I am building a node.js application. When a user connects, they transmit their state as online. When they disconnect, the state gets switched to offline. Also, when a user connects, he gets a list of his connections, along with the states of these friends(online or offline). Ok, great, no problem with this.
But the trouble I'm running into is how to dynamically send the state of a user's freind while I'm logged in. So, the steps would look like:
I log in.
I get a list of all my friends: user1, user2, user3
I get a list of states for all users: user1:online, user2:online user3:offline
user1 logs off
How do I dynamically get the state change of user 1?
Remember that these users aren't connected to each other, and also remember that a user's state needs to get transmitted to all the other user's who have him as a friend. Let's say that user1 has 100 friends, I am one of them. I (along with the other 99) need to get the notification that user1 is now offline. Also note that these people may not be on the same physical server (multiple servers handling X number of connections each). What is the best theoretical way to solve this problem?
The only answer that I could come up with is that I save the state in a DB, and that node will query the db, and when the state changes, it will update. But it seems crazy that node would have to watch for changes to the db, that can't scale. There's got to be a better way to do this - maybe its an obvious one and I'm missing it.
As hinted by ebohlman in the comments, the publish/subscribe pattern (also known as "pub/sub") sounds applicable to your problem. In a pub/sub architecture, instead of the publisher sending messages to specific clients, the publisher instead sends messages to some central pub/sub server. Subscribers notify this server for the kinds of messages, or "topics," they're interested in. The key is that the publishers and the subscribers never know, or even care, about each other, making the pattern very scalable (there can be any number of publishers or subscribers).
In your case, each of the servers that handle user login/logout might subscribe to the "login_changed" topic. Whenever a user logs in or out, the server that processed the change would send a message to the pub/sub server with the topic "login_changed" along with some data indicating what happened to whom (e.g. with a JSON-encoded string). Each of the servers that have subscribed to the "login_changed" topic will receive the message, and they can each determine if they're responsible for tracking the user in question.
There are several common tools for managing the pub/sub pattern. Redis, a common key-value store, has support for pub/sub and is a personal favorite tool of mine for the task. AMQP is an open standard that includes many kinds of messaging-related behavior, including pub/sub, and has many implementations, the most common of which is probably RabbitMQ. If you're interested in hooking up web clients over HTTP, you might be interested in something like Faye. There's a lot of literature on the subject, as well as many implementations in many languages; a Google search should take you far.
Related
This might be more of a question towards push messages in general for the web, but I'm having trouble sending messages to users using Firebase Cloud Messaging (FCM). The issue arises in the boundary case that there are multiple accounts or users on a single browser. Generally, people don't share their phones, so this wouldn't be an issue for Android/iOS, and I think it's less common now, but people still do share their computers, either with their families or maybe at an internet cafe or something.
Currently, every time somebody who is logged in enables notifications, I add that push token for that device to an array in a database, then send a notification to all of those tokens when I send a push message to that user. Although this is fine for mobile and most computers, what about shared computers? Other users of that computer would keep getting notifications about things that are irrelevant to them. If this is a family computer, I guess it's still alright, but what about at a web cafe if somebody uses it once then never again? Would that not be incredibly irritating for everyone else? Also, what if there are multiple different users on the same device that all use notifications? Then, both users would get notifications for both users. This is not good either.
I can't think of any way to circumvent this problem gracefully. Maybe I can store the userID of the current logged in user in IndexedDB, replace it whenever a new user logs in instead, and only pop up the notifications if the userIDs match? I could also add an "expiration date" so that the userID in IndexedDB times out after a while if the user doesn't log in for let's say 2 weeks.
Then again, it doesn't completely solve the problem either, it just makes it slightly less annoying, and also adds in other issues with multiples users with an account on the same device as well (ie: receiving the other person's notifications but not your own in certain scenarios). Also, I would have to switch around my entire backend as well since I'm currently sending messages to FCM topics which users subscribe to, hence can't send specialized messages to certain tokens containing the user IDs.
