docs: define minimal vehicle storage contract

This commit is contained in:
lingniu
2026-07-02 19:38:24 +08:00
parent 8fd38eb87e
commit be559c0f99
3 changed files with 448 additions and 0 deletions

View File

@@ -41,6 +41,8 @@ flowchart LR
## Minimal Storage Contract ## Minimal Storage Contract
详细的落库边界见 [车辆数据最小落库合约](storage-minimal-contract.md)。
| Store | Table or key | Purpose | Keep | Avoid | | Store | Table or key | Purpose | Keep | Avoid |
| --- | --- | --- | --- | --- | | --- | --- | --- | --- | --- |
| TDengine | `raw_frames` | Replay and audit evidence | raw hex/text, full parsed JSON, parse status, protocol tags | business-only duplicated columns | | TDengine | `raw_frames` | Replay and audit evidence | raw hex/text, full parsed JSON, parse status, protocol tags | business-only duplicated columns |

View File

@@ -0,0 +1,70 @@
# 车辆数据最小落库合约
## 目标
车辆接入系统只持久化能回答业务问题、能回放纠错、能支撑高频查询的数据。协议细节默认保存在 raw不向上层业务表扩散。
## 第一性原则
1. raw 是证据层,必须能证明收到过什么、解析成什么。
2. Kafka 是回放层,消费者失败后优先靠 Kafka lag 追平。
3. Redis 是当前态缓存,不是历史库。
4. TDengine 只承担高写入时间序列raw、位置历史。
5. MySQL 只承担低基数业务状态:身份映射、实时快照、日指标。
6. 同一个事实只落一个主表,其他表只存查询必要的投影。
7. 新字段先进入 `parsed_json`,只有稳定查询需求出现后才提升为列。
## 目标表边界
| 存储 | 表/Key | 保留内容 | 不保留内容 |
| --- | --- | --- | --- |
| TDengine | `raw_frames` | raw hex/text、完整 `parsed_json`、解析状态、协议标签、车辆标识标签 | `fields_json` 这类可从 `parsed_json`/业务表再得到的重复字段 |
| TDengine | `raw_frame_payload_chunks` | 超长 raw/parsed payload 分片 | 业务查询字段 |
| TDengine | `vehicle_locations` | 时间、VIN/协议标签、经纬度、速度、方向、SOC、总里程等位置核心字段 | 完整协议 JSON、注册鉴权信息 |
| MySQL | `vehicle_realtime_snapshot` | 每个协议+VIN 的最新事件时间、接收时间、车牌、事件 ID | phone、device、source endpoint、完整 JSON |
| MySQL | `vehicle_realtime_location` | 每个协议+VIN 的最新位置核心字段 | raw、parsed JSON、消息头内部字段 |
| MySQL | `vehicle_daily_metric` | 日期、VIN/临时 vehicle key、协议、指标名、指标值、首末总里程、样本数 | 每帧细节、位置点列表 |
| MySQL | `vehicle_identity_binding` | 人工维护或导入的 VIN、车牌、phone、device_id 映射 | 注册历史 |
| MySQL | `jt808_registration` | JT808 phone 主键下的注册、鉴权、VIN 匹配状态 | GB32960/MQTT 注册信息 |
| Redis | `vehicle:realtime-raw:{protocol}:{vin}` | 每协议最新完整 parsed 状态 | 历史数据、统计结果 |
## 当前应收敛的重复点
1. `vehicle_mileage_points``vehicle_locations.total_mileage_km` 重复。
- 目标:停止写入 `vehicle_mileage_points`
- 兼容:`/api/history/mileage-points` 改为从 `vehicle_locations` 读取 `total_mileage_km IS NOT NULL`
- 删除:确认线上无直接读表后,再删除 TDengine stable 和子表。
2. `raw_frames.fields_json``raw_frames.parsed_json``vehicle_locations` 重复。
- 目标raw 只保留完整 `parsed_json`,核心查询字段进入 `vehicle_locations`
- 兼容:先让 raw 查询返回不依赖 `fields_json`
- 删除:等历史查询和前端不再展示 `fields_json` 后删除列。
3. `vehicle_daily_metric``vehicle_key``vin` 同时存在。
