Forum Discussion

Son_of_Tom_1379's avatar
Son_of_Tom_1379
Icon for Nimbostratus rankNimbostratus
Nov 19, 2014

F5 Virtual Edition sensitive to SAN controller failover

We have a virtual edition running on a problematic SAN in one of our data centers. From time to time the primary SAN controller crashes and restarts, causing a failover to the secondary controller. We're working with the vendor to resolve this issue but in the meantime when it does occur, the F5 VE file systems suffer with errors and mounting as read only.

 

I know this problem will go away once the vendor resolves the controller bugs, but it's still an issue for upgrades and when we do need to failover the controllers. All the other guests, Linux included, have no issues (so far) with the failover of the controller.

 

My question is this, can we control how the F5 VE reacts to temporary disk access issues? Can we increase timeouts or something along those lines?

 

Thanks in advance

 

availability
storage
TMOS
TMSH
virtualization

2 Replies

Sort By
  • Chase_Abbott's avatar
    Chase_Abbott
    Icon for Employee rankEmployee
    Nov 21, 2014

    We've seen similar issues when using VMWare against NFS mounts (non-SAN), with an extended failure of +10 seconds causing stale mounts/guests going to a read-only disk. A quicker failure (5 second failover) did not reproduce this symptom.

     

    How long does the failover take and what hypervisor are you using against the SAN?

     

  • Chase_Abbott's avatar
    Chase_Abbott
    Icon for Employee rankEmployee
    Nov 24, 2014

    Odd. With Linux (in this case CentOS), the guest will mark the failed disk as read-only when it's scsi drivers times out usually 2ish times. In older versions of Linux, this was a moderately short time (2 minutes) but a lot of vendors upped this timer because VM I/O doesn't easily match up against traditional disk drivers. The last time we ran into this failure, the NFS was unavailable for the VMDK files for literally 5 minutes. Our debian installs obviously timed out and marked their volumes as read only and we had to reset servers once NFS was back up.

     

    20 seconds should be way to short a time for a VMWare guest to mark a volume as read-only. " target="_blank">The issue here if you didn't see this already.

     

    What version of BigIP are you on as CentOS is of course a RHEL derived?

     

Related Content