Doc suggestion - clarifying behaviour when routes are moving between protocols

[Maria Matejka]

Hello Mark,

you are having a lot of useful suggestions. There are even more thoughts 
alike yours, regarding documentation. And indeed you are right that our 
documentation is kinda good as a reference manual but it kinda sucks if 
you are a newbie.

The thing is, we have quite a long backlog and some technological debt 
to resolve (see e.g. multithreading or our still-non-existent API), and 
all of these are very time-consuming. Writing good documentation, or 
better to say, creating good paedagogical materials for self-learning, 
is time-consuming as well. And even though we are very much in favor of 
having better documentation, the aforementioned problems have somehow 
higher priority than documentation (money being the main reason).

Anyway, I'll later try to split your suggestions into independent pieces 
and create our internal issues for each of them, to not lose them.

Now to the suggestions (inline):

On 2024-02-09 06:13, Mark Shuttleworth wrote:
>
> Admittedly I am not a networking specialist, but perhaps that makes my 
> feedback useful in terms of gaining new users and bringing Bird to a 
> wider audience. My previous experience has been playing with Babel for 
> this same personal project, and Bird is obviously a step up in 
> complexity given its support for many different protocols. As an 
> aside, it's great to see Babel support in Bird :)
This is actually a very good position to read the docs and see what 
makes you confused. I'm very happy that you didn't freak out.
>
> At the time, here's what I struggled with. Hopefully some of this will 
> be helpful in shaping docs for new users.
>
> It would be useful early in the doc to have a visualisation of the 
> relationship between a table such as master4 and a protocol, like 
> kernel or ospf. I found drawing a picture for myself helped to make 
> import vs export clear. Perhaps the docs can have a picture like that! 
> My suggestion would be to make that visualisation include multiple 
> protocols such as the kernel protocol and one of the routing protocols 
> like RIP or OSPF, so that it can help explain how a route might move 
> between routers (OSPF) and into the local machine routing table. I 
> would walk through a full example, showing movement of routes in both 
> directions from both protocols. Describe, step by step, how a kernel 
> route would be learned and imported into a table like ipv4, and then 
> exported by OSPF, and vice versa. Perhaps this is obvious to a routing 
> expert but it wasn't obvious to me as someone playing with a proper 
> routing daemon for the first time.
We indeed don't have many pictures in the documentation and we shall add 
some. Anyway, this reminds me of … you are a visual learner, more than a 
textual one, aren't you? Would you appreciate a video tutorial, 
explaining these things step by step?
>
> The multi-protocol case (kernel protocol and ospf protocol) may be 
> less interesting to someone who's focused on BGP or OSPF across many 
> routers, but it seems to me to be critical to the new user experience. 
> A new user is likely to have one, two or three routers, and they need 
> to move routes between them AND get those routes installed into the 
> kernel, so they have to understand the multi-protocol interaction 
> dynamics. This is where things like preference, which are not things 
> one can learn about just by reading about OSPF or RIP, come into play.
>
> The part of the doc you refer to is critical - it covers the main deal 
> which is 'which routes from the various protocols will move into 
> position where they can be actioned', but it's also very terse, and 
> it's just a paragraph or two in a long stretch of detail. It also uses 
> the term 'preference' twice, and I'm not sure those two uses refer to 
> the same thing. It says 'Preferences of the routes are compared' and 
> 'Source protocol instance preferences are compared' but neither 'route 
> preference' nor 'source protocol instance preference' has been defined 
> before this text. I'm still not sure how I would go and look at a 
> setup with birdc and find out easily what the route preference or 
> source protocol instance preference would be.
>
> So I think all of this warrants a section on its own, early in the 
> overall documentation, that walks through the behaviours of protocols, 
> channels, and tables, and how they interact. It's notable that other 
> people have found a need to try and explain this more clearly 
> themselves - digging around to compose this email I found another 