- 当前保留原因JT808 可能先用 phone 作为临时 key后续才绑定 VIN。
- 收敛方向:业务查询以 VIN 为主;没有 VIN 的数据只用于排查和待绑定,不作为正式车辆指标。
## 字段提升规则
协议字段进入系统后分三层:
1. `parsed_json`:默认入口,保存完整结构化解析。
2. 核心列只有跨协议稳定查询需要时才提升例如经纬度、速度、SOC、总里程。
3. 指标表:只有聚合口径稳定、产品需要分页/排序/报表时才持久化。
反例:
- 不因为某个协议字段存在就增加 MySQL 列。
- 不为了方便调试在 snapshot/location 放完整 JSON。
- 不为 telemetry field 配置服务提前复制一份字段表;字段配置服务应从 raw/Kafka 回放生成自己的结果。
## 删除或迁移顺序
1. 先写测试证明旧 API 能从新主表读到同等结果。
2. 改代码停止写重复表或重复列。
3. 部署后观察 Kafka lag、writer 成功计数、API 结果。
4. 查询生产库确认旧表不再新增。
5. 做一次备份或导出。
6. 删除旧表或旧列。
任何一步无法证明安全,就停在兼容状态,不直接删生产数据。

View File

@@ -0,0 +1,376 @@
# Storage Simplification Implementation Plan
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
**Goal:** 精简车辆接入落库模型,去掉重复持久化,同时保持 raw 可回放、实时可查、历史位置和日指标可用。
**Architecture:** TDengine 只保留 raw 证据和位置历史两类高写入时间序列MySQL 只保留身份映射、实时当前态和日指标。先兼容 API再停止重复写入最后在生产验证后删除旧表/旧列。
**Tech Stack:** Go, TDengine, MySQL, Redis, Kafka, NATS JetStream, `go test`.
---
## File Structure
- Modify: `go/vehicle-gateway/internal/history/schema.go`
- Stop creating `vehicle_mileage_points` after API compatibility is in place.
- Modify: `go/vehicle-gateway/internal/history/writer.go`
- Stop calling `AppendMileagePoint`; keep total mileage in `vehicle_locations`.
- Modify: `go/vehicle-gateway/internal/history/query.go`
- Make `MileagePointRepository` read from `vehicle_locations` with `total_mileage_km IS NOT NULL`.
- Modify: `go/vehicle-gateway/internal/history/query_test.go`
- Update mileage point query expectations to `vehicle_locations`.
- Modify: `go/vehicle-gateway/internal/history/writer_test.go`
- Prove no `vehicle_mileage_points` stable/table is created or written.
- Modify: `go/vehicle-gateway/internal/history/schema.go`
- In a later task, remove `fields_json` from new `raw_frames` schema.
- Modify: `go/vehicle-gateway/internal/history/writer.go`
- In a later task, stop writing `fields_json`.
- Modify: `go/vehicle-gateway/internal/history/query.go`
- In a later task, make raw query tolerate schemas without `fields_json`.
- Modify: `docs/architecture/storage-minimal-contract.md`
- Keep the target storage contract updated as implementation lands.
### Task 1: Move mileage point queries onto `vehicle_locations`
**Files:**
- Modify: `go/vehicle-gateway/internal/history/query.go`
- Modify: `go/vehicle-gateway/internal/history/query_test.go`
- [ ] **Step 1: Write the failing test**
Update `TestMileagePointHandlerReturnsMileageByVehicleKey` in `go/vehicle-gateway/internal/history/query_test.go` so it expects `vehicle_locations`, not `vehicle_mileage_points`:
```go
mock.ExpectQuery("SELECT COUNT\\(\\*\\) FROM lingniu_vehicle_ts.vehicle_locations").
