Stackwise-480 behavior with a simple 2-switch C9300 stack

[RB6502]

I just finished reading the Stackwise-480 Whitepaper . It's a very informative document, but it did not answer the question I have about my simple scenario.  I currently do not have access to these switches, but I need to understand their expected behavior so I can create a test plan to verify proper Stackwise-480 operation.

The scenario is pretty simple...there will be two C9300 switches connected together via Stackwise-480 cables.  The stack will have six Etherchannels consisting of one port from each switch; five Etherchannels serve as downlinks to five access-layer switches, and one Etherchannel servers as an uplink to a higher-level switch.

So once the two switches in the stack get past the election process and are fully booted up, there will be one Active switch and one Standby switch.  My questions are:

•  Assuming a steady-state, with both switches in the stack powered on in each scneario, what should I expect to happen when:
  • I power off one of the switches in the stack without disturbing the Stackwise-480 cables?
  • I restore power to the powered-off switch in the stack?
  • I disconnect one of the Stackwise-480 cables? (I'm assuming no significant impact to downstream forwarding but with a reduction to half-duplex speed over the Stackwise-480 fabric)?
  • I disconnect both Stackwise-480 cables (I'm assuming some kind of undesirable "split-brain" behavior)?
  • I reconnect one of the Stackwise-480 cables? (Does either/both switches reboot)?
  • I reconnect the remaining Stackwise-480 cable? (Does either/both switches reboot)?

[balaji.bandi · Accepted solution]

STACK - Physically 2 and Virtually 1 - Hope you got this concept.

Assuming a steady-state, with both switches in the stack powered on in each scneario, what should I expect to happen when:

• I power off one of the switches in the stack without disturbing the Stackwise-480 cables?   -  if this is master switch - standby will take over start processing the data packet, end device will not see any interruption - if the config did best practice
• I restore power to the powered-off switch in the stack. - since standby has already become master, this will be standby even though it was set up switch1 and high priority- that is the nature of the stack (port information will not be changed)
• I disconnect one of the Stackwise-480 cables? (I'm assuming no significant impact to downstream forwarding but with a reduction to half-duplex speed over the Stackwise-480 fabric)?  - stack ring will break - you should not see any traffic loss, but you see logs.
• I disconnect both Stackwise-480 cables (I'm assuming some kind of undesirable "split-brain" behavior)? yes you guessed correct - this is split brain, but become master - in this state (quick solution - go to Core switch - shutdown the interface connected to standby switch, so master will be able to serve traffic (when I said standby switch2 - when you break both the cables both become master - so this is quick solution before engineer go to site and fix)
• I reconnect one of the Stackwise-480 cables? (Does either/both switches reboot)? - based on the switch priority they check any one has master already, stand reboot and join the master. (since you looking to test in lab, so you can test this )
• I reconnect the remaining Stackwise-480 cable? (Does either/both switches reboot)? - since they already formed stack - there is no effect when you connect the cable - this time you have full ring

I replied as per I have experienced so far.

[Joseph W. Doherty]

• I power off one of the switches in the stack without disturbing the Stackwise-480 cables?   -  if this is master switch - standby will take over start processing the data packet, end device will not see any interruption - if the config did best practice

I've bolded part of what @balaji.bandi wrote, above, to emphasis additional manual config may be need to accomplish this.  In particular, review the parts of the document about NSF.

Also, quickly reviewing the document, unclear to me whether an issue found going back to 3750 stacks has been addressed, by default, i.e. what happens to the stack MAC when standby takes over stack.  (On the 3750 series, by default, stack MAC would change, but this behavior could be changed by configuration.)

BTW, OP mentions disconnecting and connecting cables, in many cases, with both switches having power.  Again, unsure about this series of switches, but on earlier stacks, it was recommended that one of the two switches being connected should not be powered, and only powered on after stack cables were connected.

[RB6502]

Agreed, but I'm trying to test "what if/oops" situations so the customer knows what to expect, worst-case.

[Joseph W. Doherty]

Well, just be aware, I believe why Cisco suggests having one side of the stack cable connected powered off is to avoid possible equipment damage.  I.e. somethings are designed for "hot" connecting/disconnecting, some not.  I.e. oops are just that, but some "intentional" tests, to emulate some "oops", should be avoided.

[Thomas Schmitt]

---

@RB6502 wrote:

So once the two switches in the stack get past the election process and are fully booted up, there will be one Active switch and one Standby switch.  My questions are:

•  Assuming a steady-state, with both switches in the stack powered on in each scneario, what should I expect to happen when:
  • I power off one of the switches in the stack without disturbing the Stackwise-480 cables?
    • --> In Case this was the master or standby, new election process will start, otherwise what line card remains provisioned, but not functional
  • I restore power to the powered-off switch in the stack?
    
    • --> new member will be added to stack, regardless of configured priority
  • I disconnect one of the Stackwise-480 cables? (I'm assuming no significant impact to downstream forwarding but with a reduction to half-duplex speed over the Stackwise-480 fabric)?
    • --> yes, it will be half
  • I disconnect both Stackwise-480 cables (I'm assuming some kind of undesirable "split-brain" behavior)?
    • --> you should connect connect switches like a ring, so even if you remove both cables from a single switch, there won't be a split brain situation
  • I reconnect one of the Stackwise-480 cables? (Does either/both switches reboot)?
    • --> stack member do not reboot, in case you reconnect previously in this  stack known switch, it will just came up with one stack port. In Case you replaced that stack member, it is possible (bun not have to), it will reboot
  • I reconnect the remaining Stackwise-480 cable? (Does either/both switches reboot)?
    • --> the same as before, but it's recommended no to do so

it is recommended to have the stack cables connected to the newly added switch prior to powering it on. If the stack cables are connected after the switch has been powered on, it will result in reload of the newly added switch.

---

[RB6502]

• I disconnect both Stackwise-480 cables (I'm assuming some kind of undesirable "split-brain" behavior)?
• --> you should connect connect switches like a ring, so even if you remove both cables from a single switch, there won't be a split brain situation

But, now you will have two switches that were previously stacked, with Ethernchannels created from ports on both switches, turn into two independent switch "stacks" of one switch each, that are still both operational & connected to downstream & upstream switches.  I think there's the potential for some downstream traffic to be blackholed in this situation?

[Thomas Schmitt]

If 2 stack fails apart, both parts continue to switch traffic. In case you use MEC, LACP/PaGP will detect split and err-disable portchannel, I think also mode ON won’t remain up - so I don’t see a possibility for blackholing