I'm having my project for my "distributed system development" class, and my project is making a minimal version of cloud storage (sth like Google Drive).
my approach here is using 2 backend server written in Rails, with 1 proxy server to control requests sent to servers, 2 two Postgres server with master-slave replication relationship.
But problem here is how to storage real assets (video, pdf , mp3 ...). I have no experience in this.
example:
if 1 user opens 2 browser tab, then in each tab he uploads 1 video with same name to 1 directory, what will happend?
Since you probably want to upload asynchronously, this is pretty easy to handle: generate some sort of token before uploading (i.e. filename + hash), then hand the upload off to the delayed job. If the user tries uploading the second file, it will generate the same token and be rejected.
Example for keeping track of the uploads in the DB. Generate a record before upload starts and save the filename and the hash.
Asset.create(filename: ..., hash: ...)
Once the upload finishes you can update the record with the S3 URL or whatever you use for storage (pass the asset id to the delayed job). The validation then is easy:
validates uniqueness: { scope: :filename }
Related
The problem at hand
I have a rails app.
Users will be uploading files. Anywhere between 1 file to 3000 files. Sometimes they are zip files, and sometimes they are not. I do not want hold up the server with these files uploads, so I am looking for a solution around this problem.
The zipped files will have to be unzipped.
I then want to check whether: the user has previously uploaded the same files? i.e. if the user has already uploaded the same file(2) one week ago, then this is a problem: (i) either we don’t allow that particular file to be uploaded, or we ask the user: are you sure you want to upload the same file again?
Then I want to store the keys/links to the files within the appropriate models/records on the back end.
Was wondering what the best workflow for handling the above could be: i.e a very general overview: in other words, could AWS Lambda / Google cloud computing etc. etc be best employed to handle the above problem? How would we use the Shrine gem, to best handle this situation? Would it make sense to use AWS Lambda rather than using background jobs?
My preferences are to use the Shrine gem for uploading.
My Ideas:
In the client side, the user drags and drops the files the user
wants to upload.
All the files are then uploaded (whether zipped or otherwise) to a temporary bucket location via the Shrine gem.
IF the zip files are uploaded then perhaps an AWS lambda function must be triggered to unzip the files. If that’s the case,then at the end of the day, the keys for these files must somehow be returned to the client, to handle validation issues – but then how would the AWS lambda function be able to return this request to the original client side where the request was originated? Or rather,should the AWS lambda function be generated from the client side,passing in the IDs of the unzipped blobs?
Then we need to run some validations: we want to handle the situation where there are duplicate files. We will need to check with our rails backed as to whether those files have already been uploaded.
After those validation issues are handled, then user submits the form, and all the keys are stored within the appropriate records.
These ideas are by no means prescriptive
Am seeking some very general advise on what the best way is of doing this all. I am by no means constrained to AWS: I could use Google or Azure just as easily. Any guidance on the above would be much appreciated.
Specific questions:
How would the AWS lambda function get triggered?
How would be be able to return the keys of the uploaded files back to the client?
What do I mean by general overview?
Here are some examples of general overviews:
(1) Uploading & Unzipping files to S3 through Rails hosted on Heroku?
(2) https://www.quora.com/How-do-I-extract-large-zip-files-in-AWS-Lambda
Any pointers in the right direction would be much appreciated.
Cheers!
This isn't a really difficult problem to solve if you are willing to change the process flow a little bit.
In the client side, the user drags and drops the files the user wants to upload.
When the user requests the upload operation to begin you can make HTTP GET requests to an API Gateway endpoint, backed with a Lambda. The Lambda can query for previous files uploaded by the client and send back a result set showing what files already exist. You then filter those out and send only what is considered new from the client to the server. This will save the user time in waiting for the upload to happen and save you time on the S3/Lambda side of not having to store duplicates or process them. This isn't a substitute for server-side validation though, you'll still want to do that. For legit clients, this will save you and them a lot of bandwidth and storage.
All the files are then uploaded (whether zipped or otherwise) to a temporary bucket location via the Shrine gem.
