gothinkster · martavoloshyn · Aug 21, 2026 · Aug 21, 2026 · Aug 28, 2026 · Aug 28, 2026
diff --git a/architecture/baseline-metrics.json b/architecture/baseline-metrics.json
@@ -0,0 +1,199 @@
+{
+  "seq": [
+    {
+      "name": "GET /tags",
+      "n": 200,
+      "mean": 3.503,
+      "p50": 3.395,
+      "p95": 5.419,
+      "p99": 7.611,
+      "min": 1.396,
+      "max": 14.686,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "GET /articles?limit=20 (anon)",
+      "n": 200,
+      "mean": 10.371,
+      "p50": 10.305,
+      "p95": 14.657,
+      "p99": 16.947,
+      "min": 3.22,
+      "max": 22.462,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "GET /articles?limit=20 (auth)",
+      "n": 200,
+      "mean": 11.09,
+      "p50": 10.809,
+      "p95": 17.332,
+      "p99": 19.97,
+      "min": 3.759,
+      "max": 22.963,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "GET /articles/{slug} (anon)",
+      "n": 200,
+      "mean": 3.987,
+      "p50": 4.121,
+      "p95": 5.621,
+      "p99": 5.826,
+      "min": 1.737,
+      "max": 6.787,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "GET /articles/{slug} (auth)",
+      "n": 200,
+      "mean": 4.854,
+      "p50": 4.512,
+      "p95": 7.485,
+      "p99": 8.093,
+      "min": 2.029,
+      "max": 15.39,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "GET /articles/feed",
+      "n": 200,
+      "mean": 9.845,
+      "p50": 9.423,
+      "p95": 13.56,
+      "p99": 17.758,
+      "min": 4.828,
+      "max": 23.265,
+      "codes": {
+        "500": 200
+      }
+    },
+    {
+      "name": "GET /profiles/bob (auth)",
+      "n": 200,
+      "mean": 4.558,
+      "p50": 4.51,
+      "p95": 6.926,
+      "p99": 7.393,
+      "min": 2.386,
+      "max": 13.947,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "GET /user",
+      "n": 200,
+      "mean": 4.243,
+      "p50": 4.227,
+      "p95": 6.462,
+      "p99": 7.072,
+      "min": 1.831,
+      "max": 7.205,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "POST /users/login",
+      "n": 50,
+      "mean": 66.946,
+      "p50": 70.937,
+      "p95": 74.841,
+      "p99": 75.814,
+      "min": 54.448,
+      "max": 75.814,
+      "codes": {
+        "200": 50
+      }
+    },
+    {
+      "name": "GET /articles/{slug}/comments",
+      "n": 200,
+      "mean": 3.416,
+      "p50": 3.295,
+      "p95": 5.012,
+      "p99": 5.534,
+      "min": 1.77,
+      "max": 5.777,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "POST /graphql { tags }",
+      "n": 200,
+      "mean": 5.536,
+      "p50": 5.116,
+      "p95": 8.053,
+      "p99": 9.428,
+      "min": 2.298,
+      "max": 49.26,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "POST /graphql articles(first:20)",
+      "n": 200,
+      "mean": 9.591,
+      "p50": 8.675,
+      "p95": 13.767,
+      "p99": 40.277,
+      "min": 3.63,
+      "max": 61.983,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "POST /graphql article(slug)",
+      "n": 200,
+      "mean": 4.71,
+      "p50": 4.359,
+      "p95": 6.937,
+      "p99": 8.2,
+      "min": 2.024,
+      "max": 58.051,
+      "codes": {
+        "200": 200
+      }
+    },
+    {
+      "name": "POST /graphql article+author+comments",
+      "n": 200,
+      "mean": 6.365,
+      "p50": 6.24,
+      "p95": 8.23,
+      "p99": 9.688,
+      "min": 3.03,
+      "max": 67.955,
+      "codes": {
+        "200": 200
+      }
+    }
+  ],
+  "concurrent_articles": {
+    "n": 200,
+    "workers": 20,
+    "wall_s": 0.1689,
+    "mean": 11.005,
+    "p50": 10.164,
+    "p95": 19.18,
+    "rps": 1184.2
+  },
+  "create_article_mean_ms": 26.601,
+  "create_article_p95_ms": 34.171,
+  "vmrss_kb": 617192,
+  "db_bytes": 110592
+}
diff --git a/architecture/baseline-report.md b/architecture/baseline-report.md
@@ -0,0 +1,184 @@
+# Pre-refactor baseline (hexagonal)
+
+Measured **2026-08-28** at git `30d3c11` (`Add hexagonal refactor plan as agent handoff docs.`).