I guess I could also use a "dirty fix" where I simply warn users that notifications are sent to the DEVICE, not the USER, and to be thoughtful when enabling this on shared computers.
Has anyone encountered (or thought of this problem) before and come up with a reasonable solution? I would be very happy to have suggestions. I think I'll go with the "dirty fix" for now, and might consider doing something else later if I can come up with a reasonable solution. Of course, this is probably only an issue that affects ~1% of users at most anyway, so maybe I'm just thinking too much, and fixing this isn't worth the effort.
I'm using mqtt.js and mosca. I would like to know how many clients subscribed to a topic. How can i achive this goal?
At a MQTT protocol level you can not do this, there is no way for a publishing client to know if there are any subscribers to a given topic.
While you probably could implement something with the mosca hooks and callbacks and work out a way to publish the information under the $sys topic space, you will still need to not only count clients that have subscribed to the topic explicitly but also matching wildcard subscriptions.
This get's tricky because they only get matched at the point a message is published, so really at best you could only count the number of clients a message was delivered to at the point of publishing. This then ignores any clients with QOS 1/2 subscriptions that are offline at the time of publishing and the message will be queued for.
All in all you need to think why you want this number as it normally goes against the concept of pub/sub messaging for the publisher to care if there are any subscribers.
I have built a web application using AngularJS (front-end) and PHP/MySQL (back-end).
I was wondering if there is a way to "watch" the MySQL database (without Node.js), so if one user adds some data to it, the changes are synced to other users too.
E.g. I know Firebase does that, but it's object oriented database and I am unable to do the advanced queries there like I do with SQL.
I was thinking to use $interval and $http and do ajax requests, so that way I could detect changes in the database. Well, that's possible, but it'll then do thousands of http requests to the server everyday and plus interpret php on each request.
I believe nothing is impossible, I just need an idea to do this, which I don't have, so that's why I am asking for a help here.
If you want a form of "real-time communication" you'll likely have to incorporate some form of long-polling from the client. Unless you use web sockets, but that's a big post about a bunch of different things. You're right to be concerned about bandwidth and demand on the DB though. So here's my suggestion:
If you don't have experience with web sockets then log your events in a separate table/view and use the pub/sub method to subscribe entities to an event, and broadcast that event to the table. Then long-poll against the watcher view to see when changes may have occurred. If one did occur then you query for the exact value.
Another option would be to use some query system with "deciders" that hold messages. Take a look at Amazon's SQS platform for a better explanation of how this could work. Basically you have a queue that holds messages and a decider chooses where to store the message using some hash or sorting method (to reduce run time). When the client requests an update, the decider finds any messages that would apply based on the hash/sort and returns them. Then you just have to decide how and when to destruct the messages.
The second option would require a lot more tinkering though, so it's really about your preference. I think what you'll find the difficulty to be is that most solutions have to deal with the fact that the message has to be delivered 1 or More times and you'll need to track when someone received the message and if it can now be deleted from the queue/event table or if you still need to wait. Otherwise you'll consume a lot of memory.
I'm working on a pretty fun web app project that can become rather big, and I have a chance to play around with this handy thing called PubNub as main real-time engine of application.
So it's a web application with Node.js backend, involves potentially huge amount of chat rooms between users and realtime notifications sent to users by backend when some data in DB is updated.
Usually, developing with Sockets.io, I will just subscribe each user to channel of his unique DB id, and also to chanels representing different chat rooms.
This way I can handle chat rooms and authentication on backend and after storing some personal notification in DB I can easily push them to channel named by user id, so if user is online - he gets it, if not - fine, he will see it on next login, notification is already in DB. And theoretically this monstrocity should scale just fine horizontally with help of redis pub/sub.
Thing that worries me about PubNub in this case is scalability. As I obviously have no insight on what is going on in PubNub backend's dark corners, I want to make sure that app is built in the way that it will be prepared to handle some obscure enormously huge amount of simultaneous users.
My question is, what is the best approach to building such a system with PubNub?
