[Pvfs2-developers] Re: the halloween bug fixed

[Rob Ross]

Sam Lang wrote:
> 
> On Oct 9, 2007, at 8:02 AM, Scott Atchley wrote:
> 
>> On Oct 8, 2007, at 3:56 PM, Sam Lang wrote:
>>
>>> The patch implements a complementary callback function: 
>>> bmi_method_addr_forget_callback, to be called by the individual 
>>> methods when the connection is closed or the address is removed from 
>>> the method's own management structures.  In the case of tcp, a 
>>> connection is closed when poll or epoll returns an IN event on a 
>>> socket (through testcontext) and the following read returns zero 
>>> (EOF).  Its at this point the method now (with my patch) calls 
>>> bmi_method_addr_forget_callback to remove the address reference from 
>>> the list.
>>
>> I think it is stretching a bit for the server's BMI method to make an 
>> assumption about the fate of a client due to a loss of communication. 
>> The loss could be permanent or a re-connect attempt may already be 
>> initiated.
> 
> Its not a loss of communication.  When the client closes its socket, a 0 
> byte message is sent to the server.  Reading that zero byte message is 
> how the server knows the client is requesting the connection gets closed.

Same thing happens when a client dies. That's just the way sockets let 
an app know that a socket has been closed. You can't tell the difference 
between an app dying and an app closing a socket, unless I'm remembering 
incorrectly.

Rob