+
+Purpose: after the work in `architecture/hexagonal-plan.md`, re-run the same commands on the same machine and compare this file (and `architecture/baseline-metrics.json`). Hexagonal refactor is **behavior-preserving**; latency and test counts should stay in the same band unless you change logging, JVM, or data volume.
+
+## How to re-measure (keep these knobs identical)
+
+| Knob | Baseline value |
+| --- | --- |
+| JDK | Temurin **17.0.20.1** (project `sourceCompatibility` is 11; Gradle 7.4 cannot run on JDK 25) |
+| Gradle | Wrapper **7.4**, `./gradlew test bootJar --no-daemon` |
+| HTTP | `java -jar build/libs/*SNAPSHOT.jar --server.port=18080 --spring.datasource.url=jdbc:sqlite:/tmp/rw-baseline.db` |
+| Dataset | 2 users (alice/bob), alice follows bob, **50** articles, 10 favorites, 5 comments on first article |
+| Bench | 20 warmup + **200** sequential requests (login: 5 + 50) |
+| Logging | `application.properties` has MyBatis mapper/read DEBUG — **this inflates HTTP times**. Do not change log levels between runs. |
+
+Warm JVM HTTP numbers below are localhost, single process, SQLite file DB, DEBUG SQL. They are for **regression**, not production SLOs.
+
+---
+
+## 1. Build and tests
+
+| Metric | Baseline |
+| --- | --- |
+| `./gradlew test bootJar --no-daemon` (cold-ish, deps already cached) | **74 s**, `BUILD SUCCESSFUL` |
+| JUnit 5 test methods | **68** |
+| Failures / errors / skipped | **0 / 0 / 0** |
+| Sum of JUnit suite times | **2.619 s** (excludes Spring context start shared across classes) |
+| Boot JAR | **68.9 MiB** (`72278681` bytes) |
+
+### Tests by layer (method count / summed method time)
+
+| Layer | Tests | Wall in XML |
+| --- | --- | --- |
+| `api` (`@WebMvcTest`) | 35 | 1.894 s |
+| `application` (MyBatis + SQLite via `DbTestBase`) | 13 | 0.253 s |
+| `infrastructure` | 14 | 0.135 s |
+| `core` (`ArticleTest` only) | 5 | 0.001 s |
+| `RealworldApplicationTests` (`@SpringBootTest` contextLoads) | 1 | 0.296 s |
+
+Slowest tests: `ArticleApiTest#should_read_article_success` **1.04 s** (first WebMvc context), `contextLoads` **0.30 s**, `UsersApiTest#should_login_success` **0.15 s**.
+
+### Coverage gaps (quality, not hexagon)
+
+- **No GraphQL tests** (DGS adapters are untested).
+- Almost all REST tests are **WebMvc + mocks**, not HTTP against a real Spring context.
+- **No ArchUnit** yet (plan step 1).
+- Feed success is mocked in `ListArticleApiTest`; the empty-feed SQL bug below is **not** caught.
+
+---
+
+## 2. HTTP / GraphQL performance
+
+Startup: **Started RealWorldApplication in 3.81–3.93 s** (JVM running ~4.2–4.4 s). Process RSS after bench: **~603 MiB** (`VmRSS` 617192 kB) — typical fat Spring + DGS heap, not app logic.
+
+Seed: create article mean **26.6 ms**, p95 **34.2 ms** (includes BCrypt only on user create, not each article).