> example at https://blog.kintone.io/entry/bird which may be helpful to you.
This probably needs much more rewriting, considering that there are also 
MPLS rules now, EVPN upcoming … good point from you.
>
> It would also help if logging of decisions wrt import or export 
> captured more of the rationale for the decision. As I dug into this I 
> saw messages like 'route rejected by protocol', but no rationale for 
> that rejection (it turned out to be the preference).  Because the 
> relationship between tables and protocols, import and export is 
> unclear initially, it would be helpful to be explicit in these log 
> messages. Would it be possible to say something like this:
> /<TRACE> 10.2.21.0/24 <http://10.2.21.0/24> from kernel import to 
> master4 accepted //<TRACE> //10.2.34.0/24 <http://10.2.34.0/24> from //master4 export to ospf1 rejected because protocol preference 100<30 /
> In this example I'm replacing angle brackets with explicit import / 
> export, and listing the table names explicitly, to try and make it 
> easier for a new user to follow that log and not be confused. I'm 
> trying to show direction in the order - from kernel import to master4, 
> and from master4 export to ospf1. The reason for the rejection would 
> help to then dig in to details like protocol preferences.
People actually quite often parse the logs so we don't have so much 
space to change the format, but you're right that the log format is very 
dense and sometimes misleading. Regarding the "because something" part … 
it's quite challenging to actually achieve this but I'll make an issue 
for it and we'll see later what can be done with that. Especially with 
the BIRD 3 multithreaded architecture, there may be some funny corner 
cases and race conditions which have to be checked for.
>
> In debugging with birdc, I found the layouts rather inscrutable and 
> could not find documentation. For example:
>
> bird> show route
> Table master4:
> 0.0.0.0/0 <http://0.0.0.0/0>            unicast [ospf1 2024-01-31] * 
> E2 (150/40/900) [192.168.12.2]
> via 192.168.4.248 on bond-lan weight 1
> via 192.168.4.249 on bond-lan weight 1
> 192.168.254.48/30 <http://192.168.254.48/30>    unicast [ospf1 
> 2024-01-31] * I (150/40) [192.168.254.57]
> via 192.168.4.248 on bond-lan weight 1
> via 192.168.4.249 on bond-lan weight 1
>
> What is each of those fields? It would be nice if the docs for the 
> protocol also explained how birdc represents routes from that 
> protocol. In this case, the docs at 
> https://bird.network.cz/?get_doc&v=20&f=bird-6.html#ss6.11 
> <https://bird.network.cz/?get_doc&v=20&f=bird-6.html#ss6.11> could 
> show the various permutations of an OSPF route, and point out where 
> preference and metrics are (it took a lot of googling to understand 
> "E2 (150/40/900)" :)
>
> Similarly, while preference is a critical element of the table import 
> / export route selection algorithm, it's hard to know what it actually 
> is. For example, where is the preference of this protocol instance?
>
> bird> show ospf
> ospf1:
> RFC1583 compatibility: disabled
> Stub router: No
> RT scheduler tick: 1
> Number of areas: 1
> Number of LSAs in DB: 32
> Area: 0.0.0.0 (0) [BACKBONE]
> Stub: No
> NSSA: No
> Transit: No
> Number of interfaces: 3
> Number of neighbors: 4
> Number of adjacent neighbors: 4

Documenting these is one solution (but you have to look it up), creating 
a web-browser-friendly interface is another one. (Both can be done.) And 
with our new shiny API (which is still mainly on the drawing board), the 
web-browser interface will be possible to create quite smoothly. Also, 
there are so-called looking-glasses, which are basically web interfaces 
over BIRD with added features. See e.g. Alice-LG: 
https://github.com/alice-lg/alice-lg

With that, I'm quite convinced that we're gonna have some web-browser 
interface with all of the routes and parameters visualised, well styled 
and explained with context help. (Funding of such an effort, or even 
just getting a commercial support package to actually steer our 
development this way, would be very much appreciated, though.)

>
> Apologies if this is newbie stuff, and hopefully it helps you get your 
> future newbies through that learning process faster with fewer 
> questions to your list :)

You did the best thing you could, and thank you for all your thoughts. 
If you find any other suggestion, feel free to write it.

Have a nice day!
Maria

-- 
Maria Matejka (she/her) | BIRD Team Leader | CZ.NIC, z.s.p.o.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://trubka.network.cz/pipermail/bird-users/attachments/20240209/3b8353dd/attachment.htm>