refactor: migrate integration tests to Go, split monolith, clean up seed data

.github/workflows/ci.yml

@@ -36,12 +36,12 @@ jobs:
   integration_test:
     runs-on: ubuntu-latest
     steps:
-      - name: Check out source code
-        uses: actions/checkout@v4
-      - run: cd integration_tests && make build-image
-      - run: cd integration_tests && make down
-      - run: cd integration_tests && make up
-      - run: cd integration_tests && make test
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v5
+        with:
+          go-version: "1.26"
+      - name: Build, run Go integration tests, and benchmarks against all instances
+        run: cd integration_tests && make test-and-bench
   e2e_test:
     runs-on: ubuntu-latest
     steps:

.gitignore

@@ -33,3 +33,5 @@ coverage.txt
 /.cursor/
 /.sem/
 progress.md
+integration-bench.txt
+

AGENTS.md

@@ -14,6 +14,8 @@ Flagr — Go feature flag service with Vue 3 UI.
 | `make test` | Go unit tests |
 | `make test-e2e` | Build Go binary → start servers via `scripts/e2e-server.sh` → Playwright → cleanup |
 | `make swagger` | Regenerate `swagger_gen/` from OpenAPI spec |
+| `make test-integration` | Run Go integration tests (auto-starts local server, SQLite) |
+| `make bench-integration` | Run HTTP eval benchmarks against local server |
 
 **UI-only** (`browser/flagr-ui/`): `npm run dev` (Vite), `npm run build`, `npm run test:e2e` (needs servers running).
 
@@ -45,6 +47,11 @@ Flagr — Go feature flag service with Vue 3 UI.
 
 **E2E tests:** `make test-e2e` — single command. Uses `scripts/e2e-server.sh` (idempotent, port-safe) and Playwright's `webServer` lifecycle. Always works regardless of leftover processes.
 
+**Integration tests:** Three modes:
+- **Local** (`make test-integration`): Auto-starts server on random port with SQLite `:memory:`. Run `go test -tags=integration ./integration_tests/` directly.
+- **Docker Compose** (`cd integration_tests && make test`): Builds Go test binary, loops over 6 flagr instances (sqlite, mysql, mysql8, postgres9, postgres13, checkr), runs suite against each.
+- **Benchmarks** (`make bench-integration`): HTTP eval benchmarks against auto-started server.
+
 **Process management** uses `lsof -ti:<port>` not `pkill -f`, so it never touches processes from other projects.
 
 ## Constraints

Makefile

@@ -41,6 +41,18 @@ test-e2e: build
 	@echo "Running Flagr UI e2e tests..."
 	@cd ./browser/flagr-ui/; npx playwright test
 
+.PHONY: test-integration
+test-integration: build
+	@echo "Running Go integration tests (local auto-start mode)..."
+	@go test -tags=integration -count=1 -v ./integration_tests/
+
+.PHONY: bench-integration
+bench-integration: build
+	@echo "Running Go integration benchmarks (local auto-start mode)..."
+	@go test -tags=integration -bench=. -benchmem -count=1 -run=^$$ ./integration_tests/ > integration-bench.txt
+	@echo "Benchmarks saved to integration-bench.txt"
+
+
 stop-ui:
 	@-kill $$(lsof -ti:18000 2>/dev/null) 2>/dev/null; kill $$(lsof -ti:8080 2>/dev/null) 2>/dev/null; sleep 1; echo "Stopped UI services"
 

docs/_sidebar.md

@@ -5,6 +5,8 @@
     - [Overview](flagr_overview.md)
     - [Use Cases](flagr_use_cases.md)
     - [Debug Console](flagr_debugging.md)
+- Development & Testing
+    - [Testing](home?id=testing)
 - Server Configuration
     - [Env](flagr_env.md)
     - [Notifications](flagr_notifications.md)

docs/home.md

@@ -61,3 +61,61 @@ And alternatively to just run the UI service:
 ```
 make run_ui
 ```
+
+## Testing
+
+Flagr has three kinds of tests, each serving a different purpose.
+
+### Unit tests
+
+Run the Go unit tests (no external services required):
+
+```bash
+make test
+```
+
+Or directly:
+
+```bash
+go test ./pkg/...
+```
+
+### E2E tests (Flagr UI)
+
+Playwright-based end-to-end tests for the Vue 3 UI. Builds the Go
+server, starts the backend and UI servers, runs Playwright, then cleans up:
+
+```bash
+make test-e2e
+```
+
+### Integration tests (API, multi-DB)
+
+HTTP-level integration tests covering all CRUD and eval endpoints.
+Seeds ~50 realistic flags across all 12 constraint operators.
+
+**Local mode** — SQLite `:memory:`, auto-starts server on random port:
+
+```bash
+make test-integration
+```
+
+**Docker Compose mode** — runs the same test suite against 6 flagr
+instances (SQLite, MySQL, MySQL 8, PostgreSQL 9, PostgreSQL 13,
+checkr/flagr):
+
+```bash
+cd integration_tests && make test
+```
+
+**HTTP eval benchmarks** — measures end-to-end eval latency through HTTP:
+
+```bash
+make bench-integration
+```
+
+To run against a single Docker Compose instance:
+
+```bash
+cd integration_tests && make test-instance INSTANCE=flagr_with_mysql
+```