[mythtv-users] mythfilldatabase getting bad start date message from SD

f-myth-users at media.mit.edu f-myth-users at media.mit.edu
Mon Jan 11 19:47:31 UTC 2010


    > Date: Thu, 07 Jan 2010 13:14:12 -0500
    > From: Tony Lill <ajlill at ajlc.waterloo.on.ca>

    > The interesting thing is that the tcp packet is split into two IP
    > datagrams, the second's payload starting right at 8T05:00:00Z 

I haven't seen anything more on this thread, and don't know if you're
still having problems, but the fact that you're seeing this is a
smoking gun for a bad load balancer at the SD end or some similar
shenanigan---I've seen cheaply-made proxies/loadbalancers/whatever
that assume that an entire GET is going to fit into the first
datagram, even though TCP explicitly makes no such assurance since
the MTU could well be tiny or the GET could be huge.

(Of course, the -correct- behavior is to actually read until the end
of the request, no matter how many datagrams and/or fragments that is,
do reassembly, and -then- decide what to do with it, but that requires
a possibly-large amount of buffer memory and the smarts to do the
requisite reassembly of packets or fragments thereof, and many vendors
just punt and assuem they'll never get in trouble for doing so---they
just assume that -of course- MTU is always Ethernet's 1500 max and -of
course- all GETs fit in that size, etc.)

If I'm right, one way to test this would be to artificially decrease
the MTU at the requesting side to something tiny (100 bytes? 50?) and
then see if the SD end breaks.


More information about the mythtv-users mailing list