# CorpusFetch Method Slice (Zero-Prose Python Stdlib Reference) # Canonical Residual: https://corpusfetch.com/c/what-window-commit-invariants-must-survive-partial-state-checkpoints-in-stateful-stream-processing.md # Remnant Warrant: https://corpusfetch.com/c/what-window-commit-invariants-must-survive-partial-state-checkpoints-in-stateful-stream-processing.warrant.json # Markdown SHA-256: 17c96ec6a614f2fdaf3ff4128f9b64127bc1df0c39856a9600a5bfbd61b19088 # Code AST SHA-256: b40f37bac25a8fdda6236b952b4f702c27a87e25c2dd11db0fab5708b3620643 # Verified Classes: CheckpointVerdict, WindowCheckpoint, ResumePolicy, CheckpointManifest, WindowCheckpointGuard # Defensive Guards: 0 fail-closed invariants (zero external deps) from dataclasses import dataclass from enum import Enum, auto import hashlib import json class CheckpointVerdict(Enum): ATOMIC_CHECKPOINT = auto() PARTIAL_CHECKPOINT_REJECT = auto() INCONSISTENT_STATE = auto() @dataclass(frozen=True) class WindowCheckpoint: window_id: str watermark_epoch: float aggregation_state_hash: str output_emission_receipt: str | None atomicity_signature: str checkpoint_epoch: float @dataclass(frozen=True) class ResumePolicy: require_output_emission_receipt: bool max_checkpoint_age_sec: float current_epoch: float @dataclass(frozen=True) class CheckpointManifest: verdict: CheckpointVerdict resume_watermark: float | None directive: str manifest_digest: str class WindowCheckpointGuard: @staticmethod def _hash(data: str) -> str: return hashlib.sha256(data.encode("utf-8")).hexdigest() @classmethod def evaluate_checkpoint( cls, cp: WindowCheckpoint, policy: ResumePolicy ) -> CheckpointManifest: # Gate 1: Atomicity signature presence if not cp.atomicity_signature: return CheckpointManifest( verdict=CheckpointVerdict.PARTIAL_CHECKPOINT_REJECT, resume_watermark=None, directive=f"HALT: Checkpoint for window {cp.window_id} lacks atomicity signature. Fall back to prior checkpoint.", manifest_digest=cls._hash(f"NO_ATOMIC:{cp.window_id}") ) # Gate 2: Output emission receipt if policy.require_output_emission_receipt and not cp.output_emission_receipt: return CheckpointManifest( verdict=CheckpointVerdict.PARTIAL_CHECKPOINT_REJECT, resume_watermark=None, directive="HALT: Output emission receipt missing. Cannot determine if window result was emitted.", manifest_digest=cls._hash(f"NO_EMISSION:{cp.window_id}") ) # Gate 3: Freshness age = policy.current_epoch - cp.checkpoint_epoch if age > policy.max_checkpoint_age_sec: return CheckpointManifest( verdict=CheckpointVerdict.PARTIAL_CHECKPOINT_REJECT, resume_watermark=None, directive=f"HALT: Checkpoint is {age:.0f}s old, exceeding {policy.max_checkpoint_age_sec}s freshness limit.", manifest_digest=cls._hash(f"STALE:{cp.window_id}") ) # Gate 4: Verified digest = cls._hash(json.dumps({ "window": cp.window_id, "watermark": str(cp.watermark_epoch), "aggregation": cp.aggregation_state_hash, "atomicity": cp.atomicity_signature }, sort_keys=True)) return CheckpointManifest( verdict=CheckpointVerdict.ATOMIC_CHECKPOINT, resume_watermark=cp.watermark_epoch, directive=f"PROCEED: Resume window at watermark {cp.watermark_epoch}.", manifest_digest=digest )