Jay,

 

Configured distance is already considered by Zebra/RTM and that design stays as-is. Weight/Local preference alone are not enough to address this issue. Deterministic in the sense, given a set of events, FRR stack will download same route to RIB/FIB irrespective of sequence in which the events happened.

 

Some of the details of the discussion were captured in the bug comments. Please check for more details.

 

Thanks,

Mohan

 

 

From: Jia Chen <jchen1@paloaltonetworks.com>
Date: Monday, September 10, 2018 at 10:08 AM
To: Mohan Singamsetty <msingamsetty@vmware.com>, Donald Sharp <sharpd@cumulusnetworks.com>, "ntriantafillis@gmail.com" <ntriantafillis@gmail.com>, FRRouting-Dev <dev@lists.frrouting.org>, Don Slice <dslice@cumulusnetworks.com>, Russ White <russ@riw.us>
Subject: Re: [dev] [FRRouting/frr] Static blackhole route does not honor distance attribute (#2230)

 

Just a question,

 

Should configured distance be considered by Zebra/RTM stage rather than inside of BGP best path selection. If BGP best path selection need to be influenced, why using weight, local preference etc not enough?

 

In another note, the deterministic behavior, are we talking about ordered/prioritized FIB downloading depending on route types?

 

Thanks,

Jay

 

 

 

Configured distance

 

From: dev <dev-bounces@lists.frrouting.org> on behalf of Mohan Singamsetty <msingamsetty@vmware.com>
Date: Monday, September 10, 2018 at 9:40 AM
To: Donald Sharp <sharpd@cumulusnetworks.com>, "ntriantafillis@gmail.com" <ntriantafillis@gmail.com>, FRRouting-Dev <dev@lists.frrouting.org>, Don Slice <dslice@cumulusnetworks.com>, Russ White <russ@riw.us>
Subject: Re: [dev] [FRRouting/frr] Static blackhole route does not honor distance attribute (#2230)

 

We had this meeting to discuss issue #2230 and conclude on the fix.

 

Date: 09-05-2018

Attendees: Donald, Nikos, Don, Mohan, Soman.

Minutes:

  1. Two issues were discussed.
    1. FRR stack not being deterministic in downloading routes to RIB/FIB and the routes downloaded to RIB/FIB depends on sequence of commands executed or sequence of network events happened. Ideally the routes in RIB/FIB should be deterministic and consistent at a given time after executing set of same commands.
    2. FRR stack not honoring the admin distance configured for routes and downloading routes with higher admin distance (higher in numerical value). Ideally routes with lower admin distances should be considered as best routes and should get downloaded to RIB/FIB.
  2. Two Proposals were considered.
    1. Consider admin distance configured for each route in BGP best bath algorithm to make FRR stack behave more deterministic and consistent with respect to downloading routes to RIB/FIB.
    2. Use route maps to control the routes to be downloaded to RIB/FIB depending on the requirement and use case.

 

Conclusion:

  1. It was concluded that the FRR stack behavior should be deterministic even without using any route maps (without any additional configuration from user).
  2. Team agreed to use proposal #1 (Using the admin distance of route/protocol inside the BGP best path algorithm to determine the best route) to move forward and address the above issues mentioned.
  3. Admin distance will be the first thing to compare in FRR best bath path algorithm. This is in addition to the other 13 checks that FRR stack currently has.
  4. There will be a #def to control the default behavior from build time. i.e. either to have the old behavior where admin distance of each route is not used inside the BGP best path algorithm or to have new behavior where admin distance is used inside BGP best path algorithm to fix the above issue. User can use this #def to build the FRR stack and get the default behavior that is needed.
  5. There will be a CLI command also to control the default behavior as explained above.

 

We will publish the design and start implementing this soon.

 

Guys, please add more if I missed to add anything here.

 

Thanks,

Mohan

 

From: Mohan Singamsetty <msingamsetty@vmware.com>
Date: Tuesday, September 4, 2018 at 10:15 AM
To: Donald Sharp <sharpd@cumulusnetworks.com>, "ntriantafillis@gmail.com" <ntriantafillis@gmail.com>, FRRouting-Dev <dev@lists.frrouting.org>
Subject: Re: [FRRouting/frr] Static blackhole route does not honor distance attribute (#2230)

 

I am setting up a meeting to discuss this further on Wednesday (5th September) at 8 AM PST.

 

Link for the bug and comments

https://github.com/FRRouting/frr/issues/2230

 

Donald, Nikos, Russ, Please let me know if this time works for you.

 

If anyone else is interested to join, please let me know and I will add you to the meeting.

 

Thanks,

Mohan

 

 

From: David Lamparter <notifications@github.com>
Reply-To: FRRouting/frr <reply+009b114fa2e0109bec00750471786212a516d3548fe6032192cf0000000117a665b892a169ce133f1a52@reply.github.com>
Date: Tuesday, September 4, 2018 at 8:24 AM
To: FRRouting/frr <frr@noreply.github.com>
Cc: Mohan Singamsetty <msingamsetty@vmware.com>, Comment <comment@noreply.github.com>
Subject: Re: [FRRouting/frr] Static blackhole route does not honor distance attribute (#2230)

 

=> meeting this week
=> rough consensus on call is that deterministic behaviour is desired

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or mute the thread.mage removed by sender.