Can an app crash from a single TCP packet lost in transmission?

[Peter Much]

<cswiger at mac.com> aka Chuck Swiger  schrieb
mit Datum Fri, 17 Jul 2009 13:49:10 -0700 in m2n.fbsd.stable:

|On Jul 17, 2009, at 12:12 PM, Peter Much wrote:
|[ ... ]
|> One other thing did happen between 03:51 and 03:52 - the DSL
|> internet connection did disconnect/reconnect and obtained a new
|> IP adress. Afterwards, a script does flush and reload an ipfw table()
|> with the new local adresses - and during this process one(!) packet
|> of the database session was dropped.
|
|Well, there you go: having your IP change is certainly going to break  
|existing network connections; 

Uh, is that so? 
Maybe I wasnt clear enough: the failing database connection is a 
LOCALHOST-LOCALHOST connection - and it uses the (fixed) IP of one 
of the LAN interfaces, not the dynamic internet IP on the tun/PPP 
interface.

Changing the IP on one interface while using another interface is,
to all my knowledge, normal business.

And even IF there were problem with that, then I would expect the 
connection to timeout and fail, I would NOT expect a memory leak
to appear and drive the system out of swap within seconds.

!I don't believe there is anything which  
|is going to move the existing connection state in a NAT translation  
|layer or whatever over to the new IP. 

Nay, that doesnt work.

|Presumably you can obtain a  
|static IP and avoid such issues.

I have a static internet IP on the machine, also, for HTTPS. I have 
lots of interfaces there.

And I did run some tests in the meantime. The problem is in the 
PostgresQL library; it doesnt depend on a changed IP; - I only need
to steal one packet out of the transmission, then the database client
will grow to VSZ=1500GB and bigger. That's perfect reproduceable.

The postgresQL is 8.2, rather old by now - but I really wonder that
noone else did get into this during all the time.

rgds,
PMc