[Aionw]

Description

Follow-up to #4406. That PR removed the Master-generated snapshot and restore path, but the serialization code only that path called is still compiled into the master. None of it has a caller outside its own tests; this PR deletes it.

Removed:

• serialize/serializer.{h,cpp}: the msgpack Serializer<T> specializations for allocators, buffers, replicas and mounted segments, plus SerializationHelper and SerializationError.
• ha/snapshot: StoreResourceSnapshotCodec, AllocatorSnapshotCodec, LocalSsdCodec.
• SegmentSerializer and SegmentView (segment.h); TaskManagerSerializer together with restore_task/clear_all, which only it used; LocalSsdManager::ExportPersistedState/RestorePersistedState and LocalSsdPersistedState.
• The detached snapshot path from [Store] Add fork-safe SegmentPool snapshot codecs #3706: SegmentPool::CaptureSnapshot/RestoreSnapshot, OffsetBufferAllocator and OffsetAllocator CaptureSnapshot/Restore with their snapshot structs and layout validation, AdoptMode::ReplacePool, IsKnownSegmentStatus.
• The tests that only covered the above.

Kept on purpose:

• SegmentPool::WriteAccess::PrepareAdopt, back to its pre-[Store] Add fork-safe SegmentPool snapshot codecs #3706 signature: AllocationKeepsReadLockAcrossAllocatorCall uses it to install a blocking allocator.
• SegmentPoolTest.DestroyedPoolAndRetainedAllocatorRemoveMetricLabels, rewritten to mount directly instead of restoring a snapshot. [Store] Add fork-safe SegmentPool snapshot codecs #3706 also made the buffer allocator destructors remove their segment's metric labels and made remove_segment_metrics keep a label that still has usage. Both apply to ordinary mounts and had no other test.
• The byte serializer in include/serializer.h (serialize_to/deserialize_from), which storage_backend.cpp still uses.
• Error codes such as SERIALIZE_FAIL/DESERIALIZE_FAIL, which are part of the public ErrorCode enum.

The only non-deletion change is in OffsetBufferAllocator's constructor: its node cap goes back to the literal 64 * 1024 * 1024 it had before #3706 renamed it OffsetAllocatorSnapshot::kMaxNodes (same value, 1u << 26).

Overlap: #3707 (SegmentPool cutover) adapts several of these files to SegmentPool; after this lands, those hunks resolve as deletions. #4133 (stacked on #3707) removes the legacy SegmentManager itself; this PR only removes its snapshot serializer, which nothing calls on main today.

Suggested Merge Priority

• P0 — Urgent
• P1 — High
• P2 — Normal
• P3 — Low: Experimental feature, optional enhancement, refactoring, or cleanup with no urgent user impact

Module

• Mooncake Store (mooncake-store)

Type of Change

• Refactor

How Has This Been Tested?

Release build, BUILD_UNIT_TESTS=ON, USE_NOF=OFF.

Test commands:

cmake --build build -j14            # full build
ctest --test-dir build -j12 --timeout 300 \
  -E '^(ordered_oplog_writer_test|tcp_transport_test|transfer_metadata_test)$'
pre-commit run --files $(git diff --name-only --diff-filter=ACMR origin/main...HEAD)

Test results:

• Unit tests pass: 243/244. The one failure, DfsSyncClientTest.DfsOnlyBatchFailureOnlyRevokesFailedKey, fails the same way on unmodified origin/main.
• The full build has two failures that also happen on origin/main and are unrelated: tests/e2e/chaosctl.cpp (sleep undeclared) and ordered_oplog_writer_test (libaio link). The three excluded tests need a network/RDMA environment.
• cmake-format was skipped: it rewrites unrelated, pre-existing lines in tests/CMakeLists.txt; this PR only deletes lines there.

Checklist

• I have performed a self-review of my own code
• I have formatted my code using ./scripts/code_format.sh (clang-format 20 on changed lines)
• I have run pre-commit on the files changed in this PR and all hooks pass (except cmake-format, see above)
• I have updated the documentation (if applicable) — no docs reference the removed code
• I have added tests to prove my changes are effective — removal only
• For changes >500 LOC: I have filed an RFC issue — ~5.9k lines, all deletions of code with no production caller; no behavior change

AI Assistance Disclosure

• AI tools were used (specify below)

Claude Code traced the callers, made the deletions and ran the build and tests. The human submitter is responsible for reviewing every changed line and defending the change.

[Icedcoco]

I noticed that serializer_test.cpp also includes tests for the generic byte serializer in include/serializer.h, which remains in use by storage_backend.cpp. Would it be worth considering whether those tests should be kept in a separate target, since they cover code that this PR leaves in place?

[Icedcoco]

LGTM pending CI