作者: Jerry Fan

JSONL 流式处理:实时处理数据

实时流式处理 JSONL 数据的全面指南。学习使用 Node.js readline、Python 生成器、Server-Sent Events、WebSocket 和生产级日志流式处理模式构建流式管道。

为什么 JSONL 非常适合流式处理

JSONL (JSON Lines) 每行存储一个完整的 JSON 对象,用换行符分隔。这种行分隔的结构使其成为流式处理的完美格式,因为每一行都是一个自包含的、可解析的单元。与单个大型 JSON 数组不同,您无需将整个数据集加载到内存中即可开始处理记录。消费者可以读取一行、解析它、处理它,然后继续下一行,无论数据源增长到多大,内存使用量始终保持恒定。

这一特性使 JSONL 在日志聚合系统、实时分析管道、机器学习数据流和 API 流式响应中占据主导地位。OpenAI 的 API、AWS CloudWatch 和 Apache Kafka 等工具都使用换行分隔的 JSON,因为它将 JSON 的通用可读性与面向行协议的流式效率相结合。在本指南中,您将学习如何在 Node.js 和 Python 中构建 JSONL 流式解决方案、使用 Server-Sent Events 向浏览器推送实时数据、通过 WebSocket 交换 JSONL,以及实现生产级日志流式系统。

在 Node.js 中流式处理 JSONL

Node.js 建立在事件驱动、非阻塞 I/O 模型之上,这使其非常适合流式工作负载。内置的 readline 和 stream 模块提供了逐行处理 JSONL 文件和网络流所需的一切,无需将完整负载加载到内存中。

readline 模块与 fs.createReadStream 配合使用,是在 Node.js 中消费 JSONL 文件的标准方式。它分块读取文件并发出完整的行,因此即使对于数 GB 的文件,内存也保持平稳。

使用 readline 进行流式读取
import { createReadStream } from 'node:fs';
import { createInterface } from 'node:readline';


async function streamJsonl(filePath, onRecord) {
  const rl = createInterface({
    input: createReadStream(filePath, 'utf-8'),
    crlfDelay: Infinity,
  });


  let count = 0;
  for await (const line of rl) {
    const trimmed = line.trim();
    if (!trimmed) continue;
    try {
      const record = JSON.parse(trimmed);
      await onRecord(record);
      count++;
    } catch (err) {
      console.error(`Skipping invalid line: ${err.message}`);
    }
  }
  return count;
}


// Usage: process each record as it arrives
const total = await streamJsonl('events.jsonl', async (record) => {
  console.log(`Event: ${record.type} at ${record.timestamp}`);
});
console.log(`Processed ${total} records`);

Node.js Transform 流允许您构建可组合的数据管道,读取 JSONL、过滤或丰富每条记录,然后将结果写回。此模式自动处理背压,当目标无法跟上时暂停源。

用于管道的 Transform 流
import { createReadStream, createWriteStream } from 'node:fs';
import { Transform } from 'node:stream';
import { pipeline } from 'node:stream/promises';


// Split incoming chunks into individual lines
class LineSplitter extends Transform {
  constructor() {
    super({ readableObjectMode: true });
    this.buffer = '';
  }
  _transform(chunk, encoding, cb) {
    this.buffer += chunk.toString();
    const lines = this.buffer.split('\n');
    this.buffer = lines.pop();
    for (const line of lines) {
      if (line.trim()) this.push(line.trim());
    }
    cb();
  }
  _flush(cb) {
    if (this.buffer.trim()) this.push(this.buffer.trim());
    cb();
  }
}


// Parse, transform, and re-serialize each record
class JsonlTransform extends Transform {
  constructor(transformFn) {
    super({ objectMode: true });
    this.transformFn = transformFn;
  }
  _transform(line, encoding, cb) {
    try {
      const record = JSON.parse(line);
      const result = this.transformFn(record);
      if (result) cb(null, JSON.stringify(result) + '\n');
      else cb(); // Filter out null results
    } catch (err) {
      cb(); // Skip invalid lines
    }
  }
}


// Build the pipeline: read -> split -> transform -> write
await pipeline(
  createReadStream('input.jsonl', 'utf-8'),
  new LineSplitter(),
  new JsonlTransform((record) => {
    if (record.level === 'error') {
      return { ...record, flagged: true, reviewedAt: new Date().toISOString() };
    }
    return null; // Filter out non-error records
  }),
  createWriteStream('errors.jsonl', 'utf-8')
);
console.log('Pipeline complete');

在 Python 中流式处理 JSONL

Python 生成器提供了一种优雅的方式来流式处理 JSONL 数据。因为生成器一次只 yield 一条记录,并且仅在消费者请求下一个值时才向前推进,内存使用量保持最低。结合内置的 json 模块,您只需几行代码就可以构建强大的流式管道。

Python 生成器函数使用标准 for 循环逐行读取文件。每次调用 next() 时,它恰好读取一行、解析它并 yield 结果。文件句柄保持打开状态,位置仅在消费者拉取记录时才前进。

