[Info-vax] Correct way to dismount shadowed volume set as late as possible?

Phillip Helbig undress to reply helbig at asclothestro.multivax.de
Fri Mar 26 05:37:44 EDT 2021


In article <1bf1f7ab-6216-4a8a-bdd0-069bd0cd822dn at googlegroups.com>,
Matthew Wilson <mwilson at mattwilson.org> writes: 

First, try to avoid long lines so that they aren't quoted-printable 
split.  Just because it looks OK to you doesn't mean that it looks OK in 
all newsreaders.  :-|

> I have a 3-node simh VAX cluster. Because there isn't a great way to do sha=
> red-system-disk clusters with simh (except netbooting satellite nodes), eac=
> h node has its own system disk.

Similar situation, but with real Alpha hardware.

> To share things like the UAF database, audit log and journal, etc., I set u=
> p a volume shadow disk between 2 of the nodes. So, for example, early in bo=
> ot I check that the cluster disks are available mount with:
> 
> mount/sys/noassist dsa1: /shadow=3D($10$DUA1,$20$DUA1) CMN01 DISK$CMN01

What happens if they are not available?  I have a loop with F$GETDVI and 
so on to determine when to mount.

> Then start using the shared data in SYS$STARTUP:SYSECURITY.COM to set VMS$A=
> UDIT_SERVER and VMS$OBJECTS, then all the other shared data in SYLOGICALS.C=
> OM.

Right.

> Anyway, my problem is that when I want to shut down the whole cluster, I'm =
> trying to figure out the right way to dismount that volume, DSA1, such that=
>  when the cluster starts up again, it doesn't say it was dismounted incorre=
> ctly and/or need to perform a long merge operation.

I had the same problem.

> For my other shared shadow volumes, I just do a dismount/cluster on them be=
> fore the cluster shutdown and they come back online smoothly.

Depending on what activity there is, that might not always work.  As you 
note, on your non-system-disk cluster disk, there are open files.

> But I can't just dismount DSA1 before running the @SYS:SYSTEM:SHUTDOWN comm=
> and, because it obviously has data that's in use all the way through most o=
> f the shutdown process until the audit server is shut down.=20

Right.

> Hence my question -- what is the correct way to cleanly dismount such a sha=
> dow set that is needed all the way through audit server shutdown? Or is the=
> re an entirely different approach I should be looking at?

On my non-system-disk cluster disk, this procedure is called from 
SYS$MANAGER:SYSHUTDWN.COM; adapt to your needs.  (In general, the SY* 
procedures often call a (similarly named) procedure on the common disk.)

$  WSO := WRITE SYS$OUTPUT
$  WSO ""
$  WSO "Executing cluster-wide common SYSHUTDWN.COM"
$  WSO ""
$  SET NOON
$  X := SYS$SCRATCH:OPEN_FILES.TXT
$  PIPE WRITE SYS$OUTPUT F$GETSYI("NODENAME") > 'X'
$  PIPE WRITE SYS$OUTPUT "" | CONVERT/APPEND SYS$PIPE 'X'
$  PIPE SHOW DEVICE/FILES DISK$USER | CONVERT/APPEND SYS$PIPE 'X'
$  WRITE SYS$OUTPUT "-------------------------"
$  WAIT 0:0:10
$  DIR/DATE 'X'
$  WAIT 0:0:10
$  TYPE 'X'
$  WAIT 0:0:10
$  WRITE SYS$OUTPUT "-------------------------"
$  DEASSIGN/TABLE=LNM$SYSCLUSTER 'F$GETSYI("NODENAME")'_UP
$  WRITE SYS$OUTPUT "-------------------------"
$! dismount non-needed disks
$  @ CLUSTER_MANAGER:SHUTDOWN_DISMOUNT.COM SHOW ! change to DISMOUNT?!
$! stuff which I had to add
$! kill the DECwindows server? needed? RIGHTSLIST.DAT
$  @ CLUSTER_MANAGER:STOP_PROCESS.COM DECW$SERVER_0 KILL
$! layered products
$  @ CLUSTER_MANAGER:STOP_TCPIP.COM
$  @ SYS$STARTUP:MMOV$SHUTDOWN.COM
$  @ SYS$STARTUP:DFG$SHUTDOWN.COM
$  SET ACCOUNTING/DISABLE
$    OPEN/APPEND/SHARE BOOT_HISTORY CLUSTER_MANAGER:BOOT.HISTORY
$    WRITE BOOT_HISTORY "''F$GETSYI("BOOTTIME")' ''NODENAME' ''F$TIME()' SHUTDOWN" 
$    CLOSE BOOT_HISTORY
$  EXIT

Notes:  Some of the above might be obsolete.  It works on at least 8.4 
and 7.3-2.  There might be better ways to do it now, but if it ain't 
broke, don't fix it.  CLUSTER_MANAGER points to the equivalent of 
SYS$MANAGER on my non-system-disk cluster-common disk.  You can see a 
list of open files to give you hints if you missed shutting something 
down.  I add a line to CLUSTER_MANAGER:BOOT.HISTORY at startup and 
shutdown.  Basically, I started with your problem and eliminated the 
open files one by one until I could get a clean dismount.




More information about the Info-vax mailing list