BUILDING REPUTATION-POOL
Building reputation-pool
An evidence-led backend engineering series from a production problem to a JDK-only core, concurrency contracts, open-core boundaries, multi-tenancy, and horizontal scaling.
What this series verifies
This is not a retrospective tour of a finished architecture. Each article connects the initial assumption, the test or operational constraint that broke it, the revised contract, and the remaining limitation to real code and pull requests.
- Part 1 · Lincheck Published
Our 32-thread stress test passed. The specification was still wrong.
How Lincheck traces rewrote the contracts for fencing tokens, resource selection, and block races.
- Part 2 · Open-core boundary Published
Separating the core from SaaS revealed a missing boundary
Why the copied gRPC surface became a public adapter module shared by two different hosts.
- Part 3 · Multi-tenancy Planned
Isolation and noisy-neighbor control
Tenant-scoped state and a shared-JVM resource budget.
- Part 4 · Origin Planned
Why production needed a reputation pool
The operational failure that required resource history and recovery rather than a static proxy list.
- Part 5 · Horizontal scaling Published
Where the single-JVM contract breaks
The boundary of local atomicity and the direction for distributed leases and fencing.
- Part 6 · Shadow mode Awaiting data
Validating decisions with production traffic
A baseline-to-shadow comparison after enough real production data exists.