02-06-2015 12:04 PM
I see that I can run the no split-detection command on a 2 switch stack and have it disable split detection, but the manual says this can't be done on a stack containing 3 or more members. Is there any way to make this happen on larger stacks?
02-07-2015 10:43 AM
Is there any specific reason, why you want to have 'no split detection' configured for bigger stacks ?
- I believe there is no way to have ‘no split-detect’ for stack having more than 2 members/units.
Let me try to put perspective as to why it is "designed" like this:
Split detection is "essential” to ensure that there are “NO” identical/duplicate stacks “operational” in the network. Or else it will lead to other network issues such as duplicate IP-addresses, etc…
Hence, if there is stack split, then only one of the ‘split-part’ should be operational, while the other split-part should be ‘dormant (non-operational)’.
And there are certain rules that govern which one is put into ‘operational or dormant’
- If a group has both the primary and secondary stack units prior to the split, it becomes the active stack; the other group becomes dormant and interfaces on those units are removed from the primary’s operational state.
- If the primary and secondary prior to the split are no longer in one group, the group having the majority of the stack-units (relative to the original stack) becomes the active stack (and if necessary secondary transitions to primary). The group with minority of stack units becomes dormant.
- In case the isolated groups have equal number of stack units the group with secondary becomes the active stack.
And, stack with ONLY “two” members/units is “special” case of last scenario mentioned above.
Now, let’s understand when does split occurs
a) Stack link connectivity is lost, either due to
- admin shutdown of stack port
- deletion of stack port (to reclaim it as normal network port)
- Cable is removed
b) Stack unit/member goes away, either it is being
- Taken out, and moved to different location
- Powered down.
Now, if we have scenario (iii) above & last point in (b) above.
i.e, Only two members & Secondary is powered down.
==> In this scenario, secondary is anyway out of picture, but due to role (iii) above, even the primary also becomes ‘dormant’. Resulting in neither of the “split” parts of stack are ‘operational’ …!!
Hence it is required to configure, ‘no split detect’ only for this scenario. So that Primary can ingore split detection, an transition to be 'operational' & not dormant.
For all other scenarios, it is essential to have detect split to prevent two 'duplicate' (config-wise) stacks.
02-11-2015 08:08 AM
Thanks for the response. I'm aware of the purpose for split stack detection and the exception for 2 switch stacks. The situation one of my clients is in is that they have 3-5 switch stacks in closets with bad power and single uplink ports. I always set the switch with the uplink back to the core as the primary. What we have seen is situations where 2 switches in a 4 switch stack go down, but the primary (and consequently the uplink) is still up. With split stack detection enabled, the whole closet goes down. Given we have no possibility of 2 identically configured stacks running after a split, I want to shut down detection because if the primary and path out are up, we want the closet up (even if 3 of the 4 switches are down).
We're in the process of getting them better UPS's, so this may become a moot point, but is there any way to disable split stack detection on a stack with 3 or more members?
02-11-2015 12:06 PM
yes, Understood your deployment, and problem faced due to unreliable power...
But, Not possible to disable split stack detection.
Usually in Stack setup, there would be "multiple" uplinks drawn towards aggregation/core, whcih apart from increasing the over-all the bandwidth also help as redundant links - for the situation you explained.
And, stack split detection is purposefully inbuilt into the system, and not left to the assumption that user will ensure that, there is no duplicate stack (config) running in the network, except for two node stack.
Anyway once you have proper UPS in place, that would help come out of this problem..