this.sockets.indexOf is O(n) ? is there a reason to not use a hash array instead ?

[kapouer]

Hi,
when trying to figure out why latest socket.io has CPU load climbing a bit too fast (70% on one core with around 5.000 clients), this was found:
https://github.com/Automattic/socket.io/blob/916872587fd6e0280e9538f712b91d0460bc3a2b/lib/namespace.js#L191
before i try to see if that can be optimized, i'd welcome some advice from the author(s), please.

[nkzawa]

I think there is no big reasons for this.

[kapouer]

What do you think @manubb ? Feel like doing it ?

[manubb]

I'll have a look at it.

[manubb]

Hi. I have patched socket.io and socket.io-adapter:
https://github.com/manubb/socket.io/tree/1.3.5-efficient
https://github.com/manubb/socket.io-adapter/tree/0.3.1-efficient

To my understanding, a namespace or a room can have many sockets (10000s) connected to but a Client only have a few sockets (a socket for each namespace it is connected to). Am i right?

Should i do a pull request for this?

[kapouer]

ping @nkzawa

[nkzawa]

To my understanding, a namespace or a room can have many sockets (10000s) connected to but a Client only have a few sockets (a socket for each namespace it is connected to). Am i right?

In most cases, this should be correct.

The modifications look almost good to me. Thanks for your effort 😄

[manubb]

I have done your suggested corrections. I did two pull requests, one for socket.io and one for socket.io-adapter. Version number of socket.io-adapter should be updated in both socket.io and socket.io-adapter.

[kapouer]

For reference:
#2123
socketio/socket.io-adapter#29

[nkzawa]

Should be fixed now, thanks.