This works. As they enter the temp bucket, use a Lambda with an S3 event to process the files, unzip files, push any metadata needed into DynamoDb and delete the files from the temp bucket. In the temp bucket, I would place the files into a folder that is unique per request and user. I would take the user/client Id and a UUID of some kind and make that your folder name. Such as Johnathon+3b5339b8-c8db-4d5c-b678-406fcf073f4f, or encode this value into a Base64 string and make that your folder name. Store this in DynamoDb with each file uploaded into your permanent bucket with the Hash Key being the userid/clientid, a Sort Key being the full folder path + file name and an extra attribute of IsProcessed. The IsProcessed attribute will be updated by your Lambda that is processing the files and moving them to their permanent S3 bucket. If there are errors, you can put the error in this field. If it is successful then you put it in this field.
the keys for these files must somehow be returned to the client, to handle validation issues – but then how would the AWS lambda function be able to return this request to the original client side where the request was originated? Or rather,should the AWS lambda function be generated from the client side,passing in the IDs of the unzipped blobs?
The original API request to push the files to the temp S3 bucket would be able to return back to the client the folder name johnathon+3b5339b8-c8db-4d5c-b678-406fcf073f4f to the client. So let's say you made a HTTP POST to /jobs. You would return back 201 Created with a HTTP Header of Location /jobs/johnathon+3b5339b8-c8db-4d5c-b678-406fcf073f4f. Your client can then start polling /jobs/johnathon+3b5339b8-c8db-4d5c-b678-406fcf073f4f for the status of the process.
Your response back to /jobs/johnathon+3b5339b8-c8db-4d5c-b678-406fcf073f4f can return the DynamoDB records. This would include all DynamoDB records for the HashKey matching the folder name. Your client side can look at all of the objects in the result set and check the IsProcessed attribute to see if everything worked out ok, or if there were issues.
Then we need to run some validations: we want to handle the situation where there are duplicate files. We will need to check with our rails backed as to whether those files have already been uploaded.
Handle this with the Lambda that is executed by the temporary bucket. Grab the files from the temp bucket folder, handle your business logic and back-end queries then push them to their final permanent bucket.
After those validation issues are handled, then user submits the form, and all the keys are stored within the appropriate records.
All of this would happen asynchronously, starting when the user submits the form. The client side needs to be able to handle this by making HTTP GET requests to the endpoint mentioned above, checking for the status of the process. This gives you some more flexibility as you can also publish SNS messages on failures as well, such as sending an email to the clients if they upload 3,000 files and you need to spend 30 minutes processing them.
I am working on a project where the user joins a "stream". During stream setup, the person who is creating the stream (the stream creator) can choose to either:
Upload all photos added to the stream by members to our hosting solution (S3)
Upload all photos added to the stream by members to the stream creator's own Dropbox authenticated folder
In the future I would like to add more storage providers (such as Drive, Onesky etc)
There is a couple of different questions I have in regards to how to solve this.
What should the structure be in the database for photos? I currently only have photo_url, but that won't be easy to manage from a data perspective with pre-signed urls and when there are different ways a photo can be uploaded (s3, dropbox etc.)
How should the access tokens for each storage provider be stored? Remember that only the stream creator's access_token will be stored and everyone who is on the stream will share that token when uploading photos
I will add iOS and web clients in the future that will do a direct upload to the storage provider and bypass the server to avoid a heavy load on the server
As far as database storage, your application should dictate the structure based on the interface that you present both to the user and to the stream.
If you have users upload a photo and they don't get to choose the URI, and you don't have any hierarchy within a stream, then I'd recommend storing just an ID and a stream_id in your main photo table.
So at a minimum you might have something looking like
create table photos(id integer primary key, stream_id integer references streams(id) not null);
But you probably also want description and other information that is independent of storage.
The streams table would have all the generic information about a stream, but would have a polymorphic association to a class dependent on the type of stream. So you could use that association to get an instance of S3Stream or DropBoxStream based on what actual stream was used.
That instance (also an ActiveRecord resource) could store the access key, and for things like dropbox, the path to the folder etc. In addition, that instance could provide methods to construct a URI given your Photo object.
