consumed")
monkeypatch.setattr(recovery, "_claim", consume)
monkeypatch.setattr(values[0], "_rollback_before_activation_locked", lambda journal, barrier: False)
assert recovery.run(execute=True, confirmation=recovery.CONFIRMATION)["status"] == "unknown"
result = recovery.run(execute=True, confirmation=recovery.CONFIRMATION)
assert result["status"] == "blocked" and result["reason"] == "invocation_consumed_inspect_only"
def test_claim_write_uncertainty_is_unknown(recovery, owned, monkeypatch):
def uncertain(_):
raise OSError("secret-canary")
monkeypatch.setattr(recovery, "_claim", uncertain)
result = recovery.run(execute=True, confirmation=recovery.CONFIRMATION)
assert result["status"] == "unknown" and result["invocationConsumed"] is None
assert result["recoveryEffectsPerformed"] is None and "secret-canary" not in json.dumps(result)
assert "ensure" not in owned[0][9]
def test_all_runtime_ids_are_full_immutable_ids(recovery):
assert all(len(cid) == 64 and set(cid) <= set("0123456789abcdef") for cid in recovery.RUNTIME_IDS.values())
assert recovery.CANDIDATE_SHA == recovery.FLEET_SHA
assert recovery.CLAIM != recovery.PRIOR_CLAIM
@pytest.mark.skipif(sys.platform == "win32", reason="native POSIX ownership required")
def test_native_existing_migration_lock_rejects_missing_and_retains_lock(recovery):
import os
import tempfile
from types import SimpleNamespace
if os.geteuid() != 0:
pytest.skip("native root ownership required")
parent = Path("/data") if Path("/data").is_dir() else Path("/root")
recovery._parents(parent / "native-test-root")
with tempfile.TemporaryDirectory(prefix="appops-tail-lock-test-", dir=parent) as temporary:
root = Path(temporary)
root.chmod(0o700)
fleet = SimpleNamespace(MIGRATION_JOURNAL_PATH=root / "journal.json")
with pytest.raises(FileNotFoundError):
with recovery._migration_existing_lock(fleet):
pytest.fail("missing lock admitted")
lock = root / "specialization.lock"
lock.write_bytes(b"")
lock.chmod(0o600)
with recovery._migration_existing_lock(fleet) as barrier:
barrier()
with pytest.raises(recovery.RecoveryRejected, match="migration_lock_busy"):
with recovery._migration_existing_lock(fleet):
pytest.fail("second lock acquired")
assert lock.exists()
@pytest.mark.skipif(sys.platform == "win32", reason="native POSIX ownership required")
def test_native_untrusted_parent_refused_before_missing_lock(recovery):
import os
import tempfile
from types import SimpleNamespace
if os.geteuid() != 0:
pytest.skip("native root ownership required")
parent = Path("/data") if Path("/data").is_dir() else Path("/root")
recovery._parents(parent / "native-test-root")
with tempfile.TemporaryDirectory(prefix="appops-tail-untrusted-test-", dir=parent) as temporary:
root = Path(temporary)
root.chmod(0o777)
fleet = SimpleNamespace(MIGRATION_JOURNAL_PATH=root / "journal.json")
with pytest.raises(recovery.RecoveryRejected, match="root_path_untrusted"):
with recovery._migration_existing_lock(fleet):
pytest.fail("untrusted parent admitted")
def test_tail_uses_captured_legacy_tcp_not_specialized_unix(recovery, owned, monkeypatch):
values, _, _ = owned
fleet, _, _, driver, *_ = values
seen = []
def inventory(self, args, **kwargs):
assert args[3] == "DOCKER_HOST=tcp://127.0.0.1:2375"
assert "unix://" not in " ".join(args)
seen.append(args[3])
return ""
monkeypatch.setattr(driver, "_run", inventory)
monkeypatch.setattr(fleet, "_DIND_DOCKER_HOST", "unix:///run/user/1000/docker.sock")
assert recovery.run()["status"] == "ready"
assert seen == ["DOCKER_HOST=tcp://127.0.0.1:2375"] * 2