[freenet-dev] Minor Update Deadlock

[Robert Hailey]

On Jul 18, 2007, at 6:47 AM, Matthew Toseland wrote:

> On Tuesday 17 July 2007 20:40, Robert Hailey wrote:
>> It would appear that if two nodes are simultaneously updated across a
>> 'good' version, the 'link' between the two will forever be dead; each
>> waiting for the other to get up to speed on the version, and each not
>> wanting to connect to the other because it is too old, therefore
>> never contacting the presumed-old node to realize that it is, in
>> fact, compatible.
>
> Not sure I understand.
>
> We do send handshake attempts even if a node is TOO OLD, we just  
> send them
> less frequently (maybe this is a bad thing). And in fact if we  
> establish a
> connection we can UOM from them.

I'm not sure that the specific problem here is what I said, only that  
is/was my 'best guess'.

I have a few computers running freenet disconnected from the network  
at-large, but connected to each other. All was well and good with  
this test setup, but upon updating the node versions (a big jump,  
from about 1007->1045), they now will not connect to each other. I  
also tried deleting the peers and re-adding the references, but  
without success (they simply went from perpetually 'disconnected' to  
'never connected').

At present the peer status shows "NEVER CONNECTED*" (* = fetching  
ark?), a valid connection address (even port number matching),  
backoff "0/1" and zero entries in the more-detail/message-type table.  
Conventional network communications between the machines are fine.

--
Robert Hailey