More for Less: MikroTik Talk

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > More for Less: MikroTik Talk

Roseo: Jun 1, 2000; Forum Veteran

Weird Uncle Dave posted:

There are several different ways to do failover on Mikrotik, but none of them seem to handle the particular weird failure mode I'm trying to cover. Doing failover by just setting two default gateways, and using check-gateway is easy, and often "good enough." I want to handle the possibility that the failure is four or five hops upstream, though. (I work for an ISP and want to handle the rare possibility that all our upstreams are broken, so the end-user could still see everything within our network but not anything beyond that.)

I don't think I can just use a simple ping test to see if Upstream 1 is up, because let's say I ping something like 4.2.2.2. My script tests it, sees it can't ping that IP, switches to the secondary connection, pings, that IP suddenly is pingable again, switches back to the primary connection that's really still broken...

Meanwhile, pinging something like my network's default gateway would have the same problem in reverse if it really is a last-mile outage.

Any suggestions on getting out of this without a bunch of really complicated and fragile scripts?

Set up an ASN and use BGP on your external connections.

# ¿ May 20, 2011 02:40

Adbot: ADBOT LOVES YOU

# ¿ Apr 26, 2024 08:29

Roseo: Jun 1, 2000; Forum Veteran

Weird Uncle Dave posted:

My original question was for one of my residential customers who pays fifty bucks a month for my fixed-wireless service, and wanted to fail-over to a satellite connection. Sorry if that wasn't clear; we do run BGP in the NOC.

Edit: we actually used to use a PC with RouterOS for our BGP router, worked great, but the boss went all "CALEA compliance!" crazy and this was before they wrote their own CALEA package, so it got replaced with an Imagestream router. May replace the Imagestream with another RouterOS-based PC in the near future to keep up with how big the BGP table is growing...

Re: your failover question, I'm fairly certain you can specify the outbound interface for pings. You could probably set a dual homed router up with two default routes active simultaneously, and set the preferred connection to have a lower metric. When you start losing ping responses on an interface, you can flip the metric, and the router will then prefer the other route. This means you can do your testing on each interface, leave them up, but the actual internet bound traffic will just go out through the lower metric connection.

I'm not sure, however, how well the Mik scripting stuff would handle this.

As for swapping over to a Mikrotik BGP router, don't be in such a hurry. There's a memory leak when using BGP, to the point an 1100 with two sessions active starts dying and requires a reboot after ~6 weeks.

# ¿ May 27, 2011 03:43

Roseo: Jun 1, 2000; Forum Veteran

falz posted:

4.17 or 5.2?

4.x for certain. I'm not throwing 5 on anything till it's actually mature. It may be fixed, but I doubt it.

# ¿ Jun 4, 2011 18:28

Roseo: Jun 1, 2000; Forum Veteran

falz posted:

I enabled BGP on a few of them within the last month and haven't noticed any leaks (yet?). How many routes were you feeding it?

2x full tables on a RB1000. At ~2 weeks of uptime it's gone from 200 MB ram free to 70 MB free. After another week or two it'll sawtooth for a while, then a week or two after that randomly drop routes, not accept SSH sessions, and generally be crappy till a reboot.

Roseo fucked around with this message at 04:59 on Jun 7, 2011

# ¿ Jun 7, 2011 04:57

Roseo: Jun 1, 2000; Forum Veteran

wolrah posted:

5GHz has many more and doesn't make the mistake of allowing users to select "middle" channels which would overlap with two of the ideal non-overlapping channels.

Sure it does.

# ¿ Mar 14, 2014 01:21

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > More for Less: MikroTik Talk