Conference spezko::cluster

5298.0. "7.1 Shared SCSI boot problems" by CSC32::S_DANNEN (Live long and slobber) Fri Apr 25 1997 11:12

    I have a Customer running 7.1 with a Shared SCSI environment w/quorum
    disk on the Shared SCSI, that is unable to boot as single system.
    Each node provides 1 vote, the quorum disk provides 1 vote and when
    we set expected_votes=3 a single node booting hangs indefinately due
    to lack of quorum. 
    
    A forced crash revealed that the system had not configured the
    quorum disk. The controller that the quorum disk is attached to 
    is configured, normal status and on-line. No errors. STACONFIG
    is current, and SWAPPER has been created. The "Using remote access
    for quorum disk" was the last thing on the console.
    
    As soon as we bring the second node in, we get quorum, and configure
    the quorum disk, and it is later mounted in sylogicals. If we set
    expected_votes=1, the single node will come up, as well as the quorum
    disk. 
    
    Each system has it's own system disk on the internal SCSI controller,
    and this bus is not shared. I suspect a problem in the sys$pksdriver
    and was wondering if anyone had seen this behavior before. I will 
    obtain the actual dump, and perform a more indepth analysis,
    fortunately he did collect PC/Ps samples as well.
    
    /steve 
    
    The config is:
    2 Alphaserver 1000A 5/500 256MB of memory
    
    Each has own system disk connected to local SCSI controller, shows as
    PKC from the console, OVMS shows it as PKA, there is also a tape drive
    and a CDROM on this controller as well for each system.
    
    Memory Channel is installed, virtual hub mode, believe that BEVIS is
    the virtual master
    
    They have 10MB E'net between them
    Votes = 1
    Expected_votes = 3
    Disk_quorum $3$dka500 -> connected off of the PKB controller
    QDSKVOTES = 1
    QDISKINTERVAL = 1
    RECNXINTERVAL = 20
    NiSCS_LOAD_PEA0 = 1
    NiSCS_LOAD_PEA0 = 1
    ALLOCLASS -> bevis = 2
    ALLOCLASS -> buthd = 1
    DEVICE_NAMING = 1
    Port Allocation Class is set for the 3 SCSI adapters
    bevis$pka = 0
    buthd$pka = 0
    bevis$pkb = 3 ->shared SCSI
    buthd$pkb = 3
    bevis$pkc = 4 ->shared SCSI
    buthd$pkc = 4
    
    The PKB and PKC devices are KZPSA (sys$pksdriver)
    Each PKB is cabled to DWZZB in a BAxxx cab w/Rz29's
    Each PKC is cables to DWZZB in a BAxxx cab w/RZ29's
    
    
    
     

This is not a known problem.  We usually test with the system disk on the
shared SCSI bus, though, so it is possible we missed something, maybe
related to Port Allocation Class.

Just to be sure though, be aware that the quorum file is not created on the
quorum disk until it is mounted for the first time.  If you've never
mounted it, then the file is not there and it can't contribute the
necessary vote when one node boots alone.

Keep us posted.

Thanks.

Tom Coughlan
OVMS SCSI Team Leader.