基于生成器的流式处理
import json
from typing import Generator, Any


def stream_jsonl(file_path: str) -> Generator[dict[str, Any], None, None]:
    """Stream JSONL records one at a time using a generator."""
    with open(file_path, 'r', encoding='utf-8') as f:
        for line_num, line in enumerate(f, 1):
            line = line.strip()
            if not line:
                continue
            try:
                yield json.loads(line)
            except json.JSONDecodeError as e:
                print(f"Skipping line {line_num}: {e}")


# Usage: process records lazily
for record in stream_jsonl('events.jsonl'):
    if record.get('level') == 'error':
        print(f"ERROR: {record['message']}")


# Or collect a subset with itertools
import itertools
first_100 = list(itertools.islice(stream_jsonl('large.jsonl'), 100))
print(f"Loaded first {len(first_100)} records")

Python itertools 允许您在不具体化中间结果的情况下组合过滤、批处理和切片等流式操作。链中的每一步一次处理一条记录,因此您可以处理大于可用 RAM 的文件。

使用 itertools 进行链式处理
import json
import itertools
from typing import Generator, Iterable


def stream_jsonl(path: str) -> Generator[dict, None, None]:
    with open(path, 'r') as f:
        for line in f:
            if line.strip():
                yield json.loads(line)


def filter_records(records: Iterable[dict], key: str, value) -> Generator[dict, None, None]:
    """Filter records where key equals value."""
    for record in records:
        if record.get(key) == value:
            yield record


def batch(iterable: Iterable, size: int) -> Generator[list, None, None]:
    """Yield successive batches of the given size."""
    it = iter(iterable)
    while True:
        chunk = list(itertools.islice(it, size))
        if not chunk:
            break
        yield chunk


# Compose a streaming pipeline
records = stream_jsonl('access_logs.jsonl')
errors = filter_records(records, 'status', 500)


for i, error_batch in enumerate(batch(errors, 50)):
    print(f"Batch {i + 1}: {len(error_batch)} errors")
    # Send batch to alerting system, database, etc.
    for record in error_batch:
        print(f"  {record['timestamp']} - {record['path']}")

Server-Sent Events 与 JSONL

Server-Sent Events (SSE) 提供了一种简单的、基于 HTTP 的协议,用于从服务器向客户端推送实时更新。每个 SSE 事件携带一个 data 字段,JSONL 是天然的载荷格式,因为每一行都是一个完整的 JSON 对象,客户端可以立即解析。与 WebSocket 不同,SSE 工作在标准 HTTP 上,不需要特殊的代理配置,并且在连接失败时自动重连。

服务器将 Content-Type 设置为 text/event-stream,并将每条 JSONL 记录作为 SSE data 字段写入。连接保持打开状态,服务器在有新事件可用时推送它们。

SSE 服务端 (Node.js / Express)
import express from 'express';


const app = express();


// Simulate a real-time data source
function* generateMetrics() {
  let id = 0;
  while (true) {
    yield {
      id: ++id,
      cpu: Math.random() * 100,
      memory: Math.random() * 16384,
      timestamp: new Date().toISOString(),
    };
  }
}


app.get('/api/stream/metrics', (req, res) => {
  // SSE headers
  res.setHeader('Content-Type', 'text/event-stream');
  res.setHeader('Cache-Control', 'no-cache');
  res.setHeader('Connection', 'keep-alive');
  res.setHeader('Access-Control-Allow-Origin', '*');
  res.flushHeaders();


  const metrics = generateMetrics();


  const interval = setInterval(() => {
    const record = metrics.next().value;
    // Each SSE event contains one JSONL record
    res.write(`data: ${JSON.stringify(record)}\n\n`);
  }, 1000);


  req.on('close', () => {
    clearInterval(interval);
    res.end();
  });
});


app.listen(3000, () => console.log('SSE server on :3000'));

浏览器使用 EventSource API 连接到 SSE 端点。每个传入的事件包含一条 JSONL 记录,实时解析和渲染。EventSource 在连接断开时自动处理重连。

SSE 客户端(浏览器)
// Connect to the SSE endpoint
const source = new EventSource('/api/stream/metrics');


// Parse each JSONL record as it arrives
source.onmessage = (event) => {
  const record = JSON.parse(event.data);
  console.log(`[${record.timestamp}] CPU: ${record.cpu.toFixed(1)}%`);


  // Update your UI in real time
  updateDashboard(record);
};


source.onerror = (err) => {
  console.error('SSE connection error:', err);
  // EventSource automatically reconnects
};


