[Libwebsockets] a bug?

"Andy Green (林安廸)" andy at warmcat.com
Sat Mar 16 05:47:40 CET 2013


On 11/03/13 20:42, the mail apparently from "Andy Green (林安廸)" included:
> On 11/03/13 11:49, the mail apparently from klervi - Alex Rhatushnyak
> included:
>> Hi Andy!
>> I tested libwebsockets-b2637a3978b1eae784b898ea4170002ab2614a97.tar.gz
>> <http://git.libwebsockets.org/cgi-bin/cgit/libwebsockets/snapshot/libwebsockets-b2637a3978b1eae784b898ea4170002ab2614a97.tar.gz>
>>
>> *
>> *
>> *test-client -u* doesn't cause segmentation fault in test-server
>>   (client is killed, as far as I can see, for trying to flood the server)
>
> Hm that isn't what's intended, stuff the server can't cope with should
> be "backpressured" into something the server can cope with by the rx
> flow control stuff not reading from the socket.  At least that's the idea.
>
>> But when test-client is run without '-u',
>> either seg.fault occurs after less than 50 seconds (more often less than
>> 20 seconds),
>> or the server stops receiving, it only sends data, while the client only
>> receives data.
>
> It sound like a real problem, unfortunately due to workload it will have
> to wait for the weekend before I can look properly at it.

The weekend rolled around and I spent the morning studying this... I 
think the improvement should help, especially

http://git.libwebsockets.org/cgi-bin/cgit/libwebsockets/commit/?id=b55451c6d20dd64c62594a91af3297a8934b0ec4

Please give it a try again and see if you can reproduce.

-Andy




More information about the Libwebsockets mailing list