Hi all, I've submitted a PR that adds an OTLP(OpenTelemetry Protocol)/HTTP receiver to IoTDB's REST service, allowing developers to store telemetry data from AI coding assistants (Claude Code, Codex, Gemini CLI, etc.) directly into a local IoTDB instance for personal analytics.
PR: https://github.com/apache/iotdb/pull/17516 ## Why This Matters Modern AI coding tools like Claude Code emit rich telemetry via OpenTelemetry — token usage, API costs, tool execution stats, and session events. By routing this data into IoTDB, you can: - Track token consumption across models (input / output / cacheRead / cacheCreation) - Monitor daily/weekly API costs - Analyze tool success rates and latencies - Correlate events within a single prompt via prompt.id ## Quick Setup (Claude Code Example) 1. Enable REST service in `iotdb-system.properties`: ```properties enable_rest_service=true rest_service_port=18080 ``` 2. Add these environment variables to `~/.zshrc` or `~/.bashrc`: ```bash export CLAUDE_CODE_ENABLE_TELEMETRY=1 export OTEL_METRICS_EXPORTER=otlp export OTEL_LOGS_EXPORTER=otlp export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:18080/rest/v1/otlp ``` 3. Start IoTDB, open a new terminal, run `claude`. That's it — data flows automatically. 4. Query examples: ```sql -- Token usage by model USE claude_code; SELECT model, type, sum(value) FROM metrics WHERE metric_name = 'claude_code.token.usage' GROUP BY model, type; -- Daily cost SELECT date_bin(1d, time) AS day, sum(value) AS cost FROM metrics WHERE metric_name = 'claude_code.cost.usage' GROUP BY 1 ORDER BY 1; -- Tool execution stats SELECT tool_name, count_if(success = 'true') AS ok, count(*) AS total FROM logs WHERE event_name = 'tool_result' GROUP BY tool_name; ``` ## Design Highlights - **Zero extra infrastructure**: reuses the existing REST service (port 18080), no gRPC or Collector needed - **Dynamic database routing**: database name is derived from `service.name` attribute — `claude-code` → `claude_code`, `codex` → `codex`, etc. - **Structured schema**: OTLP attributes are flattened into typed TAG/ATTRIBUTE/FIELD columns, not stored as JSON blobs - **Auto schema creation**: database and tables are created on first ingest Feedback welcome — especially on the table schema design and whether this should support additional OTLP features. --- 大家好， 我提交了一个 PR，为 IoTDB 的 REST 服务添加了 OTLP(OpenTelemetry Protocol)/HTTP 接收器，使开发者可以将 AI 编程助手（Claude Code、Codex、Gemini CLI 等）的遥测数据直接存入本地 IoTDB 实例，用于个人使用分析。 PR: https://github.com/apache/iotdb/pull/17516 ## 为什么有用 Claude Code 等 AI 编程工具通过 OpenTelemetry 协议发送丰富的遥测数据——Token 用量、API 成本、工具执行统计和会话事件。将这些数据路由到 IoTDB 后，你可以： - 追踪各模型的 Token 消耗（input / output / cacheRead / cacheCreation） - 监控每日/每周 API 花费 - 分析工具调用成功率和延迟 - 通过 prompt.id 关联单次提示下的所有事件 ## 快速配置（以 Claude Code 为例） 1. 在 `iotdb-system.properties` 中启用 REST 服务： ```properties enable_rest_service=true rest_service_port=18080 ``` 2. 在 `~/.zshrc` 或 `~/.bashrc` 中添加以下环境变量： ```bash export CLAUDE_CODE_ENABLE_TELEMETRY=1 export OTEL_METRICS_EXPORTER=otlp export OTEL_LOGS_EXPORTER=otlp export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:18080/rest/v1/otlp ``` 3. 启动 IoTDB，打开新终端，运行 `claude`。数据会自动流入。 4. 查询示例： ```sql -- 按模型统计 Token 用量 USE claude_code; SELECT model, type, sum(value) FROM metrics WHERE metric_name = 'claude_code.token.usage' GROUP BY model, type; -- 每日花费 SELECT date_bin(1d, time) AS day, sum(value) AS cost FROM metrics WHERE metric_name = 'claude_code.cost.usage' GROUP BY 1 ORDER BY 1; -- 工具执行统计 SELECT tool_name, count_if(success = 'true') AS ok, count(*) AS total FROM logs WHERE event_name = 'tool_result' GROUP BY tool_name; ``` ## 设计亮点 - **零额外基础设施**：复用已有 REST 服务（18080 端口），无需 gRPC 或 Collector - **动态数据库路由**：根据 `service.name` 属性自动推导数据库名——`claude-code` → `claude_code`、`codex` → `codex` - **结构化 Schema**：OTLP 属性被展平为带类型的 TAG/ATTRIBUTE/FIELD 列，而非存为 JSON 字符串 - **自动建表**：首次写入时自动创建数据库和表 欢迎反馈——特别是关于表结构设计以及是否需要支持更多 OTLP 特性。 Best regards, ---------------- Yuan Tian
