Packet loss when using multiple subnet#weight entries

[Armin Schindler]

On 11/15/2015 02:28 PM, Armin Schindler wrote:
> Hello,
> 
> I have two servers (A and B) in separate locations. Both are connected
> together via two tinc switches to provide two subnets on both servers.
> This works pretty good. I can start my VMs on any server connected
> to one of those bridges without changing any routes.
> The subnets hosted on both servers (each in a bridge) are
> 172.16.10.0/24 (mainly on A) and 172.16.11.0/24 (mainly on B)
> 
> Now I want to connect to those servers from a third location (C) via
> tinc router. The idea was to connect to both and have a failover route
> if one connection fails.
> So I added on A:
>  Subnet = 172.16.10.0/24#10
>  Subnet = 172.16.11.0/24#20
> And on B:
>  Subnet = 172.16.11.0/24#10
>  Subnet = 172.16.10.0/24#20
> 
> Basically it seems to route all correct. But I got packet loss when
> both tinc routers on A and B are running.
> If I shutdown the tinc router on B, it works.
> 
> From C I try to ping B and got about 20% packet loss. A tcpdump shows
> that tinc packets are going to B and back. I can see the port 655 packet
> arriving at C, but the decrypted ping-reply does not appear.
> 
> What is happening to that packet? Is the tinc on C dropping it? Do I
> miss a rule or something.
> 
> Two servers run tinc 1.0.24, the other 1.0.26.

Please ignore this report about packet loss. After a lot of tests I found out
that the packets got lost somewhere between C and A. Strange that it happened
only with exactly that setup. Anyway, tinc is obviuosly not the cause.

But I have a question regarding "Subnet=" possibilities. When I have more than
one Server acting as host for VMs and need to have automatic routing to
all VMs on these servers, can I use tinc to create this routing automatically?

My idea is to have a script which is started when a VM is started/stopped
which will add/remove an Subnet=x.x.x.x/32 entry for this VMs IP address in
the local hosts file of that node and send a HUP signal to tincd.
Will tinc announce this new routing destination to all other nodes immediately?

Thank you!
Armin