WillReturnRows(sqlmock.NewRows([]string{"count"}).AddRow(17))
mock.ExpectQuery("SELECT ts, event_id, frame_id, received_at, total_mileage_km, speed_kmh, longitude, latitude, protocol, vehicle_key, vin, phone, device_id FROM lingniu_vehicle_ts.vehicle_locations").
WillReturnRows(sqlmock.NewRows([]string{
"ts", "event_id", "frame_id", "received_at", "total_mileage_km", "speed_kmh", "longitude", "latitude",
"protocol", "vehicle_key", "vin", "phone", "device_id",
}).AddRow("2026-07-02 10:00:00.000", "event-1", "frame-1", "2026-07-02 10:00:01.000", 8792.8, 8.0, 121.07764, 31.22436, "JT808", "JT808:13307811350", "VIN001", "13307811350", "dev001"))
```
- [ ] **Step 2: Run test to verify it fails**
Run:
```bash
cd go/vehicle-gateway
go test ./internal/history -run 'TestMileagePointHandlerReturnsMileageByVehicleKey|TestBuildMileagePointSQLUsesLiteralsForTDengine' -count=1
```
Expected: FAIL because `MileagePointRepository.tableName()` still returns `vehicle_mileage_points`.
- [ ] **Step 3: Point mileage repository at location table**
Change `MileagePointRepository.tableName()` in `go/vehicle-gateway/internal/history/query.go`:
```go
func (r *MileagePointRepository) tableName() string {
if r.database == "" {
return "vehicle_locations"
}
return r.database + ".vehicle_locations"
}
```
Change `mileagePointWhere` so it filters only rows with total mileage:
```go
func mileagePointWhere(query MileagePointQuery) []string {
where := []string{"total_mileage_km IS NOT NULL"}
add := func(condition string) {
where = append(where, condition)
}
if query.Protocol != "" {
add("protocol = '" + quote(query.Protocol) + "'")
}
if query.VehicleKey != "" {
add("vehicle_key = '" + quote(query.VehicleKey) + "'")
}
if query.VIN != "" {
add("vin = '" + quote(query.VIN) + "'")
}
if query.Phone != "" {
add("phone = '" + quote(query.Phone) + "'")
}
if query.DeviceID != "" {
add("device_id = '" + quote(query.DeviceID) + "'")
}
if query.DateFrom != "" {
add("ts >= '" + quote(normalizeDateTimeLiteral(query.DateFrom)) + "'")
}
if query.DateTo != "" {
add("ts <= '" + quote(normalizeDateTimeLiteral(query.DateTo)) + "'")
}
return where
}
```
- [ ] **Step 4: Run tests**
Run:
```bash
cd go/vehicle-gateway
go test ./internal/history -count=1
```
Expected: PASS.
- [ ] **Step 5: Commit**
```bash
git add go/vehicle-gateway/internal/history/query.go go/vehicle-gateway/internal/history/query_test.go
git commit -m "refactor(go): read mileage points from locations"
```
### Task 2: Stop writing duplicate mileage point table
**Files:**
- Modify: `go/vehicle-gateway/internal/history/schema.go`
- Modify: `go/vehicle-gateway/internal/history/writer.go`
- Modify: `go/vehicle-gateway/internal/history/writer_test.go`
- [ ] **Step 1: Write the failing tests**
Update `go/vehicle-gateway/internal/history/writer_test.go`:
```go
for _, want := range []string{"CREATE DATABASE IF NOT EXISTS test_ts", "raw_frames", "vehicle_locations"} {
if got := countSQL(exec.calls, want); got == 0 {
t.Fatalf("expected schema statement containing %q", want)
}
}
if got := countSQL(exec.calls, "vehicle_mileage_points"); got != 0 {
t.Fatalf("schema should not create vehicle_mileage_points, got %d", got)
}
if got := countSQL(exec.calls, "INSERT INTO mil_"); got != 0 {
t.Fatalf("history writer should not insert duplicate mileage point rows, got %d", got)
}
```
- [ ] **Step 2: Run test to verify it fails**
Run:
```bash
cd go/vehicle-gateway
go test ./internal/history -run 'TestWriter' -count=1
```
Expected: FAIL because schema and writer still create/write `vehicle_mileage_points`.
