As the question title suggests, I'm having an hard time to figure out what can be improved on my application(or tuned in the os, ubuntu) to achieve an acceptable performance. But first I'll explain the architecture:
The front-end server is an 8 core machine with 8 gigs RAM running Ubuntu 12.04. The application is written entirely in javascript and run in node.js v 0.8.22 (as some modules seem to complain on newer versions of node)
I use nginx 1.4 to proxy http traffic from port 80 and 443 to 8 node workers that are managed and started using the node cluster api. I use the latest version of socket.io 0.9.14 to handle the websocket connections, on which I've enabled only websockets and xhr-polling as available transports. On this machine I also run an instance of Redis(2.2)
I store persistent data(like users and scores) on a second server on mongodb(3.6) with 4gigs RAM and 2 cores.
The app is in production since few months(it's been running on a single box until few weeks ago) and it's being used by around 18k users per day. It has always worked very well apart of one main issue: performance degradation. With use, the amount of cpu used by each process grows until it staturates the worker(which won't serve requests anymore). I have temporarily solved it checking the cpu in use by each worker every minute, and restarting it if it reaches 98%. So the problem here is mainly cpu, and not RAM. The RAM is not an issue anymore since I've updated to socket.io 0.9.14(the earlier version was leaking memory) so I doubt it to be a memory leaking issue, especially because now it is the cpu that grows fairly quickly(I have to restart each worker around 10-12 times a day!). The RAM in use grows as well to be honest, but very slowly, 1 gig every 2-3 day of use, and the strange thing is that it is not released even when I completely restart the whole application. It is only released if I reboot the server! this I cannot really understand…
I've now discovered nodefly which is amazing, so I can finally see what's happening on my production server, and I'm collecting data since a couple of days. If anyone want to see the charts I can give you access, but basically I can see that I have between 80 and 200 concurrent connections! I was expecting node.js to handle thousands, not hundreds of requests. Also the average response time for http traffic floats between 500 and 1500 milliseconds which I think is really a lot. Also, in this very moment with 1300 users online, this is the output of "ss -s":
Total: 5013 (kernel 5533)
TCP: 8047 (estab 4788, closed 3097, orphaned 139, synrecv 0, timewait 3097/0), ports 0
Transport Total IP IPv6
* 5533 - -
RAW 0 0 0
UDP 0 0 0
TCP 4950 4948 2
INET 4950 4948 2
FRAG 0 0 0
which shows that I've got a lot of closed connections in timewait. I've increased the max open files to 999999, here is the output of ulimit -a:
core file size (blocks, -c) 0
data seg size (kbytes, -d) unlimited
scheduling priority (-e) 0
file size (blocks, -f) unlimited
pending signals (-i) 63724
max locked memory (kbytes, -l) 64
max memory size (kbytes, -m) unlimited
open files (-n) 999999
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) 819200
real-time priority (-r) 0
stack size (kbytes, -s) 8192
cpu time (seconds, -t) unlimited
max user processes (-u) 63724
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited
So I thought the problem could be on http traffic that for some reasons saturates the available ports/sockets(?), but one thing does not make sense to me: why when I restart the workers, and all the clients reconnect within few seconds, the load on the cpu of the worker goes down to 1% and is capable of serving requests properly until it saturates after about 1 hour (at peak time)?
I'm mainly a javascript programmer, not a sys admin so I don't know how much load I should expect to handle with my servers, but surely it's not performing as it should. The application is stable otherwise and this last problem is preventing me to ship the mobile versions of the app which are ready, as obviously they will bring more load and eventually crash the whole thing!
Hopefully there is something obvious that I'm doing wrong, and someone will help to spot it…feel free to ask me for more info, and I'm sorry for the length of the question but was necessary I believe…thanks in advance!
Best Answer
After few days of intense trial and errors, I'm glad to be able to say that I've understood where the bottleneck was, and I'll post it here so that other people can benefit from my findings.
The problem lies in the pub/sub connections that I was using with socket.io, and in particular in the RedisStore used by socket.io to handle inter-process communication of socket instances.
After realizing that I could implement easily my own version of pub/sub using redis, I decided to give it a try, and removed the redisStore from socket.io, leaving it with the default memory store(I don't need to broadcast to all connected clients but only between 2 different users connected possibly on different processes)
Initially I declared only 2 global redis connections x process for handling the pub/sub on every connected client, and the application was using less recources but I was still being affected by a constant CPU usage growth, so not much had changed. But then I decided to try to create 2 new connections to redis for each client to handle their pub/sub only on their sessions, then close the connections once the user disconnected. Then after one day of usage in production, the cpu's were still at 0-5%...bingo! no process restarts, no bugs, with the performance I was expecting to have. Now I can say that node.js rocks and am happy to have choosen it for building this app.
Fortunately redis has been designed to handle many concurrent connections(differently by mongo) and by default it's set at 10k, that leaves room for around 5k concurrent users, on a single redis instance, which is enough for the moment for me, but I've read that it can be pushed up to 64k concurrent connections, so this architecture should be solid enough I believe.
At this point I was thinking to implement some sort of connection pools to redis, to optimize it a little further, but am not sure if that won't cause again the pub/sub events to build up on the connections, unless each of them is destroyied and recreated each time, to clean them.
Anyway, thanks for your answers, and I'll be curious to know what you think, and if you have any other suggestion.
Cheers.