[Info-vax] NEW HBVS question

[Michael Moroney]

helbig at asclothestro.multivax.de (Phillip Helbig (undress to reply)) writes:

>In article <qn3go7$um0$1 at pcls7.std.com>, moroney at world.std.spaamtrap.com
>(Michael Moroney) writes: 

>> Did you shut down the cluster properly? Meaning that shadowset was
>> correctly dismounted by the shutdown procedure?

>Yes.

>> Do you have minicopy/minimerge enabled?

>Yes.  Of course, if there was a minicopy or minimerge it might have been 
>too quick to notice.  However, if all members were powered down, then 
>the bitmaps disappear, right?

>> Or did you say the two shadowsets were mounted on the first two nodes,
>> with one member per shadowset, and the third node added a disk to each
>> of two shadowsets but there was no copy?  

>Right.  Nodes A and C came up.  All the other shadow sets behaved as
>expected.  In particular, there was a full copy on a third shadow set
>with members on A, B, and C.  Two shadow sets have members on A and B. 
>B came up after A and C were up and had mounted all shadow sets with all
>disks which were available at the time.  So, for a while, these two
>shadow sets had only the members connected to A (all local SCSI
>connections, no dual-ported disks).  When B came back up and served the
>disks which are the other members of the two shadow sets, there was no
>full copy. 

>> Again do you have minicopy? 

>Yes.

>The difference to the three-member shadow set is that there were no 
>WRITEs to the two shadow sets in question, so actually a copy was not 
>required.  My question is how HBVS could have known that there were no 
>WRITEs.

>Yes, I could have missed a minicopy.  I thought that if I shut down the 
>entire cluster (properly or not), then the bitmaps used by minicopy 
>would disappear, since they are only in RAM.  Or are they written to 
>disk (at least in case of a proper shutdown)?

It's kind of hard to figure out from what you have posted. I suggest you troll
through the OPERATOR.LOGs to see what shadowing has been up to since this
instance of the cluster was created.

As goofy as shadowing can be at times, one thing I've never seen it get wrong
is not doing copies/merges when it should. We'd know about that very quickly 
since getting that wrong will show up rapidly as bizarrely corrupted data.

What do SHOW SHADOW and SHOW DEVICE say?