If a particular technology needs to cache signed URIs, then say the S3Stream object could reference a S3SignedUrl model where the URIs are signed.
If it turns out that the signed URL code is similar between DropBox and S3, then perhaps you have a single SignedUrl model.
When you design the ios and android clients, it is critical that they are not given access to the stream owner's access tokens. Instead, you'll need to do all the signing inside your server app. You wouldn't want a compromise of a device to lead to exposing the access token creating billing problems as well as privacy exposures.
Hope this helps.
we setup a lot of rails applications with different kind of file storages behind it.
Yes, just an url is not manageable in the future. To save a lot of time you could use gems like carrierwave or paperclip. They handle all the thumbnail generation and file validation. One approach is, that you could upload the file from the client directly to S3 or Dropbox to a tmp folder and just tell your Rails App "Hey, here is the url of a new upload file" and paperclip and carrierwave will take care of the thumbnail generation and storaging. (Example for paperclip)
Don't know exactly how your stream works, so I cannot give a good answer to this -.-
With the setup I mentioned in 1. you should upload form your different clients directly to S3 or Dropbox etc. and after uploading, the client tells the Rails Backend that it should import the file from that url. (And before paperclip or carrierwave finish their processing you could use the tmp url from the file to display something directly in your stream)
Unfortunately the bounty was awarded to an answer that does not solve this problem, for those with similar issues.
I have a form with an image upload (heroku to s3). When I submit the form, my rails server waits for the background job that uploads the image to finish before returning a response to the user. This causes an application timeout every single time there is an image upload.
Current order of events:
User submits form
Server receives form
If there is an image, server starts a background job
If a background job was started, the server waits for it to complete (rails times out here)
If started, the background job completes
The server processes the request
The server responds to the user
Desired order of events:
User submits form
Server receives form
Server processes non-image fields
If there is an image, server starts a background job
The server responds to the user
The background job completes and the server processes the uploaded image (saves URL)
Uploader code
class PhotoUploader < CarrierWave::Uploader::Base
include ::CarrierWave::Backgrounder::Delay
include CarrierWave::MimeTypes
process :set_content_type
storage :fog
end
Carrierwave::Backgrounder initializer
CarrierWave::Backgrounder.configure do |c|
c.backend :sidekiq, queue: :carrierwave
end
User model
class User < ActiveRecord::Base
mount_uploader :photo, PhotoUploader, delayed: true
process_in_background :photo
end
There is no controller code because the form is handled by ActiveAdmin. I can override wherever is needed, but have not been able to figure out what needs to change.
What do I have to change to get the correct order of events?
The underlying issue here is Heroku has strict limits on how long a request may block without sending data back to the client. If you hit that limit (30 seconds for the initial byte), Heroku will timeout your request. For file uploads, it is very likely you will hit this limit.
The best approach is to have the user's browser directly upload the file to S3 first. There is some discussion here that is relevant to this: Direct Uploads to S3 using Carrierwave
If you use something like the jQuery File Upload plugin (https://github.com/blueimp/jQuery-File-Upload), the flow would be something like:
User adds one or more files to the form before clicking submit.
The files are uploaded directly to S3 and a file upload token is added for each file to your form.
User submits form with the tokens of the uploaded files rather than the contents of the files.
The server can move the files to their real home in S3 based on the submitted token.
This allows your web server to focus on just serving the request and not block on file uploads which can take a long time.
This requires a bit more work due to Heroku's limits - but ultimately I think is your only option to avoid the timeout limits.
Also, I'd recommend you create an upload S3 bucket and then set an S3 lifecycle policy to purge files older than some interval. When you are doing direct file uploads, it is common for some uploads to not be processed due to the user giving up etc, so the lifecycle does the job of cleaning up these files.
I have a Rails app that catalogues recorded music products with metadata & wav files.
Previously, my users had the option to send me files via ftp, which i'd monitor with a cron task for new .complete files and then pick it's associated .xml file and a perform metadata import and audio file transfer to S3.
