Files
aios/active-runtime/app/task_plane/governance_recovery_controller.py
T

475 lines
17 KiB
Python

"""Event-driven governed recovery reconciler with polling fallback.
The controller cannot create a task and cannot bypass admission. It resumes
only an exact, already-existing child through an injected governed executor.
"""
from __future__ import annotations
from dataclasses import dataclass
from datetime import datetime, timezone
import hashlib
import json
from typing import Mapping, Protocol
from .governance_admission import (
AdmissionOperation,
AdmissionPreflight,
GovernanceFacts,
)
from .governance_checkpoint import (
CheckpointStore,
RecoveryCheckpoint,
RetryBudget,
)
from .governance_resolver import ExactTaskResolver, ResolutionQuery, ResolutionState
def _canonical(value) -> str:
return json.dumps(
value, sort_keys=True, separators=(",", ":"), ensure_ascii=False
)
def _digest(value) -> str:
return hashlib.sha256(_canonical(value).encode("utf-8")).hexdigest()
@dataclass(frozen=True)
class GovernanceEvent:
kind: str
revision: str
occurred_at: str
@dataclass(frozen=True)
class RecoveryReceipt:
receipt_ref: str
task_id: str
accepted: bool
@dataclass(frozen=True)
class ReconcileResult:
status: str
next_legal_action: str
task_id: str | None
blocker_code: str | None
mechanism_repair_required: bool
receipt_ref: str | None = None
class GovernanceFactsProvider(Protocol):
def snapshot(
self,
*,
task_id: str | None,
actor_identity: str,
operation: AdmissionOperation,
) -> GovernanceFacts: ...
class ExistingChildRecoveryExecutor(Protocol):
def recover_existing_child(
self,
*,
task_id: str,
actor_identity: str,
idempotency_key: str,
facts: GovernanceFacts,
) -> RecoveryReceipt: ...
# Optional at runtime. If implemented, this is preferred after an
# unknown-result write so the controller reconciles before re-executing.
def reconcile_recovery_result(
self,
*,
task_id: str,
actor_identity: str,
idempotency_key: str,
) -> RecoveryReceipt | None: ...
class GovernanceRecoveryController:
def __init__(
self,
*,
goal_id: str,
query: ResolutionQuery,
actor_identity: str,
resolver: ExactTaskResolver,
facts_provider: GovernanceFactsProvider,
executor: ExistingChildRecoveryExecutor,
checkpoint_store: CheckpointStore,
retry_budget: RetryBudget = RetryBudget(),
) -> None:
self.goal_id = goal_id
self.query = query
self.actor_identity = actor_identity
self.resolver = resolver
self.facts_provider = facts_provider
self.executor = executor
self.checkpoints = checkpoint_store
self.retry_budget = retry_budget
self.preflight = AdmissionPreflight()
def _timestamp(self, event: GovernanceEvent) -> str:
return event.occurred_at or datetime.now(timezone.utc).isoformat()
def _save_blocked(
self,
*,
event: GovernanceEvent,
task_id: str | None,
mission_id: str | None,
root_task_id: str | None,
parent_task_id: str | None,
principal_id: str | None,
blocker_code: str,
blocker_details: tuple[str, ...],
) -> ReconcileResult:
previous = self.checkpoints.load()
fingerprint = self.retry_budget.blocker_fingerprint(
blocker_code, blocker_details
)
cycles, repair = self.retry_budget.observe(
previous=previous,
blocker_code=blocker_code,
blocker_fingerprint=fingerprint,
)
next_action = "REPAIR_MECHANISM" if repair else "WAIT_DEPENDENCY"
checkpoint = RecoveryCheckpoint(
schema="aios.governance-recovery-checkpoint.v1",
goal_id=self.goal_id,
repository=self.query.repository,
source_oid=self.query.base_oid,
candidate_digest=(
self.query.patch_digest
or self.query.artifact_sha256
or self.query.candidate_oid
or ""
),
task_id=task_id,
mission_id=mission_id,
root_task_id=root_task_id,
parent_task_id=parent_task_id,
principal_id=principal_id,
blocker_code=blocker_code,
blocker_fingerprint=fingerprint,
next_legal_action=next_action,
last_verified_at=self._timestamp(event),
evidence_refs=(f"event:{event.kind}:{event.revision}",),
unchanged_cycles=cycles,
mechanism_repair_required=repair,
)
self.checkpoints.save(checkpoint)
return ReconcileResult(
status="BLOCKED",
next_legal_action=next_action,
task_id=task_id,
blocker_code=blocker_code,
mechanism_repair_required=repair,
)
def _candidate_digest(self) -> str:
return (
