Our problem: we have more than 300 000 online clients to application (50 000 to one node). Because we use load balancer we don't now where (which nginx instance) alive connection to client. So we need pub message to all nodes.
How we can get linear scalability?
Current functionality:
Comet(for us comet is required) clients (mobile/web browsers) and server (java) with pub-sub functionality.
Client can send notifications to other (single) client;
Server can send notifications to single client or group of clents;
Clients recieve information about active clients (by group) from nginx+memcached
Now we use: nginx + https://github.com/wandenberg/nginx-push-stream-module + memcached.
Client before set up on comet channel send request to server: "I am alive and my commet channel id ...". Each 10 min client reopen pipeline, but register as alive on server only once (it's other problem).
I think we can write some logic with Lua on nginx(+redis or memcached). And each client request for reopen pipeline update information in redis (+add additional information about nginx instance ip which bind connection).
But may be exist other variants or practice for scalability comet functionality?
Related
When speaking from a conceptual point of view, is it standard practice to mix WebSockets and HTTP requests when making a chat application (or any application that requires real-time communication between devices)?
Imagine a scenario with a client and a server in a chat app. What would be the best approach for connecting and sending data between the client and the server? Would it be using sockets for both sending and receiving or HTTP requests for sending (so the client would get a response and then know if the message was received), and then using WebSocket for only receiving new messages?
No this is not standard practice.
If you need real-time communication between client and server you normally just use a websocket connection and keep that one open. The client can send messages to the server and receive messages through the same connection.
Using HTTP requests for sending messages to the server and receiving new messages via websocket seems odd and just adds unnecessary complexity.
Now if your server has some endpoints to subscribe for real-time data e.g. a chat room and endpoints for getting information you don't necessary want to subscribe to e.g. information about a certain user, than you can use the appropriate protocol for each endpoint
https://developer.mozilla.org/en-US/docs/Web/API/EventSource
The EventSource interface is web content's interface to server-sent events. An EventSource instance opens a persistent connection to an HTTP server, which sends events in text/event-stream format. The connection remains open until closed by calling EventSource.close().
From what I understand server-sent events require persistent HTTP connection (Connection: keep-alive) so similarly to keeping the connection alive like in case of web sockets.
If the connection is persistent, why server-sent events are unidirectional? Web socket connections are persistent as well.
In this case, what happens if I send a request to my HTTP service and I have persistent connection opened due to EventSource. Will it re-use HTTP connection opened by EventSource or open a new connection?
If it re-uses the connection opened by EventSource how is it considered unidirectional?
Might be trivial, but I had to ask because it is not clear. Because nothing mentions what happens to subsequent HTTP requests when there's existing connection opened by EventSource.
For example, it seems possible to me to implement centralized chat app using SSE:
User 1 sends message to User 2(by sending it to HTTP server). Server sends event to user 2 with a new message, user 2 sends another request to HTTP server with message for User 1, server sends event to user 1.
How is that not considered bi-directional?
Related:
What's the behavioral difference between HTTP Stay-Alive and Websockets?
SSE is unidirectional because when you open a SSE connection, only the server can send data to the client (browser, etc.). The client cannot send any data. SSE is a bit older than WebSockets, hence may be the difference between the unidirectional and bi-directional support between these two technos.
In your use-case, if you open a SSE connection (which is an HTTP connection), only the server will be able to send data. If you wish to send a request to your HTTP service, you will need to open a new "classical" HTTP connection. You will see your browser opening two HTTP connections: 1 for the SSE connection and 1 for the classical HTTP request (short live).
You can implement a chat with SSE. You can have a SSE connection (hence HTTP) to let the user receives the messages from the server. And you can use POST HTTP requests to enable the user to send his/her messages.
Note that most of the browsers can open around 6 HTTP/1.x connections to the same host. So, if you use 1 SSE connection, it will remain potentially 5 HTTP/1.x connections. This is only true with HTTP/1.x. With HTTP 2.x, the connections to the same host are multiplexed: so, in theory, you can send as many HTTP requests at the same time as you wish or you can open as many SSE connections as you wish and thus, by passing the limitation of the 6 connections.
You can have a look at this article (https://streamdata.io/blog/push-sse-vs-websockets/) and this video (https://www.youtube.com/watch?v=NDDp7BiSad4) to get an insight about this technology and whether it could fit your needs. They summarize pros & cons of both SSE and WebSockets.
My understanding regarding network model communication:
Application layer:
1. HTTP(Not Persistent or stateless): For exchanging messages like get, post, put etc. Here connection is made to webserver and disconnected after sending response. So server will not keep track of the previous requests.
2. Websockets(Persistent or statefull): For creating a communication channel that will be open to exchange data. Here we can keep track of the previous requests. Like we can know how many users are currently connected to our server.
Transport layer:
TCP(Persistant and Statefull): Will let the server know to which application to connect using port number. Both HTTP and web sockets will work upon this layer.
Considering working with HTTP and TCP:
I make a HTTP request from browser(application layer):
Connects to web server sends all files requested and also makes a TCP connection with the application(transport layer).
After sending response it's disconnected.
My confusion:
I got totally confused when I heard, read that TCP is Statefull and Persistant connection.
Q1. Now after step three is browser still connected to webserver because of TCP?
Q2. Is the context object we get in server side in c# code is the complete request packet with HTTP info, TCP info, function to invoke or controller to invoke in MVC etc?
Q3. If client and server are still connected with TCP. Then on next HTTP request does it will use the available TCP connection or will create new TCP and HTTP connection? Why can't it use previous TCP to communicate? Or TCP will be destroyed after HTTP? Or what's happening?
I am new to SignalR and I have a question on SignalR communication when we introduce a load balancer.
Lets assume we want to execute a void method on server side which receives some data as a parameter from client. Server takes that data and processes further. Lets say after processing for a while, it identifies that it has to send the notification back to client.
Case 1(Between client and server): Client calls void method on server side(Hub) by passing some data. Connection gets disconnected. Server processes the client data further. When it identifies that it has to push the notification back to client, it recreates the connection and pushes back the data to client.
Case 2(Between client and server with load balancer in between): How does the above scenario(Case 1) work here?. When server sends the push notification back to load balancer after processing client data, how does it know to which client it has to send the notification back?
You should read the scaleout docs. Short version: messages get sent to all servers, so if the client reconnects (it's not the server that establishes the connection!) before the connection times out , it will get the message.
Quote from the docs:
The cursor mechanism works even if a client is routed to a different
server on reconnect. The backplane is aware of all the servers, and it
doesn’t matter which server a client connects to.
At the moment I have an existing application which basically consists of a desktop GUI and a TCP server. The client connects to the server, and the server notifies the client if something interesting happens.
Now I'm supposed to replace the desktop GUI by a web GUI, and I'm wondering if I have to rewrite the server to send http packets instead of tcp packets or if I can somehow use some sort of proxy to grab the tcp packets and forward them to the web client?
Do I need some sort of comet server?
If you can make your client ask something like "Whats new pal?" to your server from time to time you can start implementing HTTP server emulator over TCP - its fun and easy process. And you could have any web based GUI.
You can just add to your TCP responds Http headers - itll probably do=)
So I mean HTTP is just a TCP with some headers like shown in here.
You should probably install fiddler and monitor some http requests/ responses you normally do on the web and you'll get how to turn your TCP server into http emulator=)
If you want keep sockets based approche use flash (there is some socket api) or silverlight (there is socket API and you can go for NetTcpBinding or Duplexbinding something like that - it would provide you with ability to receive messages from server when server wants you to receive them (server pushes messages))
So probably you should tall us which back end you plan to use so we could recomend to you something more usefull.