+
+Sequential latency (ms), localhost, 50 articles in SQLite:
+
+| Endpoint | n | mean | p50 | p95 | p99 | HTTP |
+| --- | --- | --- | --- | --- | --- | --- |
+| GET `/tags` | 200 | 3.50 | 3.40 | 5.42 | 7.61 | 200 |
+| GET `/articles?limit=20` anon | 200 | 10.37 | 10.31 | 14.66 | 16.95 | 200 |
+| GET `/articles?limit=20` auth | 200 | 11.09 | 10.81 | 17.33 | 19.97 | 200 |
+| GET `/articles/{slug}` anon | 200 | 3.99 | 4.12 | 5.62 | 5.83 | 200 |
+| GET `/articles/{slug}` auth | 200 | 4.85 | 4.51 | 7.49 | 8.09 | 200 |
+| GET `/articles/feed` (alice follows bob, bob has 0 articles) | 200 | 9.85 | 9.42 | 13.56 | 17.76 | **500** |
+| GET `/profiles/bob` auth | 200 | 4.56 | 4.51 | 6.93 | 7.39 | 200 |
+| GET `/user` | 200 | 4.24 | 4.23 | 6.46 | 7.07 | 200 |
+| POST `/users/login` | 50 | 66.95 | 70.94 | 74.84 | 75.81 | 200 |
+| GET `/articles/{slug}/comments` | 200 | 3.42 | 3.29 | 5.01 | 5.53 | 200 |
+| POST `/graphql` `{ tags }` | 200 | 5.54 | 5.12 | 8.05 | 9.43 | 200 |
+| POST `/graphql` `articles(first:20)` | 200 | 9.59 | 8.68 | 13.77 | 40.28 | 200 |
+| POST `/graphql` `article(slug)` | 200 | 4.71 | 4.36 | 6.94 | 8.20 | 200 |
+| POST `/graphql` article + author + comments | 200 | 6.37 | 6.24 | 8.23 | 9.69 | 200 |
+
+Concurrent: **200** GET `/articles?limit=20` with **20** threads: wall **169 ms**, ~**1184 req/s**, p50 **10.2 ms**, p95 **19.2 ms**.
+
+Login (~67 ms) is dominated by **BCrypt**, not routing. GraphQL list p99 (~40 ms) is a tail (GC / SQLite / DEBUG logging), not a different algorithm vs REST list.
+
+Raw JSON: `architecture/baseline-metrics.json`.
+
+---
+
+## 3. Read-path cost (why list is slower than slug)
+
+Anonymous `GET /articles` (DEBUG logs):
+
+1. `queryArticles` — `DISTINCT` ids with **five** `LEFT JOIN`s (tags, favorites, two user aliases), then `LIMIT`.
+2. `countArticle` — same join shape, `count(DISTINCT A.id)`.
+3. `findArticles` — article + tags + author; 20 ids produced **30** rows (tag fan-out).
+
+Authenticated list adds `UserMapper.findById` in `JwtTokenFilter` plus batched extras in `ArticleQueryService.fillExtraInfo` (favorite counts, user favorites, following authors). Auth vs anon list: **~0.7 ms** mean — extra queries, not N+1 per article.
+
+`GET /articles/feed` with follows but **zero** matching articles: `fillExtraInfo` still runs `articlesFavoriteCount` with an empty id list → invalid `IN ()` SQL → **500**. Empty follow list short-circuits and returns 200. **Do not “fix” this during hexagonal steps** unless you treat it as out of scope; it is current behavior for the follow-but-empty-feed case.
+
+Schema: 7 tables, **no indexes** except UNIQUE on `users.username`, `users.email`, `articles.slug`. Fine at this size; list SQL will dominate if data grows.
