Make weave-kube work with AWSVPC mode

[awh]

From @bboreham on October 8, 2016 21:15

This may be as simple as adding the flag, but will need testing, and all the setup preamble needs documenting.

Copied from original issue: weaveworks-experiments/weave-kube#31

[awh]

From @chrislovecnm on October 9, 2016 20:28

@bboreham more details on this please

[awh]

From @bboreham on October 9, 2016 20:55

https://www.weave.works/documentation/net-latest-using-weave/net-latest-awsvpc/

[awh]

From @chrislovecnm on October 9, 2016 21:15

The number of hosts in a cluster is limited by the maximum size of your AWS route table. This is limited to 50 entries though you can request an increase to 100 by contacting Amazon.

You want me to open another ticket about how to get around that?? That is a big concern of ours.

[awh]

From @bboreham on October 9, 2016 22:40

We have spoken with Amazon; they do not have any plans to change it. I guess it's down to performance; larger routing tables take time to scan.

The idea of working around the limit is noted at #2322; I don't mind having a separate issue.

[awh]

From @chrislovecnm on October 9, 2016 22:46

Did not ask a good question ;) Can you run weave and get around the limit? Not run in aws vpc mode? Not sure if u have to use vpc mode.

[awh]

From @bboreham on October 9, 2016 22:51

Yes, you can run in the default overlay mode, where we don't set any limits on number of nodes. No further configuration is required; this is the default.

AWSVPC mode is for people who want no overlay.

[awh]

From @chrislovecnm on October 10, 2016 0:53

Perfect. Any idea how much of a performance hit you get with the overlay? I know native networking is always faster.

[awh]

From @bboreham on October 10, 2016 10:32

As always with performance, you should measure the real system you care about.

However we posted some fairly simple benchmark numbers at https://www.weave.works/weave-fast-datapath/, and more graphically at https://www.weave.works/container-networking-no-overlay-aws-vpc/

[chrislovecnm]

The big thing is the 50 VM limit and can we upgrade or migrate between those patterns on the love cluster.

I know this is a short statement.