Ubiquiti discovery and STP mess

[aurelien personnaz]

Hello,

I want to report an issue we've had since a few weeks with the ubiquiti discovery and our switches running STP.
It is very probably related to your message v1.4.0 - CRITICAL BUG ALERT from the 28th of may.

We are in the process of splitting a big bridged network into small routed subnets, but the bridged part is still pretty big.

We have some scripts running once a day to discover the topology of the network and detect the new/dropped nodes. This script makes an extensive use of both CDP and the ubiquiti discovery to explore the network.

Since a few weeks, we have noticed that the script was very harmful for the network, generating a huge amount of packet losses. We found out that these packet losses were due to some wrong STP behavior on the switches when some ubiquiti discoveries were running (unwanted flipping between links).

The workaround we have found is to disable the ubiquiti discovery on all the switches running STP on the switched network. The switches are now only discovered by CDP and ignore the ubnt discovery packets. And they don't get crazy anymore.

All the switches are on 1.4.2rc6.
So there seems to be two possible reasons. Either the problem you talked about in your message has not been completely fixed, or the switch is simply dropping some STP packets when submitted to some ubnt discovery traffic and requests.

[sirhc]

So you are saying even with v1.4.2rc6 you still have a problem when you run your scripts asking all UBNT devices to advertise?

[aurelien personnaz]

Yes we do, when the switches have the ubiquiti discovery enabled.

[sirhc]

OK, I pinged Eric on this.

How many UBNT Discovery devices do you think you have on the segment?

How often are you running a script to have UBNT device re-advertise?

[aurelien personnaz]

We have a little more than 80 ubiquiti devices on that network and 15 netonix switches.

The discover command is supposed to run only once on this subnet for each of our script run, but in certain (unlikely) circumstances, it can be two or three times.

[sirhc]

We have a little more than 80 ubiquiti devices on that network and 15 netonix switches.

The discover command is supposed to run only once on this subnet for each of our script run, but in certain (unlikely) circumstances, it can be two or three times.

OK the number of devices is understood [95 devices].

But how often is this script automatically run?

And when you say unlikely circumstances it is run 2 or 3 times on this interval do you mean this:
Meaning if you said it [the script] runs every hour but sometimes when it runs on the hour the script is executed 2-3 times by mistake instead of once?
This could cause 95 devices to be asked to report up to 3 times so there would be up to 285 advertisements?

[aurelien personnaz]

Our script is run once a day, but can be manually called if needed.
And in the end I'm pretty sure that the discover command is ran only once per subnet.

And I forgot to mention it, but during the tests I did to identify the problem, even a single discovery call from the Web interface of a uibiquiti device on the bridged network was messing up the STP tree for the time it was running.