- [ ] **Step 3: Remove duplicate write path**
Change `AppendAll` in `go/vehicle-gateway/internal/history/writer.go`:
```go
func (w *Writer) AppendAll(ctx context.Context, env envelope.FrameEnvelope) error {
if err := w.AppendRawFrame(ctx, env); err != nil {
return err
}
return w.AppendLocation(ctx, env)
}
```
Remove `vehicle_mileage_points` from `SchemaStatements` in `go/vehicle-gateway/internal/history/schema.go`. Keep `AppendMileagePoint` only if tests still use it directly; otherwise remove the function and its helper after compile errors guide cleanup.
- [ ] **Step 4: Run tests**
Run:
```bash
cd go/vehicle-gateway
go test ./internal/history ./cmd/history-writer -count=1
```
Expected: PASS.
- [ ] **Step 5: Commit**
```bash
git add go/vehicle-gateway/internal/history/schema.go go/vehicle-gateway/internal/history/writer.go go/vehicle-gateway/internal/history/writer_test.go
git commit -m "refactor(go): stop writing duplicate mileage history"
```
### Task 3: Remove `fields_json` from new raw writes
**Files:**
- Modify: `go/vehicle-gateway/internal/history/schema.go`
- Modify: `go/vehicle-gateway/internal/history/writer.go`
- Modify: `go/vehicle-gateway/internal/history/writer_test.go`
- Modify: `go/vehicle-gateway/internal/history/query.go`
- Modify: `go/vehicle-gateway/internal/history/query_test.go`
- [ ] **Step 1: Write failing writer test**
Add this assertion in the raw insert test:
```go
rawInsert := findSQL(exec.calls, "INSERT INTO raw_")
if strings.Contains(rawInsert, "fields_json") {
t.Fatalf("raw insert should not write duplicate fields_json: %s", rawInsert)
}
```
- [ ] **Step 2: Run test to verify it fails**
Run:
```bash
cd go/vehicle-gateway
go test ./internal/history -run 'TestWriter' -count=1
```
Expected: FAIL because raw insert still includes `fields_json`.
- [ ] **Step 3: Remove `fields_json` from schema and writer**
In `go/vehicle-gateway/internal/history/schema.go`, remove:
```sql
fields_json BINARY(4096),
```
In `go/vehicle-gateway/internal/history/writer.go`, remove the `fieldsJSON` variable and `fields_json` insert column. Keep `parsed_json` complete.
