无法使用 Node.js 将大量数据填充到 mongodb

Question

I am asked to import a big chunk of weather data collected from many sites all over the city. Each site has 1 computer having one folder, which is being synced to a central server each 5 mins. Everyday, a new file is created. So, basically the structure is like this. One txt file has format as a csv file, which has the 1st line as fields, and the rest are numbers.

folder_on_server
|__ site1 __ date1.txt
| |__ date2.txt
|
|__ site2 __ date1.txt
|__ date2.txt
I wrote a small node.js app to populate those data onto mongoDB. However, currently, we have only 3 sites, but each site has almost 900 txt files, each file contains 24*20 = 288 rows (as data is recorded each 5 mins). I tried to run the node app, but after reading about 100 files of the first folder, the program crashes with an error about memory allocation failure.

I have tried many ways to improve this:

1. Increase memory size of nodejs to 8GB => a litle better, more files read in but still not able to move on to the next folder.
2. Set some variable to null and undefined at the end of the _.forEach loop (I use underscore) => does not help.
3. Shift the files array (use fs.readdir), so that the first element will be deleted => does not help either.

Is there any ways to force js to clean up memory each time it finishes reading a file? Thanks

Update 1: I ended up adding 100 files in each folders at a time. This seems to be tedious but it worked, and this is like one time job. However, I still want to find a solution for this.

Answer 1

Try using streams instead of loading each file into memory.

I've sent you a pull request with an implementation using streams and async i/o.

This is most of it:

var Async = require('async');
var Csv = require('csv-streamify');
var Es = require('event-stream');
var Fs = require('fs');
var Mapping = require('./folder2siteRef.json');
var MongoClient = require('mongodb').MongoClient;

var sourcePath = '/hnet/incoming/' + new Date().getFullYear();

Async.auto({
  db: function (callback) {
    console.log('opening db connection');
    MongoClient.connect('mongodb://localhost:27017/test3', callback);
  },
  subDirectory: function (callback) {
    // read the list of subfolder, which are sites
    Fs.readdir(sourcePath, callback);
  },
  loadData: ['db', 'subDirectory', function (callback, results) {
    Async.each(results.subDirectory, load(results.db), callback);
  }],
  cleanUp: ['db', 'loadData', function (callback, results) {
    console.log('closing db connection');
    results.db.close(callback);
  }]
}, function (err) {
  console.log(err || 'Done');
});

var load = function (db) {
  return function (directory, callback) {
    var basePath = sourcePath + '/' + directory;
    Async.waterfall([
      function (callback) {
        Fs.readdir(basePath, callback); // array of files in a directory
      },
      function (files, callback) {
        console.log('loading ' + files.length + ' files from ' + directory);
        Async.each(files, function (file, callback) {
          Fs.createReadStream(basePath + '/' + file)
            .pipe(Csv({objectMode: true, columns: true}))
            .pipe(transform(directory))
            .pipe(batch(200))
            .pipe(insert(db).on('end', callback));
        }, callback);
      }
    ], callback);
  };
};

var transform = function (directory) {
  return Es.map(function (data, callback) {
    data.siteRef = Mapping[directory];
    data.epoch = parseInt((data.TheTime - 25569) * 86400) + 6 * 3600;
    callback(null, data);
  });
};

var insert = function (db) {
  return Es.map(
    function (data, callback) {
      if (data.length) {
        var bulk = db.collection('hnet').initializeUnorderedBulkOp();
        data.forEach(function (doc) {
          bulk.insert(doc);
        });
        bulk.execute(callback);
      } else {
        callback();
      }
    }
  );
};

var batch = function (batchSize) {
  batchSize = batchSize || 1000;
  var batch = [];

  return Es.through(
    function write (data) {
      batch.push(data);
      if (batch.length === batchSize) {
        this.emit('data', batch);
        batch = [];
      }
    },
    function end () {
      if (batch.length) {
        this.emit('data', batch);
        batch = [];
      }
      this.emit('end');
    }
  );
};

I've updated your tomongo.js script using streams. I've also changed it to use async instead of sync for its file i/o.

I tested this against the structure defined in your code with small data sets and it worked really well. I did some limited testing against 3xdirs with 900xfiles and 288xlines. I'm not sure how big each row of your data is, so i threw a few random properties in. Its quite fast. See how it goes with your data. If it causes issues, you could try throttling it with different write concerns when executing the bulk insert operation.

Also check out some of these links for more information on streams in node.js:

http://nodestreams.com - a tool written by John Resig with many stream examples.

And event-stream a very useful streams module.

Answer 2

As Robbie said, streams are the way to go with this. fs.createReadStream() should be used instead of .readFileSync() . I'd start with creating a line reader that takes a path and whatever string/regex you want to split on:

linereader.js

var fs = require("fs");
var util = require("util");
var EventEmitter = require("events").EventEmitter;

function LineReader(path, splitOn) {

    var readStream = fs.createReadStream(path);
    var self = this;
    var lineNum = 0;
    var buff = ""
    var chunk;

    readStream.on("readable", function() {

        while( (chunk = readStream.read(100)) !== null) {
            buff += chunk.toString();
            var lines = buff.split(splitOn);

            for (var i = 0; i < lines.length - 1; i++) {
                self.emit("line",lines[i]);
                lineNum += 1;
            }
            buff = lines[lines.length - 1];
        }
    });
    readStream.on("close", function() {
        self.emit("line", buff);
        self.emit("close")
    });
    readStream.on("error", function(err) {
        self.emit("error", err);
    })
}
util.inherits(LineReader, EventEmitter);
module.exports = LineReader;

This will read a text file, and emit "line" events for each line read, so you won't have all of them in memory at once. Then, using the async package (or whatever async loop you want to use), loop through the files inserting each document:

app.js

var LineReader = require("./linereader.js");
var async = require("async");

var paths = ["./text1.txt", "./text2.txt", "./path1/text3.txt"];
var reader;

async.eachSeries(paths, function(path, callback) {

    reader = new LineReader(path, /\n/g);

    reader.on("line", function(line) {
        var doc = turnTextIntoObject(line);
        db.collection("mycollection").insert(doc);
    })
    reader.on("close", callback);
    reader.on("error", callback);
}, function(err) {
    // handle error and finish;
})