在不超出堆栈限制的情况下迭代或递归大量函数的最佳方法是什么？

I have an application that I'm writing in Node.js which needs to make a lot of configuration and database calls in order to process user data. The issue I'm having is that after 11,800+ function calls Node will throw an error and exit the process.

我有一个应用程序,我正在Node.js编写,需要进行大量的配置和数据库调用,以处理用户数据。我遇到的问题是,在11,800多个函数调用之后,Node会抛出错误并退出进程。

The error says: RangeError: Maximum call stack size exceeded

错误说明:RangeError:超出最大调用堆栈大小

I'm curious if anyone else has had this situation arise and to know how they handled this. I've already started to break up my code into a couple of extra worker files but even so each time I process a data node it needs to touch 2 databases (at most 25 calls to update various tables) and do a number of sanitization checks.

我很好奇是否有其他人已经出现这种情况,并知道他们是如何处理这个问题的。我已经开始将我的代码分解为几个额外的工作文件,但即便如此,每次处理数据节点时,它都需要触摸2个数据库(最多25次调用以更新各种表)并进行一些清理检查。

I am totally willing to admit that I'm possibly doing something non-optimal if that is the case but would appreciate some guidance if there is a more optimal manner.

我完全愿意承认,如果是这种情况,我可能会做一些非最优的事情,但如果有更优化的方式,我会很感激。

Here is an example of the code I'm running on data:

以下是我在数据上运行的代码示例:

app.post('/initspeaker', function(req, res) {
    // if the Admin ID is not present ignore
    if(req.body.xyzid!=config.adminid) {
        res.send( {} );
        return;
    }

    var gcnt = 0, dbsize = 0, goutput = [], goutputdata = [], xyzuserdataCallers = [];

    xyz.loadbatchfile( xyz.getbatchurl("speakers", "csv"), function(data) {
        var parsed = csv.parse(data);
        console.log("lexicon", parsed[0]);

        for(var i=1;i<parsed.length;i++) {
            if(typeof parsed[i][0] != 'undefined' && parsed[i][0]!='name') {
                var xyzevent = require('./lib/model/xyz_speaker').create(parsed[i], parsed[0]);
                xyzevent.isPresenter = true;
                goutput.push(xyzevent);
            }
        }
        dbsize = goutput.length;

        xyzuserdataCallers = [new xyzuserdata(),
                                    new xyzuserdata(),
                                    new xyzuserdata(),
                                    new xyzuserdata(),
                                    new xyzuserdata(),
                                    new xyzuserdata(),
                                    new xyzuserdata(),
                                    new xyzuserdata()
                                ];
        // insert all Scheduled Items into the DB                   
        xyzuserdataCallers[0].sendSpeakerData(goutput[0]);
        for(var i=1;i<xyzuserdataCallers;i++) {
            xyzuserdataCallers[i].sendSpeakerData(8008);
        }

        //sendSpeakerData(goutput[0]);
    });

    var callback = function(data, func) {
        //console.log(data);
        if(data && data!=8008) {
            if(gcnt>=dbsize) {
                res.send("done");
            } else {
                gcnt++;
                func.sendSpeakerData(goutput[gcnt]);
            }
        } else {
            gcnt++;
            func.sendSpeakerData(goutput[gcnt]);
        }
    };

    // callback loop for fetching registrants for events from SMW
    var xyzuserdata = function() {};
    xyzuserdata.prototype.sendSpeakerData = function(data) {
        var thisfunc = this;

        if(data && data!=8008) {
            //console.log('creating user from data', gcnt, dbsize);
            var userdata = require('./lib/model/user').create(data.toObject());
            var speakerdata = userdata.toObject();
            speakerdata.uid = uuid.v1();
            speakerdata.isPresenter = true;

            couchdb.insert(speakerdata, config.couch.db.user, function($data) {
                if($data==false) {
                    // if this fails it is probably due to a UID colliding
                    console.log("*** trying user data again ***");
                    speakerdata.uid = uuid.v1();
                    arguments.callee( speakerdata );
                } else {
                    callback($data, thisfunc);
                }
            });
        } else {
            gcnt++;
            arguments.callee(goutput[gcnt]);
        }
    };

});

