Deep Review

Workflow

1. Confirm inputs

   • Identify the repository root, target output folder, and output filename.
   • If the review scope is not explicit, use the current branch diff against origin/HEAD.

2. Collect the change set

   • Prefer the diff provided by the user.
   • Otherwise run git diff origin/HEAD...HEAD from the repository root.
   • If origin/HEAD is unavailable or the diff command fails, stop and ask for guidance.

3. Understand the intent

   • Read the PR description, issue link, commit messages, or nearby code comments when available.
   • State the concrete system problem the change is trying to solve.
   • If intent is still unclear, infer from code and label the inference as an assumption.

4. Read the affected TiKV subsystems

   • Follow changed code into the owning modules, not just the diff hunk.
   • Typical subsystems include:
     • src/storage, src/storage/mvcc, src/storage/txn
     • src/server
     • components/raftstore, components/raftstore-v2
     • components/cdc
     • components/pd_client
     • tests/
   • Read enough nearby code to understand invariants, concurrency assumptions, error propagation, and test coverage.

5. Explain how the change works

   • Walk through each non-trivial logic change.
   • Explain control flow, data flow, state transitions, and failure paths in plain language.
   • Reference concrete files, symbols, and lines when discussing findings.

6. Evaluate costs and negative impacts

   • Explicitly assess:
     • correctness
     • security
     • robustness and failure modes
     • compatibility and behavioral shifts
     • CPU cost
     • memory cost
     • log volume or log-signal quality
     • operability and debuggability
     • cognitive load and maintainability

7. Run TiKV static validation from ./Makefile

   • Run make format from the repository root.
   • Then run make clippy from the repository root.
   • Treat the behavior as part of the review:

Review Output Template

### Deep Review

#### Problem Summary
- [Explain the concrete problem the change targets]

#### Solution Walkthrough
- [Explain how the change solves the problem; cover all non-obvious logic]

#### Findings (ordered by severity)
- [Issue or risk with file/line references]

#### Costs and Negative Impacts
- Correctness:
- Security:
- Compatibility:
- Robustness:
- Operability:
- Cognitive Load:
- CPU:
- Memory:
- Log Volume:

#### Static Validation
- `make format`:
- `make clippy`:

#### Engineering Rules Check
- [List code or process mismatches against `AGENTS.md`, or "None"]

#### Questions and Assumptions
- [List unknowns or assumptions made]

#### Suggested Tests / Validation
- [Targeted tests or checks to validate behavior]

• make format runs pre-format, which installs rustfmt and bootstraps the pinned cargo-sort version before running cargo fmt and cargo sort
• make clippy runs repository checks including redact-log, log-style, dashboards, docker-build, license, deny, and finally scripts/clippy-all