- [ ] **Step 4: Make raw query backward compatible**
Keep API response field `fields_json,omitempty`, but query only columns that exist. Add a repository option so tests and startup can select the schema shape explicitly:
```go
type RawFrameRepository struct {
db Queryer
database string
includeFieldsJSON bool
}
func NewRawFrameRepository(db Queryer, database string) *RawFrameRepository {
return NewRawFrameRepositoryWithOptions(db, database, RawFrameRepositoryOptions{
IncludeFieldsJSON: true,
})
}
type RawFrameRepositoryOptions struct {
IncludeFieldsJSON bool
}
func NewRawFrameRepositoryWithOptions(db Queryer, database string, opts RawFrameRepositoryOptions) *RawFrameRepository {
if db == nil {
panic("raw frame query db must not be nil")
}
database = strings.TrimSpace(database)
if database != "" && !safeIdentifier(database) {
database = ""
}
return &RawFrameRepository{db: db, database: database, includeFieldsJSON: opts.IncludeFieldsJSON}
}
```
Then split raw SQL construction:
```go
func buildRawFrameSQL(table string, query RawFrameQuery, includeFieldsJSON bool) (string, []any) {
fieldsColumn := ""
if includeFieldsJSON {
fieldsColumn = "fields_json, "
}
sqlText := `SELECT ts, frame_id, event_id, message_id, event_time, received_at, raw_size_bytes, raw_hex, raw_text, parsed_json, ` +
fieldsColumn +
`parse_status, parse_error, source_endpoint, protocol, vehicle_key, vin, phone, device_id FROM ` + table
where := rawFrameWhere(query)
if len(where) > 0 {
sqlText += " WHERE " + strings.Join(where, " AND ")
}
sqlText += " ORDER BY " + rawFrameOrderColumn(query.OrderBy) + " DESC LIMIT " + strconv.Itoa(query.Limit) + " OFFSET " + strconv.Itoa(query.Offset)
return sqlText, nil
}
```
The new-schema test must scan a row without `fields_json` and assert the JSON response omits `fields_json`. The old-schema test must use `IncludeFieldsJSON: true` and assert `fields_json` is hydrated.
- [ ] **Step 5: Run tests**
Run:
```bash
cd go/vehicle-gateway
go test ./internal/history ./cmd/history-writer ./cmd/realtime-api -count=1
```
Expected: PASS.
- [ ] **Step 6: Commit**
```bash
git add go/vehicle-gateway/internal/history/schema.go go/vehicle-gateway/internal/history/writer.go go/vehicle-gateway/internal/history/writer_test.go go/vehicle-gateway/internal/history/query.go go/vehicle-gateway/internal/history/query_test.go
git commit -m "refactor(go): remove duplicate raw fields json"
```
### Task 4: Production migration and cleanup gate
**Files:**
- Modify: `docs/architecture/storage-minimal-contract.md`
- Modify: `docs/ops/vehicle-ingest-runbook.md`
- [ ] **Step 1: Deploy the code changes**
Run the existing build/deploy flow for Go services, then restart:
```bash
systemctl restart lingniu-go-history-writer.service
systemctl restart lingniu-go-realtime-api.service
```
- [ ] **Step 2: Verify service health**
Run on ECS:
```bash
for port in 20212 20200; do
curl -fsS "http://127.0.0.1:${port}/readyz"
echo
done
curl -fsS http://127.0.0.1:20212/metrics | grep vehicle_history_kafka_lag
curl -fsS http://127.0.0.1:20212/metrics | grep vehicle_history_writes_total
```
Expected:
- both readiness endpoints return `status=ok`
- history lag drains to `0` or remains bounded
- history writes continue increasing
- [ ] **Step 3: Verify no new duplicate mileage table writes**
Run TDengine checks on ECS:
```sql
SELECT COUNT(*) FROM lingniu_vehicle_ts.vehicle_locations WHERE total_mileage_km IS NOT NULL;
SELECT COUNT(*) FROM lingniu_vehicle_ts.vehicle_mileage_points;
```
Expected: `vehicle_locations` grows; `vehicle_mileage_points` stops increasing after deployment time.
- [ ] **Step 4: Drop old table only after observation**
After at least one production observation window with no direct reads from `vehicle_mileage_points`, run:
```sql
DROP STABLE IF EXISTS lingniu_vehicle_ts.vehicle_mileage_points;
```
- [ ] **Step 5: Commit docs update**
```bash
git add docs/architecture/storage-minimal-contract.md docs/ops/vehicle-ingest-runbook.md
git commit -m "docs: record storage simplification migration"
```
## Self-Review
- Spec coverage: The plan addresses minimal storage, duplicate mileage storage, raw field duplication, production verification, and delayed destructive cleanup.
- Red-flag scan: The plan has no open-ended task. Task 3 specifies the compatibility option and the new/old schema test expectations.
- Type consistency: The plan keeps existing public API types and changes only their backing tables.