Comware

Hello All,

We have 2xHPE switches 5945 connected in IRF and separate BGP peering are there with customer routers (router01 and router02) from each switch.

While we are doing power failover test on HPE switched by powering OFF IRF master switch01, BGP peering between IRF standby switch02 and customer router02 gets reset/flapped.  

So questions are:

1. When IRF master switch is powered down, will it impact whole IRF control plane or not  and for how long.

2. What is the switchover time interval for standby switch to become master.

Hi, what you experienced is what I would expect. In IRF there is a single control plane. In my network I have OSPF and this will 'flap' upon failover.

What I observe is that all routing starts from scratch as if the daemon only starts up upon becoming master. So the failover time will be the time BGP takes to go from off to established with all routes.

IRF is not like having two active nodes with their own IP addresses and neighbours etc. It is a single control plane.

Aruba CX works differently. Two switches with VSX pairing have their own L3 but can share specific aspects like LACP and anycast SVIs (sort of).

Hello All,

We have 2xHPE switches 5945 connected in IRF and separate BGP peering are there with customer routers (router01 and router02) from each switch.

While we are doing power failover test on HPE switched by powering OFF IRF master switch01, BGP peering between IRF standby switch02 and customer router02 gets reset/flapped.  

So questions are:

1. When IRF master switch is powered down, will it impact whole IRF control plane or not  and for how long.

2. What is the switchover time interval for standby switch to become master.

Thanks Ian for your feedback. 

Is there any solution in your mind to deal with this problem ?

Hi, what you experienced is what I would expect. In IRF there is a single control plane. In my network I have OSPF and this will 'flap' upon failover.

What I observe is that all routing starts from scratch as if the daemon only starts up upon becoming master. So the failover time will be the time BGP takes to go from off to established with all routes.

IRF is not like having two active nodes with their own IP addresses and neighbours etc. It is a single control plane.

Aruba CX works differently. Two switches with VSX pairing have their own L3 but can share specific aspects like LACP and anycast SVIs (sort of).

Hello All,

We have 2xHPE switches 5945 connected in IRF and separate BGP peering are there with customer routers (router01 and router02) from each switch.

While we are doing power failover test on HPE switched by powering OFF IRF master switch01, BGP peering between IRF standby switch02 and customer router02 gets reset/flapped.  

So questions are:

1. When IRF master switch is powered down, will it impact whole IRF control plane or not  and for how long.

2. What is the switchover time interval for standby switch to become master.

Hi, I don't think there is a simple answer to this. It is worth looking at your design as a whole. If the use of IRF is to get a failover of routers, consider two independent routers and use routing costs for the failover. Use IRF only for L2 failover and stacking.

If you need IRF for other reasons like LACP connections to downstream switches then I would consider tuning the routing to have the minimal downtime. The ability to do that depends on if you control the other end. In my network with OSPF, failover of routing with agressive timers is five seconds. 

Original Message:
Sent: May 31, 2023 05:42 AM
From: Vishank Goel
Subject: IRF failover cause BGP flap

Thanks Ian for your feedback. 

Is there any solution in your mind to deal with this problem ?

Original Message:
Sent: May 31, 2023 04:11 AM
From: IanNightingale
Subject: IRF failover cause BGP flap

Hi, what you experienced is what I would expect. In IRF there is a single control plane. In my network I have OSPF and this will 'flap' upon failover.

What I observe is that all routing starts from scratch as if the daemon only starts up upon becoming master. So the failover time will be the time BGP takes to go from off to established with all routes.

IRF is not like having two active nodes with their own IP addresses and neighbours etc. It is a single control plane.

Aruba CX works differently. Two switches with VSX pairing have their own L3 but can share specific aspects like LACP and anycast SVIs (sort of).

Original Message:
Sent: May 30, 2023 01:54 AM
From: Vishank
Subject: IRF failover cause BGP flap

Hello All,

We have 2xHPE switches 5945 connected in IRF and separate BGP peering are there with customer routers (router01 and router02) from each switch.

While we are doing power failover test on HPE switched by powering OFF IRF master switch01, BGP peering between IRF standby switch02 and customer router02 gets reset/flapped.  

So questions are:

1. When IRF master switch is powered down, will it impact whole IRF control plane or not  and for how long.

2. What is the switchover time interval for standby switch to become master.

This is expected behavior as there is no dynamic protocols state sync between IRF members. So when the Master reboots the Standby will re-establish BGP peering with its neighbors. What I can suggest you is to read "Improving BGP network reliability" section of the IP Routing Config Guide for the platform and assess the options available - https://www.arubanetworks.com/techdocs/Switches/FlexFabric/5944/5200-7955.pdf

EDIT: In your case I'd consider BGP NSR

------------------------------
Ivan Bondar

Original Message:
Sent: May 31, 2023 05:58 AM
From: IanNightingale
Subject: IRF failover cause BGP flap

Hi, I don't think there is a simple answer to this. It is worth looking at your design as a whole. If the use of IRF is to get a failover of routers, consider two independent routers and use routing costs for the failover. Use IRF only for L2 failover and stacking.

If you need IRF for other reasons like LACP connections to downstream switches then I would consider tuning the routing to have the minimal downtime. The ability to do that depends on if you control the other end. In my network with OSPF, failover of routing with agressive timers is five seconds. 

Original Message:
Sent: May 31, 2023 05:42 AM
From: Vishank Goel
Subject: IRF failover cause BGP flap

Thanks Ian for your feedback. 

Is there any solution in your mind to deal with this problem ?

Original Message:
Sent: May 31, 2023 04:11 AM
From: IanNightingale
Subject: IRF failover cause BGP flap

Hi, what you experienced is what I would expect. In IRF there is a single control plane. In my network I have OSPF and this will 'flap' upon failover.

What I observe is that all routing starts from scratch as if the daemon only starts up upon becoming master. So the failover time will be the time BGP takes to go from off to established with all routes.

IRF is not like having two active nodes with their own IP addresses and neighbours etc. It is a single control plane.

Aruba CX works differently. Two switches with VSX pairing have their own L3 but can share specific aspects like LACP and anycast SVIs (sort of).

Original Message:
Sent: May 30, 2023 01:54 AM
From: Vishank
Subject: IRF failover cause BGP flap

Hello All,

We have 2xHPE switches 5945 connected in IRF and separate BGP peering are there with customer routers (router01 and router02) from each switch.

While we are doing power failover test on HPE switched by powering OFF IRF master switch01, BGP peering between IRF standby switch02 and customer router02 gets reset/flapped.  

So questions are:

1. When IRF master switch is powered down, will it impact whole IRF control plane or not  and for how long.

2. What is the switchover time interval for standby switch to become master.