[paramiko] intermitent bug in channel

[Robey Pointer]

Sorry for the late reply...  I got caught up in work & life and left  
my email to pile up for a few weeks.


On 15 Sep 2006, at 12:22, David Guerizec wrote:

> Hello Robey,
>
> On Friday 15 September 2006 19:09, Robey Pointer wrote:
>> I can't reproduce it at all on my macbook.  On my desktop Linux box
>> at home, I was able to see the bug a few times, but very rarely --
>> too rarely for me to get many clues.
>>
>> I know it's not helpful, but the times I saw the short response, it
>> had the same symptoms as you describe: the server thought the client
>> disconnected prematurely.
>
> It is not exactly happenning the same on each boxes, unfortunatly.  
> Sometimes,
> the connection is valid to the end from the point of view of the  
> client.
>
>> I was able to see that when it gets a short response, the missing
>> lines are from the end -- it's truncated.
>
> IIRC, I've seen some occurence of missing data in the middle of the
> connection, when transfering several dozens of MB.
>
>> Just now, I was able to get it to happen while the server was logging
>> packet sizes.  It looks like the server sent all the packets, because
>> for the truncated response, the byte count is still correct.  I wish
>> I could see what was happening on the client side.
>
> Yes, I've seen the same behavior.
>
> I've spent many hours today trying my test program on any box I've  
> found, and
> I finally came to a surprising conclusion. Read on.
>
> For the test, I've made a Makefile[0] to compile python2.5 (I've  
> also tried
> with 2.4 with the same result), pycrypto-2.0.1 and paramiko-1.6.2,  
> to run the
> buggy_server script. This way I could easily test it on any UNIX.
>
> First, it's not happening on OpenBSD (I couldn't find any OSX). All  
> linux
> boxes (debian and gentoo) where affected but one. So I diffed the  
> good one
> with a similar server, and after upgrading them to the exact same  
> gcc and
> glibc, the only sensible difference was the kernel.
>
> I've upgraded the kernel from 2.6.8 to 2.6.16 on the good box, and  
> the bug
> appeared. I downgraded the kernel on a bad box, and the bug  
> disappeared.
>
> So far, I've narrowed the kernel good and bad versions to resp.  
> 2.6.13 and
> 2.6.15. I'm still in the narrowing-down process.
>
> I hope your linux kernel version is more recent than 2.6.13, else  
> my theory
> falls to pieces.

My kernel is 2.6.17.11 -- I dunno if that helps. :)

robey