dismounting and mounting shadow sets

[IFX_1]

Dear all,
How can a I dismount all shadow members while the cluster is up and running?

How can I mount the shadow set member (the one I dismount in my 1st question) after maybe half an hour?

Hope to hear any replies...

regards,
ronald

[Bart Zorn_1]

Hello Ronald,

You can dismount all members of a shadowset by dismounting the DSAnnn device itself. Or you can dismount individual members by specifying their device name in the DISMOUNT command ($1$DGAnnn, $m$DKAnnn etc). DISMOUNT will not let you dismount all members, at least one must remain.

You can remount dismounted members with the MOUNT command, like:

$ MOUNT/SYSTEM DSAn/SHADOW=$1$DGAnn volumelabel

If you are running at least V7.3-2 with the SHADOWING update, you can use mini copy. Specify /POLICY=MINICOPY with the DISMOUNT command to enable it.

HTH,

Bart Zorn

[Heinz W Genhart]

Hi Ronald

before you try to dismount a whole shadowset with dismount/cluster DSAx, I would recommend to check for open files. If there are open files you will not be able to dismount the whole shadowset.

On evey node in the cluster do (or use SYSMAN with /Env=Cluster)

SHO DEVICE/FILES DSAx

The only open file you should see is INDEXF.SYS

If there are other open files, then you have to close them prior to dismounting the Shadowset (Stop the applications).

Regards

Heinz

[Wim Van den Wyngaert]

Just tested dismount/cluster on a 6.2 1h3 node. With activity on the disk that was mounted with /SYSTEM and 2 members.

You get a WARNING that the disk can not be dismounted. Not an error. But the disk is not dismounted.

I added /override=check to the dismount and the WARNING became Informational. But still the disk was not dismounted. I could still access it but it shows "mounted dismount" and dismount sees it as "not mounted".

When all activity was stopped, the disk finally dismounted.

I remounted it and the disk went into MERGE. Funny.

BTW : without /cluster the check for open files is also done and the dismount is refused too.

Wim

[IFX_1]

Hi All,
If there is an open files, is there a way that I can dismount it and mount it again?

How can I get rid of the open file drastically just to dismount the shadow set member?

[Wim Van den Wyngaert]

On all nodes of the cluster you do a show dev dsaX/fi/nosys and you find all PID's that are in it (yes, scripting). Then you do stop/id of them. It is always possible that new processes start using it while you are killing them.

Suggestion for HP : implement dismount/stop_processes.

Wim

[Robert Gezelter]

Ronald,

As noted, dismounts will not complete until after all connections to the disk are released.

After ANY elapsed time, some class of merge/copy will be needed to bring a disconnected member back into the active shadow set.

However, let us go back to the beginning. What is the reason for breaking and reforming the shadow set in the first place?

- Bob Gezelter, http://www.rlgsc.com

[Jan van den Ende]

Re Heinz:

For completeness, note that if quota are enabled on the volume, then QUOTA.SYS joins INDEXF.SYS as file that may still be opened for DISMPOUNT to proceed.

just my EUR 0.02

Proost.

Have one on me.

jpe

[Wim Van den Wyngaert]

May be read this (dismount command)
http://h71000.www7.hp.com/doc/82FINAL/9996/9996pro_66.html#blue_76

An advantage of doing dism/over=checks is that the cache is flushed when doing the command.

Wim

[Robert Brooks_1]

How can I get rid of the open file drastically just to dismount the shadow set member?

---

For a multimember set, you can dismount a member without closing any open files.

For a single-member set, you cannot dismount the member; you must specific the virtual unit. In that case, of course, open files are an issue.


-- Rob

[IFX_1]

"However, let us go back to the beginning. What is the reason for breaking and reforming the shadow set in the first place?"

The reason why I was going to break and reform again the shadow set is due to high I/O after one node in a cluster crashed. The server performance was crawling opting my users to complain. I don't have a choice but to break the shadow set member to stop/abort the merge process.

Do you have any idea how to do it?

[Karl Rohwedder]

Ronald,

if several shadowsets are merging you may limit its number with the system parameter SHADOW_MAX_COPY to lower the systemimpact (its a DYNAMIC parameter, also look at SET SHADOW/EVALUATE).

If its just one shadowmerge, you may decreasse its system impact using the logical name
SHAD$MERGE_DELAY_FACTOR[_disk] (See shadowing manual).

regards Kalle

[Wim Van den Wyngaert]

I have shadow_max_copy on 1 on the primary node and on zero on all others. Thus maximum 1 merge/copy is done at the time.

I also defined shad$merge_delay_factor as 400 to minimize merge disturbance.

For shadow copies I raised the question too.
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=707709&admit=-682735245+1151474017231+28353475
With no real result except of manualy breaking it with a script.

Wim

[Robert Gezelter]

Ronald,

Thank you for the clarification about why. As has been suggested, there are a variety of parameters which control the rate and way that things rebuild.

The Shadowing manual does a good job of documenting these parameters. There are also have been several good sessions by Keith Parris and John Atoz on using shadowing. Many of Keith's sessions are available at http://www2.openvms.org/kparris .

Having analyzed and resolved several such situations, I would recommend care. I have seen some situations where things were straightforward, and others where the answer was a change in usage/organization of shadowing (particularly in situations where shdowing is done over remote links).

- Bob Gezelter, http://www.rlgsc.com

[Volker Halle]

Ronald,

to stop an active shadow-merge, just dismount ONE member of the shadowset (assuming a 2-member shadowset):

$ DISM/NOUNL $1$DGAx:

This will drop the member from the shadowset immediately and cluster-wide. The merge will be aborted immediately, but you will need a full shadow-copy to re-mount this member into the shadowset later. No need to be concerned about open files etc.

Volker.

[Jan van den Ende]

Ronald,

which version of VMS is this?

If you are on (at least) 7.3-2 (and for this and several other reasons you really should) and you have your VOLSHAD patches in place, you simply configure HBMM (HostBasedMiniMerge).
Detailed instructions are in the patch release notes. Really, the hardest part is reading those.
From that moment on, your aforementioned problem will have a duration measured in seconds.

But you still need to live through the shadow copy one more time now...

hth

Proost.

Have one on me.

jpe