self.query.patch_digest
or self.query.artifact_sha256
or self.query.candidate_oid
or ""
)
def _recovery_context_digest(
self, *, task_id: str, facts: GovernanceFacts
) -> str:
return _digest({
"task_id": task_id,
"repository": self.query.repository,
"candidate": self._candidate_digest(),
"principal": facts.authenticated_principal_id,
"branch": facts.target_branch,
"branch_oid": facts.expected_branch_oid,
"generation": facts.generation,
"task_lease_id": facts.task_lease_id,
"task_principal_id": facts.task_principal_id,
"writer_lease_id": facts.writer_lease_id,
"acceptance_revision": facts.acceptance_revision,
"file_allowlist": sorted(facts.accepted_file_allowlist),
})
def _finish_recovered(
self,
*,
event: GovernanceEvent,
resolved,
facts: GovernanceFacts,
idempotency_key: str,
context_digest: str,
receipt: RecoveryReceipt,
) -> ReconcileResult:
checkpoint = RecoveryCheckpoint(
schema="aios.governance-recovery-checkpoint.v1",
goal_id=self.goal_id,
repository=self.query.repository,
source_oid=self.query.base_oid,
candidate_digest=self._candidate_digest(),
task_id=resolved.task_id,
mission_id=resolved.mission_id,
root_task_id=resolved.root_task_id,
parent_task_id=resolved.parent_task_id,
principal_id=facts.authenticated_principal_id,
blocker_code=None,
blocker_fingerprint=None,
next_legal_action="ROOT_REVIEW",
last_verified_at=self._timestamp(event),
evidence_refs=(
f"event:{event.kind}:{event.revision}",
receipt.receipt_ref,
),
unchanged_cycles=0,
mechanism_repair_required=False,
recovery_idempotency_key=idempotency_key,
recovery_receipt_ref=receipt.receipt_ref,
recovery_context_digest=context_digest,
)
self.checkpoints.save(checkpoint)
return ReconcileResult(
status="RECOVERED",
next_legal_action="ROOT_REVIEW",
task_id=resolved.task_id,
blocker_code=None,
mechanism_repair_required=False,
receipt_ref=receipt.receipt_ref,
)
def reconcile(self, event: GovernanceEvent) -> ReconcileResult:
resolution = self.resolver.resolve(
self.query, actor_identity=self.actor_identity
)
if resolution.state is not ResolutionState.RESOLVED:
code = {
ResolutionState.NOT_FOUND: "TASK_NOT_RESOLVED",
ResolutionState.AMBIGUOUS: "TASK_RESOLUTION_AMBIGUOUS",
ResolutionState.INVALID_QUERY: "TASK_RESOLUTION_INVALID",
}[resolution.state]
details = (
resolution.detail,
*resolution.candidate_task_ids,
)
return self._save_blocked(
event=event,
task_id=None,
mission_id=None,
root_task_id=None,
parent_task_id=None,
principal_id=None,
blocker_code=code,
blocker_details=tuple(str(v) for v in details if v),
)
resolved = resolution.resolved
assert resolved is not None
facts = self.facts_provider.snapshot(
task_id=resolved.task_id,
actor_identity=self.actor_identity,
operation=AdmissionOperation.RECOVER_EXISTING_CHILD,
)
admission = self.preflight.evaluate(
resolution=resolution,
facts=facts,
operation=AdmissionOperation.RECOVER_EXISTING_CHILD,
)
if not admission.ready:
primary = admission.blockers[0]
details = tuple(
f"{blocker.code}:{blocker.detail}"
for blocker in admission.blockers
)
return self._save_blocked(
event=event,
task_id=resolved.task_id,
mission_id=resolved.mission_id,
root_task_id=resolved.root_task_id,
parent_task_id=resolved.parent_task_id,
principal_id=facts.authenticated_principal_id,
blocker_code=primary.code,
blocker_details=details,
)
context_digest = self._recovery_context_digest(
task_id=resolved.task_id, facts=facts
)
previous = self.checkpoints.load()
candidate_digest = self._candidate_digest()
unknown_previous = (
previous is not None
and previous.blocker_code == "RECOVERY_RESULT_UNKNOWN"
and previous.task_id == resolved.task_id
and previous.candidate_digest == candidate_digest
and bool(previous.recovery_idempotency_key)
)
if unknown_previous:
idempotency_key = previous.recovery_idempotency_key
reconcile = getattr(
self.executor, "reconcile_recovery_result", None
)
if callable(reconcile):
receipt = reconcile(
task_id=resolved.task_id,
actor_identity=self.actor_identity,
idempotency_key=idempotency_key,
)
if receipt is not None:
if receipt.accepted:
return self._finish_recovered(
event=event,
resolved=resolved,
facts=facts,
idempotency_key=idempotency_key,
context_digest=(
