Failover time
83 views
Skip to first unread message
Rico Gleissner
Jul 21, 2021, 5:13:34 AM7/21/21
to metallb-users
Hi,
we use MetalLB (Layer 2, CentOS 8, Calico network plugin ) in a k8s bare metal cluster. We did several failover tests and we measured the time. The default 5 minutes pass before the swivel is carried out in the k8s cluster. With a few changes to the k8s cluster, the time can be reduced to 40-60 seconds but this is not ok for use. Is there a option to reduce the failove time ?
Etienne Champetier
Jul 21, 2021, 9:13:25 AM7/21/21
to Rico Gleissner, metallb-users
Hi Rico,
You need to run more multiple pods, here the time you measure is the
time to move the pods, not for MetalLB to switch (~2s)
Le mer. 21 juil. 2021 à 05:13, Rico Gleissner <neoku...@gmail.com> a écrit :
>
> Hi,
>
> we use MetalLB (Layer 2, CentOS 8, Calico network plugin ) in a k8s bare metal cluster. We did several failover tests and we measured the time. The default 5 minutes pass before the swivel is carried out in the k8s cluster. With a few changes to the k8s cluster, the time can be reduced to 40-60 seconds but this is not ok for use. Is there a option to reduce the failove time ?
>
> --
> You received this message because you are subscribed to the Google Groups "metallb-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to metallb-user...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/metallb-users/797abd3f-5a91-4bfc-831e-ef437ad145cdn%40googlegroups.com.
You need to run more multiple pods, here the time you measure is the
time to move the pods, not for MetalLB to switch (~2s)
Le mer. 21 juil. 2021 à 05:13, Rico Gleissner <neoku...@gmail.com> a écrit :
>
> Hi,
>
> we use MetalLB (Layer 2, CentOS 8, Calico network plugin ) in a k8s bare metal cluster. We did several failover tests and we measured the time. The default 5 minutes pass before the swivel is carried out in the k8s cluster. With a few changes to the k8s cluster, the time can be reduced to 40-60 seconds but this is not ok for use. Is there a option to reduce the failove time ?
>
> You received this message because you are subscribed to the Google Groups "metallb-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to metallb-user...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/metallb-users/797abd3f-5a91-4bfc-831e-ef437ad145cdn%40googlegroups.com.
Rico Gleissner
Jul 22, 2021, 8:23:50 AM7/22/21
to metallb-users
Hi,
our setup is as followed:
- 3 k8s nodes as master and worker
- 3 MetalLB speaker
- 1 MetalLB controller
- Calico (BGP mode)
- CentOS 8
- All nodes in the same subnet
We did a every second a curl on the MetalLB VIP but failovertime is not as aspected. Approx 5 min.
As you described....We also tried 3 replicas of MetalLB controler and MetalLB speaker but the failovertime is not 2 seconds or even a minute. We thought it might be a similar problem like this: https://github.com/metallb/metallb/issues/298
But we checkt that the memberlist is active:
Etienne Champetier
Jul 22, 2021, 11:50:35 AM7/22/21
to Rico Gleissner, metallb-users
How many pods are you running for the service behind the VIP ?
To view this discussion on the web visit https://groups.google.com/d/msgid/metallb-users/835d462f-bb20-456a-bd07-09e33ed385d3n%40googlegroups.com.
Rico Gleissner
Jul 23, 2021, 3:35:34 AM7/23/21
to metallb-users
Hi,
also 3. I have picked out some information:
Etienne Champetier
Jul 23, 2021, 10:46:53 AM7/23/21
to Rico Gleissner, metallb-users
I only see 1 ingress-nginx
To view this discussion on the web visit https://groups.google.com/d/msgid/metallb-users/713f6e2b-6d23-400f-9938-a3fcafde6c73n%40googlegroups.com.
Rico Gleissner
Jul 26, 2021, 2:51:11 AM7/26/21
to metallb-users
Hi Etienne,
thank you very much.... This was the problem. Only one ingress.After the increase everything went as desired
Reply all
Reply to author
Forward
0 new messages