Response streaming in Express does not work in Azure App Service

问题

I am trying to stream responses to my client using a NodeJS Express server hosted using Azure App Service. However, I noticed that it is not really streaming but tries to send the response as a whole. When the response size is huge (>50MB), the client gets an Internal Server Error, but the server does not throw an error.

Further, when I run the server inside a Docker (Node Image: 10.22.0-alpine3.9), I see that the client gets the response as a stream even for huge responses. (This is the behavior I actually need)

My web.config file is as follows.

<?xml version="1.0" encoding="utf-8"?>
<configuration>
  <system.webServer>
    <handlers>
      <add name="iisnode" path="server.js" verb="*" modules="iisnode" responseBufferLimit="0"/>
    </handlers>
    <iisnode flushResponse="true" />
    ...
  </system.webServer>
</configuration>

This is a small explanation as to what my program does.

I have an external API that returns an object similar to the following.

{
  "title":"Test Title",
  "lastBuildDate":"1597981114347",
  "items":[
    {
      id: 'item1',
      value: 'value1'
    },
    {
      id: 'item2',
      value: 'value2'
    },
    ...
  [
}

I want to filter only the elements in items array and send those to the client. The client should get a response like below.

[
   {
     id: 'item1',
     value: 'value1'
   },
   {
     id: 'item2',
     value: 'value2'
   },
   ...
[

Sometimes this object is too large (>50MB) and because of that, I am sending the response as a stream to avoid using too much buffer memory in my server. Below here is the code I used to stream the response.

const https = require('https');
const { withParser } = require('stream-json/filters/Pick');
const { streamArray } = require('stream-json/streamers/StreamArray');
const { chain } = require('stream-chain');

exports.getStreamResponse = async function (req, res) {
  const options = {
    hostname,
    port,
    path,
    method: 'GET',
  };

  return new Promise((resolve, reject) => {
    https.request(options, (dataStream) => {
      const pipeline = chain([
        dataStream,
        withParser({ filter: 'items' }),
        streamArray()
      ]);
  
      res.write("[");
  
      let separator = '';
  
      pipeline.on('data', data => {
        res.write(separator + JSON.stringify(data.value));
        if (!separator) {
          separator = ',';
        }
      });
  
      pipeline.on('end', () => {
        res.write("]");
        res.end();
        resolve();
      });

      pipeline.on('error', (error) => {
        reject(error);
      });
    });
  })
};
            

I also noticed that if I write the code like below, I always get a stream response. However, the response is not in the correct format as needed.

https.request(options, (dataStream) => {
  dataStream.pipe(res);
});

回答1:

Like I described in the latter part of my question, directly piping the res (my response to the client) to dataStream (the data stream I got from the external API) allowed to stream without any issues.

Extending the same behavior, I created a Readable stream which is equivalent to the response I should send to my client. Then I piped it to res and it worked.

Here is my solution.

const https = require('https');
const { withParser } = require('stream-json/filters/Pick');
const { streamArray } = require('stream-json/streamers/StreamArray');
const { chain } = require('stream-chain');
const { Readable } = require('stream');

exports.getStreamResponse = async function (req, res) {
  const options = {
    hostname,
    port,
    path,
    method: 'GET',
  };

  return new Promise((resolve, reject) => {
    https.request(options, (dataStream) => {
      const pipeline = chain([
        dataStream,
        withParser({ filter: 'items' }),
        streamArray()
      ]);
  
      // create a readable stream to collect data from response 
      const readable = new Readable({
        // this empty method is to avoid 'ERR_METHOD_NOT_IMPLEMENTED'
        // error when read method is called while there is no data in the
        // readable stream
        read(size) { }
      });
  
      let separator = '';
  
      readable.pipe(res);
      readable.push("[");

      pipeline.on('data', data => {
        readable.push(separator + JSON.stringify(data.value));
        if (!separator) {
          separator = ',';
        }
      });

      pipeline.on('end', () => {
        readable.push("]");
        readable.push(null);
        resolve();
      });
            
      pipeline.on('error', reject);
    });
  })
};

However, I noticed this solution requires more memory than the solution I had issues with. Probably because I am creating a readable stream that is redundant.

来源：https://stackoverflow.com/questions/63555610/response-streaming-in-express-does-not-work-in-azure-app-service