+
+---
+
+## 4. Size and shape
+
+Non-comment lines of code (NLOC) by tree (generated DGS Java under `build/` excluded):
+
+| Area | Files | Lines | NLOC |
+| --- | --- | --- | --- |
+| `io.spring.api` | 18 | 921 | 824 |
+| `io.spring.application` | 31 | 1000 | 856 |
+| `io.spring.core` | 12 | 284 | 228 |
+| `io.spring.graphql` | 12 | 1134 | 1045 |
+| `io.spring.infrastructure` | 16 | 478 | 380 |
+| `io.spring` bootstrap | 4 | 71 | 58 |
+| MyBatis XML | 11 | 506 | 493 |
+| GraphQL schema | 1 | 177 | 150 |
+| Flyway SQL | 1 | 49 | 43 |
+| Tests | 23 | 2082 | 1785 |
+
+**93** main Java types. Stereotypes (files containing): `@RestController` 9, `@Service` 7, `@Mapper` 10, `@Repository` 4, `@Component` 4, `@Configuration` 3, `@Transactional` 1.
+
+REST handlers: **21** mappings. GraphQL `@Dgs*` methods: **~30** (queries, mutations, field resolvers).
+
+Domain (`core`) is small: `User`, `Article`, `Comment`, `Tag`, `ArticleFavorite`, `FollowRelation`, four repository interfaces, `AuthorizationService`, `JwtService`.
+
+---
+
+## 5. Coupling (hexagonal health)
+
+Target from the plan: domain/application do not import Spring/MyBatis/adapters; adapters call inbound ports only.
+
+### Cross-package imports (unique main files)
+
+| From | To | Files |
+| --- | --- | --- |
+| api | application | 8 |
+| api | core | 8 |
+| application | core | 9 |
+| application | **infrastructure** | **5** (V4) |
+| graphql | application | 9 |
+| graphql | core | 9 |
+| graphql | **api** | **8** (shared exceptions) |
+| infrastructure | core | 10 |
+| infrastructure | **application** | **4** (read mappers → `application.data`, V7) |
+| core | bootstrap (`io.spring.Util`) | **2** (V2) |
+
+### Framework on “inner” packages
+
+- **core + Spring:** `JwtService` (`@Service`), `UserRepository` (`@Repository`) — V1.
+- **application + Spring Security:** `UserService` → `PasswordEncoder` — V6.
+- **application + MyBatis types:** query services inject `*ReadService` `@Mapper`s — V4.
+- **api/graphql write orchestration:** controllers/mutations use `*Repository` + `AuthorizationService` — V5.
+
+After refactor, expect: those five application→infrastructure files **0**; core Spring annotations **0**; REST/GQL without repository fields; ArchUnit green.
+
+---
+
+## 6. Other characteristics worth tracking
+
+| Characteristic | Baseline | After refactor, watch |
+| --- | --- | --- |
+| Gradle modules | **1** | 5 (`domain`, `application`, `adapter-*`, `bootstrap`) — compile isolation |
+| ArchUnit | none | all rules enabled (plan step 7) |
+| JWT | HS512, secret in `application.properties` | same header `Authorization: Token …` |
+| Password | `BCryptPasswordEncoder` default | login p50 still ~70 ms |
+| CORS | `*` origins, credentials false | matcher behavior frozen |
+| CSRF | disabled | frozen |
+| Session | `STATELESS` | frozen |
+| Unused Boot auto-config | logs **generated security password** (`UserDetailsServiceAutoConfiguration`) | noise only |
+| Time | Joda-Time on `Article`/`Comment` | do not switch without checking JSON |
+| Persistence | MyBatis 2.2.2 + SQLite + Flyway 8 | **do not migrate to JPA** |
+| GraphQL | DGS 4.9.21, `/graphql` permitAll | same schema |
+
+---
+
+## 7. Suggested comparison checklist after step 7
+
+1. `./gradlew test` — still **68+** tests, **0** failures (ArchUnit may **add** tests).
+2. Boot JAR size — ± a few MB is noise; large jump means extra deps.
+3. Repeat HTTP table with the same seed and DEBUG flags — p50 within ~20% on this machine is “same”.
+4. `GET /articles/feed` with follow-and-empty-articles still **500** unless you explicitly fix it.
+5. Coupling table: application→infrastructure files = 0; core Spring files = 0.
+6. Startup still ~4 s class of number on this host.
+
+Do not treat hexagonal package moves as a performance project. If p95 list latency moves a lot, check log level, pool size, and SQL shape first.