I regularly hit capacity limits on the prior FTP so decided to move the user 'dropbox' to S3, with an FTP gateway to allow users to send me their files. Now it's on S3 and due to S3 not storing the object in folders i'm struggling to get my head around how to navigate the bucket, find the .complete files and then perform my imports as usual.
Can anyway recommend how to 'scan' a bucket for new .complete files.....read the filename and then pass back to my app so that I can then pick up it's xml, wav and jpg files?
The structure of the files in my bucket is like this. As you can see there are two products here. I would need to find both and import their associated xml data and wavs/jpg
42093156-5060156655634/
42093156-5060156655634/5060156655634.complete
42093156-5060156655634/5060156655634.jpg
42093156-5060156655634/5060156655634.xml
42093156-5060156655634/5060156655634_1_01_wav.wav
42093156-5060156655634/5060156655634_1_02_wav.wav
42093156-5060156655634/5060156655634_1_03_wav.wav
42093156-5060156655634/5060156655634_1_04_wav.wav
42093156-5060156655634/5060156655634_1_05_wav.wav
42093156-5060156655634/5060156655634_1_06_wav.wav
42093156-5060156655634/5060156655634_1_07_wav.wav
42093156-5060156655634/5060156655634_1_08_wav.wav
42093156-5060156655634/5060156655634_1_09_wav.wav
42093156-5060156655634/5060156655634_1_10_wav.wav
42093156-5060156655634/5060156655634_1_11_wav.wav
42093163-5060243322593/
42093163-5060243322593/5060243322593.complete
42093163-5060243322593/5060243322593.jpg
42093163-5060243322593/5060243322593.xml
42093163-5060243322593/5060243322593_1_01_wav.wav
Though Amazon S3 does not formally have the concept of folders, you can actually simulate folders through the GET Bucket API, using the delimiter and prefix parameters. You'd get a result similar to what you see in the AWS Management Console interface.
Using this, you could list the top-level directories, and scan through them. After finding the names of the top-level directories, you could change the parameters and issue a new GET Bucket request, to list the "files" inside the "directory", and check for the existence of the .complete file as well as your .xml and other relevant files.
However, there might be a different approach to your problem: did you consider using SQS? You could make the process that receives the uploads post a message to a queue in SQS, say, completed-uploads, with the name of the folder of the upload that just completed. Another process would then consume the queue and process the finished uploads. No need to scan through the directories in S3.
Just note that, if you try the SQS approach, you might need to be prepared for the possibility of being notified more than once of a finished upload: SQS guarantees that it will eventually deliver posted messages at least once; you might receive duplicated messages! (you can identify a duplicated message by saving the id of the received message on, say, a consistent database, and checking newly received messages against the same database).
Also, remember that, if you use the US Standard Region for S3, then you don't have read-after-write consistency, you have only eventual-consistency, which means that the process receiving messages from SQS might try to GET the object from S3 and get nothing back -- just try again until it sees the object.
Can one of the Amazon services (their S3 data service, or otherwise) be used to offload server of static files for a Ruby on Rails app, but still support the app's authentication & authorization?
That is such that when the user browser downloaded the initial HTML for one page of the Ruby on Rails application, when it went back for static content (e.g. an image or CSS file), that this request would be:
(a) routed directly to the Amazon service (no RoR cycles used to serve it, or bandwidth), BUT
(b) the browser request for this item (e.g. an image) would still have to go through an authentication/authorization layer based on the user model in the Ruby on Rails application - in other words to ensure not just anyone could get the image...
thanks
The answer is a yes with a but. You can use a feature of S3 that allows you to create links to secure S3 objects that has a small time to live, default is 5 minutes. This will work for any S3 object that is uploaded as private. This means that the browser will only have X seconds or whatever to request the file from S3. Example code from docs for the AWS gem:
S3Object.url_for('beluga_baby.jpg', 'marcel_molina')
You can also specify an expires_in or expires option per file. The bad thing is that you would need to create a helper for your stylesheet, image, and js links to create the proper S3 URLs.
I would recommend that you setup a domain name for your S3 bucket, like "examples3.amazonaws.com" and put all your standard image files and CSS there as public. Then set that as the asset host in your rails config. Then, only use the secure links for static files that really need it.