Module changelog 

Change lifecycle tracking and durable write-ahead log.

When a storage mutation spans both the high-volume (HV) and long-term (LT) backends, several non-atomic steps must happen in sequence: uploading to LT, committing a tombstone in HV via compare-and-swap, and cleaning up unreferenced blobs. A crash at any point can leave orphaned LT blobs.

This module provides two layers of protection:

1. In-process tracking — ChangeGuard is an RAII guard that tracks the current ChangePhase of an operation. When dropped, it spawns a background task to clean up whichever blob is unreferenced based on the phase reached before the drop. This handles normal errors and early returns within a running process.

2. Durable write-ahead log — The ChangeLog trait records a Change to durable storage before any LT side effects begin. If the process crashes, a recovery scan reads outstanding entries and cleans up orphaned blobs. Recovery is garbage collection — it never replays CAS mutations or finishes incomplete operations.

Structs

Change

Describes the LT blobs involved in a multi-step storage change.

ChangeGuard

RAII guard that tracks cleanup state for a multi-step storage change.

ChangeId

Unique identifier for a change log entry.

ChangeManager

Manager for multi-step storage changes, including backends and durable log.

InMemoryChangeLog

In-memory ChangeLog for tests and deployments without durable logging.

NoopChangeLog

ChangeLog implementation that discards all entries.

Enums

ChangePhase

Phase of a multi-step storage change.

Traits

ChangeLog

Durable write-ahead log for multi-step storage changes.