// For Fetch-based SSE (more control over the stream)
async function fetchSSE(url, onRecord) {
  const response = await fetch(url);
  const reader = response.body
    .pipeThrough(new TextDecoderStream())
    .getReader();


  let buffer = '';
  while (true) {
    const { done, value } = await reader.read();
    if (done) break;
    buffer += value;
    const events = buffer.split('\n\n');
    buffer = events.pop();
    for (const event of events) {
      const dataLine = event.split('\n').find(l => l.startsWith('data: '));
      if (dataLine) {
        const record = JSON.parse(dataLine.slice(6));
        onRecord(record);
      }
    }
  }
}


await fetchSSE('/api/stream/metrics', (record) => {
  console.log('Metric:', record);
});

WebSocket + JSONL 流式处理

SSE 处理服务器到客户端的推送,而 WebSocket 则支持全双工通信,双方可以随时发送 JSONL 记录。这非常适合协作编辑、带用户命令的实时仪表板和双向数据同步等交互式应用。每个 WebSocket 消息包含一条 JSONL 记录,使两端的解析都很简单。

使用 JSONL 消息的 WebSocket 服务器
import { WebSocketServer } from 'ws';


const wss = new WebSocketServer({ port: 8080 });


wss.on('connection', (ws) => {
  console.log('Client connected');


  // Send JSONL records to the client
  const interval = setInterval(() => {
    const record = {
      type: 'metric',
      value: Math.random() * 100,
      timestamp: new Date().toISOString(),
    };
    ws.send(JSON.stringify(record));
  }, 500);


  // Receive JSONL records from the client
  ws.on('message', (data) => {
    try {
      const record = JSON.parse(data.toString());
      console.log(`Received: ${record.type}`, record);


      // Echo back an acknowledgment
      ws.send(JSON.stringify({
        type: 'ack',
        originalId: record.id,
        processedAt: new Date().toISOString(),
      }));
    } catch (err) {
      ws.send(JSON.stringify({ type: 'error', message: 'Invalid JSON' }));
    }
  });


  ws.on('close', () => {
    clearInterval(interval);
    console.log('Client disconnected');
  });
});


// Client-side usage:
// const ws = new WebSocket('ws://localhost:8080');
// ws.onmessage = (e) => {
//   const record = JSON.parse(e.data);
//   console.log(record);
// };
// ws.send(JSON.stringify({ type: 'subscribe', channel: 'metrics' }));

每个 WebSocket 消息都是一个单独的 JSON 对象,遵循 JSONL 每条消息一条记录的约定。服务器以 500 毫秒的间隔向客户端流式传输指标,同时也接受来自客户端的命令。两个方向使用相同的 JSON 序列化,因此协议是对称的且易于调试。

日志流式处理实践

JSONL 流式处理最常见的实际用途之一是集中式日志收集。应用程序将结构化日志条目作为 JSONL 行写入 stdout 或文件,收集器代理实时跟踪输出,将记录转发到中央系统。Docker、Kubernetes 和大多数云日志服务都使用这种模式。

实时 JSONL 日志收集器
import { createReadStream, watchFile } from 'node:fs';
import { createInterface } from 'node:readline';
import { stat } from 'node:fs/promises';


class JsonlLogCollector {
  constructor(filePath, handlers = {}) {
    this.filePath = filePath;
    this.handlers = handlers;
    this.position = 0;
    this.running = false;
  }


  async start() {
    this.running = true;
    // Read existing content first
    await this.readFrom(0);
    // Watch for new appends
    this.watch();
  }


  async readFrom(startPos) {
    const rl = createInterface({
      input: createReadStream(this.filePath, {
        encoding: 'utf-8',
        start: startPos,
      }),
      crlfDelay: Infinity,
    });


    for await (const line of rl) {
      if (!line.trim()) continue;
      try {
        const record = JSON.parse(line);
        await this.dispatch(record);
      } catch (err) {
        this.handlers.onError?.(err, line);
      }
    }
    const info = await stat(this.filePath);
    this.position = info.size;
  }


  watch() {
    watchFile(this.filePath, { interval: 500 }, async (curr, prev) => {
      if (!this.running) return;
      if (curr.size > this.position) {
        await this.readFrom(this.position);
      }
    });
  }


  async dispatch(record) {
    const level = record.level || 'info';
    this.handlers[level]?.(record);
    this.handlers.onRecord?.(record);
  }


  stop() {
    this.running = false;
  }
}


// Usage
const collector = new JsonlLogCollector('app.log.jsonl', {
  error: (r) => console.error(`[ERROR] ${r.message}`),
  warn: (r) => console.warn(`[WARN] ${r.message}`),
  onRecord: (r) => forwardToElasticsearch(r),
  onError: (err, line) => console.error('Bad line:', line),
});


await collector.start();

此收集器实时跟踪 JSONL 日志文件,根据日志级别将记录分发给处理器。它跟踪文件位置,因此在初始追赶读取之后只读取新内容。在生产环境中,您需要添加文件轮转检测、优雅关闭和批量转发以减少到集中式日志系统的网络调用。

在线处理 JSONL 文件

在浏览器中查看、验证和转换最大 1GB 的 JSONL 文件。无需上传,100% 私密。

常见问题

参考资料