问题描述
我正在编写一个备份脚本，该脚本仅下载特定Azure帐户的所有blob容器中的所有blob.
I am writing a backup script that simply downloads all the blobs in all the blob containers of a specific Azure account.
该脚本使用async.js来确保只能同时运行那么多线程，因此不会使服务器超载.当我运行此脚本时，它可以正常运行，但是当它击中大型文件时，它将用尽内存.我猜下载的运行速度快于磁盘可以写入的速度，并且最终填充内存的缓冲区非常严重，以至于我完全耗尽了内存，但是到目前为止，调试确切的原因是不可能的.
The script uses async.js to make sure only so much threads can run at the same time so it doesn't overload the server. When I run this script it works fine, but when it hits large files it runs out of memory. I'm guessing the download runs faster than the disk can write, and it eventually fills up the in-memory buffer so badly that I run out of memory entirely, but debugging the exact cause has been impossible so far.
似乎使用大量内存的特定功能称为:
The specific function which appears to use a lot of memory is called as follows:
blobService.getBlobToStream(
containerName,
blob.name,
fs.createWriteStream(fullPath),
function(error) {
if(error){ //Something went wrong, write it to the console but finish the queue item and continue.
console.log("Failed writing " + blob.name + " (" + error + ")");
callback();
}
else if(!error) { //Write the last modified date and finish the queue item silently
fs.writeFile(fullPath + ".date", blobLastModified, function(err)
{ if(err) console.log("Couldn't write .date file: " + err); });
callback();
}
});
即使只有700MB的下载量，也很容易占用我这一边的1GB内存.
Even a single 700MB download will easily fill up 1GB of memory on my side.
有什么办法解决吗?我是否缺少一个可以神奇地阻止Azure SDK缓冲所有内容和厨房水槽的参数?
Is there any way around this? Am I missing a parameter which magically prevents the Azure SDK from buffering everything and the kitchen sink?
完整代码:
#!/usr/bin/env node
//Requires
var azure = require('azure');
var fs = require('fs');
var mkdirp = require('mkdirp');
var path = require('path');
var async = require('async');
var maxconcurrency = 1; //Max amount of simultaneous running threads of getBlobsAndSaveThem() running through async.js.
var blobService = azure.createBlobService();
backupPrefix='/backups/azurebackup/' //Always end with a '/'!!
//Main flow of the script is near the bottom of the file.
var containerProcessingQueue = async.queue(
function getBlobsAndSaveThem(containerName) {
console.log(containerName); //DEBUG
blobService.listBlobs(containerName,
function(error, blobs) {
if(!error){
var blobProcessingQueue =
async.queue(function(index,callback) {
var blob = blobs[index];
console.log(blob); //DEBUG
var fullPath = backupPrefix + containerName + '/' + blob.name;
var blobLastModified = new Date(blob.properties['last-modified']);
//Only create if the directoy doesn't exist, since mkdirp fails if the directory exists.
if(!fs.existsSync(path.dirname(fullPath))){ //And do it sync, because otherwise it'll check 99999 times if the directory exists simultaneously, doesn't find it, then fails to create it 99998 times.
mkdirp.sync(path.dirname(fullPath), function(err) { console.log('Failed to create directory ' + path.dirname(fullPath) + " ("+ err + ")"); });
}
if(fs.existsSync(fullPath + ".date")){
if(blobLastModified == fs.readFileSync(fullPath + ".date").toString()) {
callback();
return; //If the file is unmodified, return. No this won't exit the program, because it's called within a function definition (async.queue(function ...))
}
}
blobService.getBlobToStream(
containerName,
blob.name,
fs.createWriteStream(fullPath),
function(error) {
if(error){ //Something went wrong, write it to the console but finish the queue item and continue.
console.log("Failed writing " + blob.name + " (" + error + ")");
callback();
}
else if(!error) { //Write the last modified date and finish the queue item silently
fs.writeFile(fullPath + ".date", blobLastModified, function(err)
{ if(err) console.log("Couldn't write .date file: " + err); });
callback();
}
});
},maxconcurrency);
for(var blobindex in blobs){
blobProcessingQueue.push(blobindex);
} //Push new items to the queue for processing
}
else {
console.log("An error occurred listing the blobs: " + error);
}
});
},1);
blobService.listContainers(function(err, result){
for(var i=0;i<result.length;i++) {
containerProcessingQueue.push(result[i].name);
}
});
推荐答案
您可能要做的一件事是，仅将一部分数据读取到流中，而不是整个blob数据，将其附加到文件中并读取下一块.Blob存储服务支持该功能.如果您查看 getBlobToStream 的源代码( https://github.com/WindowsAzure/azure-sdk-for-node/blob/master/lib/services/blob/blobservice.js )，您可以指定选项中的from/to字节- rangeStartHeader 和 rangeEndHeader .看看是否有帮助.
One thing that you could possibly do is read only a chunk of data into stream instead of whole blob data, append that to the file and read next chunk. Blob Storage service supports that. If you look at the source code for getBlobToStream (https://github.com/WindowsAzure/azure-sdk-for-node/blob/master/lib/services/blob/blobservice.js), you can specify from/to bytes in the options - rangeStartHeader and rangeEndHeader. See if that helps.
我已经破解了一些可以做到这一点的代码(正如您从我的代码中看到的那样，我对node.js的了解非常原始:).[请使用此代码只是为了了解如何进行分块下载，因为我认为它仍然存在一些故障]
I have hacked some code which does just that (as you can see from my code, my knowledge about node.js is quite primitive :)). [Please use this code just to get an idea about how you can do chunked download as I think it still has some glitches]
var azure = require('azure');
var fs = require('fs');
var blobService = azure.createBlobService("account", "accountkey");
var containerName = "container name";
var blobName = "blob name";
var blobSize;
var chunkSize = 1024 * 512;//chunk size -- we'll read 512 KB at a time.
var startPos = 0;
var fullPath = "D:\\node\\";
var blobProperties = blobService.getBlobProperties(containerName, blobName, null, function (error, blob) {
if (error) {
throw error;
}
else {
blobSize = blob.contentLength;
fullPath = fullPath + blobName;
console.log(fullPath);
doDownload();
}
}
);
function doDownload() {
var stream = fs.createWriteStream(fullPath, {flags: 'a'});
var endPos = startPos + chunkSize;
if (endPos > blobSize) {
endPos = blobSize;
}
console.log("Downloading " + (endPos - startPos) + " bytes starting from " + startPos + " marker.");
blobService.getBlobToStream("test", blobName, stream,
{ "rangeStartHeader": startPos, "rangeEndHeader": endPos-1 }, function(error) {
if (error) {
throw error;
}
else if (!error) {
startPos = endPos;
if (startPos <= blobSize - 1) {
doDownload();
}
}
});
}
这篇关于节点js azure SDK getBlobToStream使用大量内存的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！