stack-mac persistent timer lacp port-channels C3750

[velo84]

I am having an issue on a Cisco 3750 stack where when the stack master is rebooted, all my lacp port-channels drop and then come back up again. After doing some investigation It seems that it is happening because of lacp using the stack master mac-address as part of the system-id, so when the stack master reboots, the stack mac changes. I see that there is the command: stack-mac persistent timer 0

http://www.cisco.com/en/US/docs/switches/lan/catalyst2960/software/release/12.2_53_se/configuration/guide/swstack.html

There is this warning about using this command:

When you configure this feature, a warning message displays the consequences of your configuration. You should use this feature cautiously. Using the old master MAC address elsewhere in the domain could result in lost traffic.

My question are:

1. Are there any other consequences to using this command (apart from moving the switch/mac to another location in the network)
2. It mentions 'If the entire switch stack reloads, it acquires the MAC address of the master as the stack MAC address' Is this still the case if you have the stack-mac persistent timer to 0?
3. Does using channel-group mode on for the port-channels still use the same mechanism of having a system-id? (Will the channels flap using 'mode on' when rebooting the stack master.

Many thanks

Mark

[Giuseppe Larosa · Accepted solution]

Hello Mark,

1) the only case that the command does not cover is when the current master switch is isolated from the stack but still alive and connected to the rest of the network.

If this happens the stack becomes partitioned with the old master becoming a single switch stack and the rest of stack electing a new master that will use the same MAC address.

The same MAC address would be used causing communication problems.

A part from this the MAC persistency should provide a better behaviour for LACP bundles as you have noted.

2) with stack mac timer to 0 the stack mac address should be always the same, the sentence is true in any case with and without persistency enabled the mac stack is chosen by the stack master, with persistency the last used stack mac address is remembered and used even if the master switch changes.

3) mode on means unconditionally bundling no control messages are exchanged on member links so no system-id concept is present here. However LACP based bundles are to be preferred for their capability to detect problems on each member link and also errors in cabling

Hope to help

Giuseppe

[velo84]

Giuseppe

Thanks very much for taking the time to answer. Sorry just one more thing, to confirm, are you saying that if you set the stack-mac time to 0 (indefinitely) it will still be the same mac for the stack even if the old master got removed and replaced by another switch and the whole stack got rebooted and a new master was elected for example?

The reason I ask this is because on the Cisco article:

http://www.cisco.com/en/US/docs/switches/lan/catalyst3750x_3560x/software/release/12.2_55_se/configuration/guide/swstack.html#wp1233369

There is a note which says:

Note

If the entire switch stack reloads, it acquires the MAC address of the master as the stack MAC address.

Thanks

Mark

[Giuseppe Larosa]

Hello Mark,

I haven't tested this aspect so I may be wrong in assuming a so high level of persistency.

However, my reasoning is that MAC persistency should still be there if the master is powered off, as this is one of the possible failures of the current master, and the new master will use the previous master MAC address for some time or permanently ( if the timer is 0)

I admit that the case of power off of all member switches may be different. It all depends on how this MAC persistency is implemented. The stack MAC address might be stored in NVRAM of member switches or not.

However, in the same link that you have reported of configuration guide 12.2(55)SE we can see the following:

>> If you enter

0

, the stack MAC address of the previous stack master is used until you enter the

no stack-mac persistent timer

command, which immediately changes the stack MAC address to that of the current stack master.

If you do not enter the

no stack-mac persistent timer

command, the stack MAC address never changes.

This leads me to think of the stack MAC address stored in NVRAM as part of the configuration or separately

Hope to help

Giuseppe

[velo84]

That was my feeling as well, many thanks for your help.

Mark