Implement durable module recovery operations
This commit is contained in:
@@ -0,0 +1,182 @@
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import datetime, timedelta, timezone
|
||||
|
||||
from sqlalchemy import create_engine, select
|
||||
from sqlalchemy.orm import sessionmaker
|
||||
import pytest
|
||||
|
||||
from govoplan_core.core.recovery import (
|
||||
RecoveryCheckpoint,
|
||||
RecoveryGuaranteeError,
|
||||
RecoveryMode,
|
||||
RecoveryOperation,
|
||||
RecoveryPlan,
|
||||
RecoveryStatus,
|
||||
verify_recovery_evidence_chain,
|
||||
)
|
||||
from govoplan_core.core.recovery_runtime import (
|
||||
RecoveryOperationBusy,
|
||||
RecoveryOperationStateConflict,
|
||||
begin_durable_recovery_operation,
|
||||
claim_durable_recovery_operation,
|
||||
)
|
||||
from govoplan_core.core.runtime_coordination import DistributedLease, RuntimeIdentity
|
||||
from govoplan_core.db.base import Base
|
||||
|
||||
|
||||
def _fixture():
|
||||
engine = create_engine("sqlite+pysqlite:///:memory:")
|
||||
Base.metadata.create_all(
|
||||
engine,
|
||||
tables=[
|
||||
DistributedLease.__table__,
|
||||
RecoveryOperation.__table__,
|
||||
RecoveryCheckpoint.__table__,
|
||||
],
|
||||
)
|
||||
return engine, sessionmaker(bind=engine, expire_on_commit=False)
|
||||
|
||||
|
||||
def _identity(node: str, incarnation: str) -> RuntimeIdentity:
|
||||
return RuntimeIdentity(
|
||||
installation_id="installation-1",
|
||||
node_id=node,
|
||||
incarnation=incarnation,
|
||||
role="worker",
|
||||
software_version="test",
|
||||
composition_hash="a" * 64,
|
||||
)
|
||||
|
||||
|
||||
def _start(factory, identity, *, key: str = "build-1"):
|
||||
return begin_durable_recovery_operation(
|
||||
factory,
|
||||
identity=identity,
|
||||
module_id="campaigns",
|
||||
operation_type="build-artifacts",
|
||||
idempotency_key=key,
|
||||
request={"version_id": "version-1", "write_eml": True},
|
||||
recovery_plan=RecoveryPlan(
|
||||
mode=RecoveryMode.COMPENSATION,
|
||||
preconditions=("validated version is locked",),
|
||||
compensation_steps=("delete build object prefix",),
|
||||
verification_steps=("compare database and object manifests",),
|
||||
),
|
||||
precondition_evidence={"validation_sha256": "b" * 64},
|
||||
lease_resource_key="campaign:build:version-1",
|
||||
resource_type="campaign_version",
|
||||
resource_id="version-1",
|
||||
)
|
||||
|
||||
|
||||
def test_durable_operation_commits_before_caller_effect_and_replays_success() -> None:
|
||||
engine, factory = _fixture()
|
||||
try:
|
||||
started = _start(factory, _identity("worker-1", "incarnation-1"))
|
||||
assert started.operation is not None
|
||||
with factory() as session:
|
||||
persisted = session.get(RecoveryOperation, started.operation_id)
|
||||
assert persisted is not None
|
||||
assert persisted.status == RecoveryStatus.RUNNING.value
|
||||
assert persisted.checkpoint_count == 3
|
||||
|
||||
started.operation.checkpoint(
|
||||
kind="object-prefix-reserved",
|
||||
summary="Build object prefix reserved",
|
||||
evidence={"prefix": "campaign-artifacts/build-1/"},
|
||||
)
|
||||
started.operation.succeed(
|
||||
evidence={
|
||||
"verified": True,
|
||||
"checks": {"database_manifest": "matched", "object_manifest": "matched"},
|
||||
}
|
||||
)
|
||||
|
||||
replay = _start(factory, _identity("worker-2", "incarnation-2"))