previous.recovery_context_digest
or context_digest
),
receipt=receipt,
)
return self._save_blocked(
event=event,
task_id=resolved.task_id,
mission_id=resolved.mission_id,
root_task_id=resolved.root_task_id,
parent_task_id=resolved.parent_task_id,
principal_id=facts.authenticated_principal_id,
blocker_code="RECOVERY_NOT_ACCEPTED",
blocker_details=(receipt.receipt_ref,),
)
if (
previous.recovery_context_digest is not None
and previous.recovery_context_digest != context_digest
):
checkpoint = RecoveryCheckpoint(
schema="aios.governance-recovery-checkpoint.v1",
goal_id=self.goal_id,
repository=self.query.repository,
source_oid=self.query.base_oid,
candidate_digest=candidate_digest,
task_id=resolved.task_id,
mission_id=resolved.mission_id,
root_task_id=resolved.root_task_id,
parent_task_id=resolved.parent_task_id,
principal_id=facts.authenticated_principal_id,
blocker_code="RECOVERY_RESULT_UNKNOWN",
blocker_fingerprint=_digest({
"task_id": resolved.task_id,
"idempotency_key": idempotency_key,
"context_changed": True,
}),
next_legal_action="RECOVER_EXISTING_CHILD",
last_verified_at=self._timestamp(event),
evidence_refs=(f"event:{event.kind}:{event.revision}",),
unchanged_cycles=1,
mechanism_repair_required=False,
recovery_idempotency_key=idempotency_key,
recovery_context_digest=(
previous.recovery_context_digest
),
)
self.checkpoints.save(checkpoint)
return ReconcileResult(
status="RESULT_UNKNOWN",
next_legal_action="RECOVER_EXISTING_CHILD",
task_id=resolved.task_id,
blocker_code="RECOVERY_RESULT_UNKNOWN",
mechanism_repair_required=False,
)
else:
idempotency_key = "recover-" + _digest({
"task_id": resolved.task_id,
"repository": self.query.repository,
"candidate": candidate_digest,
"branch_oid": facts.expected_branch_oid,
"generation": facts.generation,
"writer_lease_id": facts.writer_lease_id,
})[:40]
try:
receipt = self.executor.recover_existing_child(
task_id=resolved.task_id,
actor_identity=self.actor_identity,
idempotency_key=idempotency_key,
facts=facts,
)
except Exception:
# Unknown-result write: retain the SAME idempotency key for the next
# reconcile. Never assume the side effect did not happen.
checkpoint = RecoveryCheckpoint(
schema="aios.governance-recovery-checkpoint.v1",
goal_id=self.goal_id,
repository=self.query.repository,
source_oid=self.query.base_oid,
candidate_digest=(
self.query.patch_digest
or self.query.artifact_sha256
or self.query.candidate_oid
or ""
),
task_id=resolved.task_id,
mission_id=resolved.mission_id,
root_task_id=resolved.root_task_id,
parent_task_id=resolved.parent_task_id,
principal_id=facts.authenticated_principal_id,
blocker_code="RECOVERY_RESULT_UNKNOWN",
blocker_fingerprint=_digest({
"task_id": resolved.task_id,
"idempotency_key": idempotency_key,
}),
next_legal_action="RECOVER_EXISTING_CHILD",
last_verified_at=self._timestamp(event),
evidence_refs=(f"event:{event.kind}:{event.revision}",),
unchanged_cycles=1,
mechanism_repair_required=False,
recovery_idempotency_key=idempotency_key,
recovery_context_digest=context_digest,
)
self.checkpoints.save(checkpoint)
return ReconcileResult(
status="RESULT_UNKNOWN",
next_legal_action="RECOVER_EXISTING_CHILD",
task_id=resolved.task_id,
blocker_code="RECOVERY_RESULT_UNKNOWN",
mechanism_repair_required=False,
)
if not receipt.accepted:
return self._save_blocked(
event=event,
task_id=resolved.task_id,
mission_id=resolved.mission_id,
root_task_id=resolved.root_task_id,
parent_task_id=resolved.parent_task_id,
principal_id=facts.authenticated_principal_id,
blocker_code="RECOVERY_NOT_ACCEPTED",
blocker_details=(receipt.receipt_ref,),
)
return self._finish_recovered(
event=event,
resolved=resolved,
facts=facts,
idempotency_key=idempotency_key,
context_digest=context_digest,
receipt=receipt,
)
def on_event(self, event: GovernanceEvent) -> ReconcileResult:
# Event-driven primary path: reconcile immediately in the same call.
return self.reconcile(event)
def poll_once(self, *, revision: str, occurred_at: str) -> ReconcileResult:
# Polling is only a fallback and uses exactly the same reconcile path.
return self.reconcile(GovernanceEvent(
kind="POLL_FALLBACK",
revision=revision,
occurred_at=occurred_at,
))