msa1000 dual controller active state flapping

[easynet]

hi,

i have an msa1000 dual controller and 1 2/8 brocade. inserting the second controller, they went in active/passive mode. i tested the failover system ejecting the active controller and the failover goes up. but when i insert the controller again, they flap the active state every 1 minutes, so now i ejected the failover controller to have stability. how can i resolve ?

[Amol Garge · Accepted solution]

ok, 2/8 seems like an internal MSA switch. If yes, then you will need two of these as the controllers do not communicate with the same switch from the single bay.

Now, assuming that you do not have any hosts connected to it(since you pulled out the active controller to check the failover).

Shut down the MSA and insert only single controller.

[easynet]

yes, it is. infact a connected an i/o module and a fiber channel cable from it to the switch 2/8. it works and the controllers comunicate with the esxi nodes well. the first time i inserted the failover one, they do no flap, only afther fail test, now they flap

[Greybeard]

You don't say what firmware your controllers have, 4.xx = pull out the controller to fail over, 5.xx do NOT as they work very diffrently and now have a shutdown/reboot command at the CLI to do failovers.

[Amol Garge]

ok, in that case the first thing I would suspect the multipath setup on the ESX.

Do you have any preference setup on the server regarding the path/controller?

What is the firmware on the MSA controllers?

[easynet]

=> controller wwn=500805F30000C141 show

MSA1000 at NWINETD001
Bus Interface: Fibre
WWN: 500805F30000C141
Serial Number: PB9840TX3TY0Y4
Chassis Serial Number: 8A23JN71B091
Chassis Name: NWINETD001
RAID 6 (ADG) Status: Enabled
SSP State: Enabled
Controller Status: OK
Chassis Slot: 1
Hardware Revision: Rev A
Firmware Version: 4.48
Rebuild Priority: High
Expand Priority: Medium
Surface Scan Delay: 3 sec
Cache Board Present: True
Cache Status: OK
Accelerator Ratio: 50/50 (read/write)
Read Cache Size: 128 MB
Write Cache Size: 128 MB
Total Cache Size: 256 MB
Battery Pack Count: 2
Battery Status: OK
Host Bus Adapter Slot: 3

[Amol Garge]

ok, now I know the firmware on MSA is 4.48.
I want to know about ESX multipathing..

[easynet]

i did not configured it on esxi nodes. infact when i ejected conttoller A, node didnt see the uuid, and only rebooting them the LUN where rescanned and esxi works again. but why moltipath can influence flapping ? before i removed controller A they where right !

[Amol Garge]

Ok, now the problem is clear:
MSA1000 controller does not fail back automatically. The only active path it can see is from the previously active controller. Once you took it out, the OS didn't see the LUNs since the other path was not setup. Once you put back the original controller, there was nothing to tell the now active controller to failback and hence the LUN ownership started fluctuating between the two controllers.

Now, go ahead and setup multipath, so that there is a proper failover and manual failback if needed.

[easynet]

ok, begin to manual failback . how can i say the msa1000 to failback primay state to controller A e failover state to controller B. i use acu-cli commands.

[easynet]

in this forum i sow Tosten to say :
Even if you have an OS that supports and expect dual active/active pathes, you need special software for active/passiv arrays, because every access to the passive ctrl would cause the controllers to switch. This would result in a endless fail-over from one controller to another.

this is my case i thing, becouse rebooting 1 esxi, it take the path from the failover controller, so having 2 nodes esxi each using a different controller path, the active state flaps at each connection request ?

[Uwe Zessin]

From what I have seen in the past, ESX automatically finds out if an array supports active/active or active/passive mode and uses an appropriate multipath policy (static vs MRU) - there is no additional filter like HP Secure Path / AutoPath necessary or even available. An end-user _can_ overwrite this (unintentionally or by not knowing better) and use a wrong policy. Can you check the policy?

Last time I checked, you were not allowed to mix different module types (switch, hub, plain I/O) in the interconnect bay. In your case I would at least set up switch zoning so that a single server HBA does not have access to both controllers - from what I have seen the MSA can act strange if that happens.