Re: [Idr] 2 week WG LC for draft-ietf-idr-shutdown-02 (1/17 to 1/31/2017)

[Job Snijders <[e-mail]>]

Dear Peter,

On Wed, Jan 18, 2017 at 10:54:19PM +0100, Peter van Dijk wrote:
> On 17 Jan 2017, at 22:04, Susan Hares wrote:
> > This begins a 2 week WG Last Call for draft-ietf-idr-shutdown-02.txt
> > (1/17/2017 to 1/31/2017).  Section 7 of this document lists the
> > implementations that have been created for this BGP functionality. These
> > include: ExaBGP, pmacct, OpenBGPD, Wireshark (packet analyzer), and tcpdump
> 
> I did the openbgpd implementation of draft-ietf-idr-shutdown. Please find my
> implementation report below.
> 
> A functional description of the implementation can be found at
> http://marc.info/?l=openbsd-tech&m=148390348126153&w=2 . Note that the final
> version of the patch is a few mails down in that thread (and phessler@ did
> some last whitespace fixes before committing it to OpenBSD CVS). In short, the
> openbgpd implementation is complete; it can send shutdown communications from
> the config file and from the command line; it logs received communications to
> syslog and exposes them in various forms of 'bgpctl .. show'.
> 
> During implementation, I have found the draft very clear to read, and there
> were no ambiguities in how an implementation was supposed to work. In every
> one of the few situations I had doubts arising from interoperability testing
> with other fresh implementations, it turned out there was a bug in the other
> implementation and each time this was quickly corrected on the other side
> after a closer reading of the draft or the involved code.
> 
> Implementation took a few evenings, spread out over weeks, showing that the
> cognitive load involved with understanding this draft is quite low.
> 
> Development was done on OpenBSD 6.0 (as openbgpd has not been ported to other
> platforms). The work was divided in logical pieces (log incoming messages;
> remember them for bgpctl show; send them from bgpctl neighbor X down REASON;
> send them from the configuration file on a reload+refresh).
> 
> Interoperability was continuously tested against the then-current
> implementations of -shutdown in exabgp and pmacct pmbgpd, running on a nearby
> Linux VM, as at that time pmacct only had support for sending a shutdown
> message, and exabgp was only able to receive shutdown messages. Various minor
> bugs in the exabgp and pmacct implementations surfaced and were fixed during
> this process.
> 
> Unprintable characters (both of the 'control' and the 'UTF-8' nature) were
> easily dealt with via the vis(3) (http://man.openbsd.org/vis.3) facility
> available on all BSDs, in octal mode as is also used in C string constants.
> Incidentally Debian has a 'libbsd' that provides it. A standalone
> implementation would not be hard to do either. vis-encoded strings can be
> decoded by 'unvis' on the commandline, or by simply pasting them into a Python
> interpreter. This way full Unicode support is achieved without having to worry
> about 'what encoding is syslog actually in'.
> 
> As an example, the Unicode unicorn (U+1F984), encoded on the wire in UTF-8 (as
> the draft specifies), yields a syslog message such as
> 
>     neighbor 165.254.255.26: received shutdown reason: "\360\237\246\204"
> 
> Running
> 
>     echo "\360\237\246\204" | unvis
> 
> on an UTF-8 capable terminal, or typing
> 
>     print("\360\237\246\204".decode('utf-8'))
> 
> into a Python2 prompt, will recover said unicorn. 🦄

Thank you for this write-up!

Kind regards,

Job