Am I correct assuming that it will be better, in need of pushing notification to specific user, to subscribe to this user's pubnub, push note and unsubscribe. As if I will keep all online-user channels open - then there is no point in PubNub instead of websockets on my server, as server will be anyway under load of all of those opened online-user channels and should be scaled just to maintain huge quantity of them.
What about user authorisation? Without involving my backend how can I be sure that user posting some message will not be able to fake his personality and will have exatly the same as he have authenticated inside application?
And generally (and via PubNub) what is the best practice to tackle huge amounts of chats per users? As say during application life each user may accumulate some decent amount of garbage chat rooms that have some users in it, though havn't been touched by anyone for a long time, and users just way too lazy to leave it manually?
Thanks for Your patience in reading this wall of text!
UPDATED Dec 5, 2021
If you are implementing a chat app, please refer to the PubNub Chat use-case documentation for full details. It has new features and UI components that are built upon the PubNub Platform.
UPDATED May 15, 2020
We have some new docs that will explain much of the below in much clearer terms.
And new features that can be applied to many of the questions/answers below:
Message Actions
Message Counts
Batch History (multi-channel message fetch)
Objects (Users, Channels and Memberships Metadata)
NOTE: I've sprinkled some of the above links down in the answers below.
First, let's address this...
Thing that worries me about PubNub, in this case, is scalability. As I
obviously have no insight on what is going on in PubNub backend's dark
corners, I want to make sure that the app is built in a way that it will
be prepared to handle some obscure enormously huge amount of
simultaneous users.
and this...
then there is no point in PubNub instead of WebSockets on my server,
as the server will be anyway under the load of all of those opened online-user
channels and should be scaled just to maintain the huge quantity of them
This is sort of backward because you would use a service like PubNub to ensure that your application scales to handle millions of users. PubNub has thousands of customers that scale to millions of users and 100's of billions of messages. Not knowing how PubNub does this frees you to implement the biz logic of your application.
But I think I get what you are saying. You are under the impression that your server has to be involved in each and every chat room interaction for every user, but that is only partially true. Mostly, your server will be used for authentication, some subscription maintenance (optional), and probably for sending messages out to one, many, or all end users, as required (depends on your requirements).
Here are some attempts to answer your questions although they are kind of a bit all over the place so I will do my best to answer what I think it is you are asking.
Question 1
This question seems to be directed at maintaining lots of subscriptions to channels and the scalability of that.
Generally speaking, every end-user initializes PubNub and subscribes to channels they need to listen to and publish to channels they need to send messages on. Typically, the channels (chat rooms in your case, I assume) they are publishing on are the same channels they are subscribing to, but they are different kinds of use cases. And you can subscribe to thousands of channels at a time (up to 20K per client). If you did this with WebSockets, how would you go about scaling this to millions of users? You would implement and operate (to scale) something similar to PubNub (not easy and not cheap).
Now, if a user is subscribed to a bunch of chat room channels but some or many are stale (user hasn't viewed or posted to in a while), you could have some code on your server (or the client) that monitors the activity of users and unsubscribes them from those stale channels. This is possible using channels groups. Each end-user would have their own channel group that contains all the channels they are listening to. And the client code or the server code and add and remove channels to/from those end users' channel groups.
Question 2
UPDATED DOCS: https://www.pubnub.com/docs/platform/security/access-control
Now this question is a bit more clear and focused and is asking about authentication (login) and how to ensure someone is who they say they are and how to handle authorization (what they can and cannot do) and where/who controls this.
The answer is, you control the authentication (login) to prove that the person is what they say they are. Your log-in process checks for a valid username/password and in the user record, you will have a list of access controls for that user. With that, you generate an auth-key that you grant read and/or write access to one or more channels. This grant is a PubNub operation that your server invokes. The auth-key is passed back to the client and the client code initializes PubNub instance using the pub/sub keys and this auth-key that PubNub servers use to check for access based on the channel and the operation being requested (subscribe to this channel, publish to that channel, etc). If the auth-key does not have the proper access, the PubNub server will deny access (403 response).