|
||||
assert replay.replayed is True
|
||||
assert replay.operation is None
|
||||
with factory() as session:
|
||||
assert verify_recovery_evidence_chain(session, started.operation_id)
|
||||
finally:
|
||||
engine.dispose()
|
||||
|
||||
|
||||
def test_same_fence_cannot_start_duplicate_running_operation() -> None:
|
||||
engine, factory = _fixture()
|
||||
try:
|
||||
identity = _identity("worker-1", "incarnation-1")
|
||||
started = _start(factory, identity)
|
||||
with pytest.raises(RecoveryOperationStateConflict, match="already running"):
|
||||
_start(factory, identity)
|
||||
assert started.operation is not None
|
||||
started.operation.release_unresolved()
|
||||
finally:
|
||||
engine.dispose()
|
||||
|
||||
|
||||
def test_other_runtime_cannot_use_an_active_fence() -> None:
|
||||
engine, factory = _fixture()
|
||||
try:
|
||||
started = _start(factory, _identity("worker-1", "incarnation-1"))
|
||||
with pytest.raises(RecoveryOperationBusy):
|
||||
_start(factory, _identity("worker-2", "incarnation-2"), key="build-2")
|
||||
assert started.operation is not None
|
||||
started.operation.release_unresolved()
|
||||
finally:
|
||||
engine.dispose()
|
||||
|
||||
|
||||
def test_expired_crash_fence_is_taken_over_as_recovery_required() -> None:
|
||||
engine, factory = _fixture()
|
||||
try:
|
||||
started = _start(factory, _identity("worker-1", "incarnation-1"))
|
||||
assert started.operation is not None
|
||||
with factory() as session:
|
||||
lease = session.execute(select(DistributedLease)).scalar_one()
|
||||
lease.expires_at = datetime.now(timezone.utc) - timedelta(seconds=1)
|
||||
session.add(lease)
|
||||
session.commit()
|
||||
|
||||
recovery = claim_durable_recovery_operation(
|
||||
factory,
|
||||
identity=_identity("worker-2", "incarnation-2"),
|
||||
operation_id=started.operation_id,
|
||||
)
|
||||
with factory() as session:
|
||||
operation = session.get(RecoveryOperation, started.operation_id)
|
||||
assert operation is not None
|
||||
assert operation.status == RecoveryStatus.RECOVERY_REQUIRED.value
|
||||
assert operation.fencing_token == 2
|
||||
assert verify_recovery_evidence_chain(session, operation.id)
|
||||
recovery.compensate(
|
||||
failure_summary="worker stopped during object publication",
|
||||
failure_evidence={"object_prefix": "campaign-artifacts/build-1/"},
|
||||
recovery_evidence={
|
||||
"verified": True,
|
||||
"checks": {"object_prefix_empty": True},
|
||||
},
|
||||
)
|
||||
finally:
|
||||
engine.dispose()
|
||||
|
||||
|
||||
def test_tampered_checkpoint_blocks_verified_success() -> None:
|
||||
engine, factory = _fixture()
|
||||
try:
|
||||
started = _start(factory, _identity("worker-1", "incarnation-1"))
|
||||
assert started.operation is not None
|
||||
with factory() as session:
|
||||
checkpoint = session.execute(
|
||||
select(RecoveryCheckpoint).order_by(RecoveryCheckpoint.sequence)
|
||||
).scalars().first()
|
||||
assert checkpoint is not None
|
||||
checkpoint.summary = "tampered"
|
||||
session.add(checkpoint)
|
||||
session.commit()
|
||||
with pytest.raises(RecoveryGuaranteeError, match="chain verification failed"):
|
||||
started.operation.succeed(
|
||||
evidence={"verified": True, "checks": {"objects": "matched"}}
|
||||
)
|
||||
finally:
|
||||
engine.dispose()
|
||||
Reference in New Issue
Block a user