A couple of classes and items are defined here that need some introduction:

这里定义了几个类和项目需要一些介绍:

• I am using Express.js + hosted CouchDB and this is responding to a POST request

我正在使用Express.js +托管的CouchDB,这是响应POST请求

• There is a CSV parser class that loads a list of events which drives pulling speaker data

有一个CSV解析器类可以加载一个驱动扬声器数据的事件列表

• Each event can have n number of users (currently around 8K users for all events)

每个活动可以有n个用户(目前所有活动的用户约为8K)

• I'm using a pattern that loads all of the data/users before attempting to parse any of them

我正在使用一种模式,在尝试解析任何数据/用户之前加载所有数据/用户

• Each user loaded (external data source) is converted into an object I can use and also sanitized (strip slashes and such)

每个加载的用户(外部数据源)都会转换为我可以使用的对象,也会被清理(条带斜线等)

• Each user is then inserted into CouchDB

然后将每个用户插入CouchDB

This code works in the app but after a while I get an error saying that over 11,800+ calls have been made and the app breaks. This isn't an error that contains a stack trace like one would see if it was code error, it is exiting due to the number of calls being done.

此代码在应用程序中有效,但过了一段时间后,我收到一条错误消息,说已经有超过11,800多个来电并且应用程序中断了。这不是包含堆栈跟踪的错误,如果它是代码错误就会看到它,由于调用次数正在退出。

Again, any assistance/commentary/direction would be appreciated.

再次,任何协助/评论/指示将不胜感激。

2 个解决方案

#1

It looks like xyzuserdata.sendSpeakerData & callback are being used recursively in order to keep the DB calls sequential. At some point you run out of call stack...

它看起来像递归使用xyzuserdata.sendSpeakerData和回调,以保持DB调用顺序。在某些时候你用完了电话堆栈......

There's several modules to make serial execution easier, like Step or Flow-JS.

有几个模块可以简化串行执行,比如Step或Flow-JS。

Flow-JS even has a convenience function to apply a function serially over the elements of the array:

Flow-JS甚至还具有一个便利功能,可以在数组元素上串行应用函数:

flow.serialForEach(goutput, xyzuserdata.sendSpeakerData, ...)

I wrote a small test program using flow.serialForEach, but unfortunately was able to get a Maximum call stack size exceeded error -- Looks like Flow-JS is using the call stack in a similar way to keep things in sync.

我使用flow.serialForEach编写了一个小测试程序,但遗憾的是能够获得超出最大调用堆栈大小的错误 - 看起来像Flow-JS以类似的方式使用调用堆栈来保持同步。

Another approach that doesn't build up the call stack is to avoid recursion and use setTimeout with a timeout value of 0 to schedule the callback call. See http://metaduck.com/post/2675027550/asynchronous-iteration-patterns-in-node-js

另一种不构建调用堆栈的方法是避免递归并使用超时值为0的setTimeout来调度回调调用。见http://metaduck.com/post/2675027550/asynchronous-iteration-patterns-in-node-js

You could try replacing the callback call with

您可以尝试使用替换回调调用

setTimeout(callback, 0, [$data, thisfunc])

#2

Recursion is very useful for synchronizing async operations -- that's why it is used in flow.js etc.

递归对于同步异步操作非常有用 - 这就是它在flow.js等中使用的原因。

However if you want to process an unlimited number of elements in an array, or buffered stream, you will need to use node.js's event emitter.

但是,如果要在数组或缓冲流中处理无限数量的元素,则需要使用node.js的事件发射器。

in pseudo-ish-code:

 ee = eventemitter
 arr = A_very_long_array_to_process
 callback = callback_to_call_once_either_with_an_error_or_when_done

 // the worker function does everything
 processOne() {
   var 
      next = arr. shift();
   if( !arr )
      ee.emit ( 'finished' )
      return

   process( function( err, response) {
      if( err )
         callback( err, response )
      else
         ee.emit( 'done-one' )
    } );
 }    

 // here we process the final event that the worker will throw when done
 ee.on( 'finished', function() { callback( null, 'we processed the entire array!'); } );

 // here we say what to do after one thing has been processed
 ee.on( 'done-one', function() { processOne(); } );

 // here we get the ball rolling
 processOne();