[lxy-9602]

Purpose

Linked issue: #158

Expose the snapshot used by each partition-bucket writer, since lazy restoration can select different snapshots for different buckets.

Initialize the value from the actual restore result and advance it after successful real-time refresh. Partial refresh failures retain the previous value for failed buckets. The getter performs no storage I/O.

Tests

Added or extended coverage for:

• Lazy restoration across snapshots and partitions, empty tables, ignored previous files, and missing writers.
• Real-time refresh failures, retries, and shared contexts.
• Partial bucket failures, concurrent refreshes, and snapshots with unchanged offsets.
• Compaction refresh failures and retries.
• Format-table rejection and deduplicate refresh preparation failures.

API and Format

Add the pure virtual API:

virtual Result<std::optional<int64_t>> GetBaseSnapshotId(
    const std::map<std::string, std::string>& partition, int32_t bucket) const = 0;

Returns nullopt when no snapshot has been used, and an error for missing bucket writers or format tables.

No storage format or protocol changes.

Documentation

Generative AI tooling

Generated-by: OpenAI Codex (GPT-6)

[zjw1111]

+1

[SteNicholas]

LGTM.