There's more to all of this but this is a good start. Read up on PubNub Access Manager for the SDK you will be using on our docs page. For example, you can start with the JavaScript SDK Access Manager docs and tutorials.
Question 3
UPDATED DOCS: https://www.pubnub.com/docs/platform/channels/receive#subscribe-to-channels
I believe I answered this sufficiently with question 1 - Channel Groups. Start with the JavaScript SDK Stream Controller (which provides Channel Group feature) docs and tutorials.
I hope I have managed to move you a few steps further along your journey to a highly successful real-time, data stream application using PubNub. Please reply with any additional questions you may still have.
*Answers to your new comments:*
Thanks for your follow-up comments. It is very clear what you are asking now.
I will need to compare chat room timestamp with personal user last-read timestamp for this, so it seems that I need to listen to those channels from back-end and update user's last-reads, or to trust into the front-end, and get timestamps from a user directly
No, you do not have to listen to the channels on your server. Yes, from the client app, you will keep the timestamp of the last received message. When the user comes back online, you use this timestamp to get history for the channels the client was subscribed to. Many have done this successfully and we are going to be releasing some amazing features in the coming months that will simplify this considerably.
pushing real-time notifications to users from the back-end. Do I need to be subscribed to all of my user channels if I want to push notes to them at any time?
You can publish on any channel without actually subscribing to it first. So your server can publish to channels as it needs to.
And as before, keep coming with more questions as you require.
*Great follow-up questions again. Here's what I suggest*
... it makes sense to not request all of those chat rooms from DB and join via pubnub all of them, but rather implement pagination... how user can be aware of new messages that may appear in his old chat rooms?
Again, you can stay subscribed to 20K channels using channel groups. You can subscribe to 10 channel groups with 2K channels per channel group - but I'd recommend just limiting the user to 100 or less because that seems like a sufficient limit to impose in your app. But pick whatever upper limit you want and when the user hits that limit, force them to leave another chat room first or suggest they leave one of the top 10 most inactive, or some algorithm that makes sense for your app.
UPDATED DOCS: https://www.pubnub.com/docs/platform/channels/receive#subscribe-to-channels
Getting the # of missed messages does require a full history fetch, but we are going to be providing improved APIs to make this simpler in the near future. But if the user is registered for push notifications on all these channels, the device would be able to receive these push messages and your app can keep that count locally. We will have a "how to update the badge count in background" article being published soon. You could also use that to keep track of the number of missed messages per channel (chat room).
For now I just want to limit the number of rooms available for users to let's say a hundred and request and join them without pagination.
UPDATED DOCS: https://www.pubnub.com/docs/platform/channels/retrieve
We do have customers that do this without worrying about pagination. They just retrieve history on the 100 channels the device is subscribed to. With the background badge count updater strategy, you will have the advantage to know which channels to fetch from when the app becomes active. I will post the link to that article here once it is published.
I am looking to implement/add realtime push notification updates from node.js server to browser (client).
I looked into socket.io (http://socket.io/docs/rooms-and-namespaces/)
business requirement is - users will visit a page displaying customer info & their orders. There will be ~10,000 users visiting the page at any given time (all 10,000 could be for different customers or sometimes a user may have opened the same page in 2 or 3 tabs)
When orders flow in to elasticsearch for a customer (my datastore) i want to push notification to users who have that customer's page opened.
Questions:
Is socket.io the correct framework for this case?
Am I correct in understanding I have to use socket.io' rooms functionality to implement this? (each room identifier equals customer ID?)
Is this implementation scalable and would it be memory intensive for 10k users on node.js server?
Thanks!
Yes, but you could consider socksJS as well and write your own simple back-end.
Yes, it's the easiest way if you need authentication.
Worst case scenario you'd need to cluster your socket.io servers and use a back-end adapter. Redis should be fast enough for 10.000 connections.
EDIT: memory will depend on your specific implementation.
Also consider https://github.com/Automattic/socket.io/issues/1393