OpenVMS will not increase error count when disk is removed from ES45 machine

[tqs_reuters]

Hi

We are using ES45 machines running OpenVMS V7.3-2.

When we pull out one disks (while it is mounted) from the rack of the ES45, we expected an error count everytime we access that disk. But the error count remains 0.

Is this normal or is there anyway to simulate error count increase.

This is for testing purposes for error handling.

Thanks

[labadie_1]

if you do

$ show device disk_in_question/fu
does it show, among other things

error logging is enabled.

if you do

$ reply/ena
before pulling the disk, when you pull it, do you get any message ?

[Jur van der Burg]

What kind of controller?

Notice that i/o may be satisfied from cache so that you don't really access the disk if you do a DIR DISK:[ABC] for example each time from the same dir.

Start an analyze/disk and see what happens.

Jur.

[Hoff]

I'd whack the error counter to a non-zero value, and move on. Either via ZDEC or other kernel-mode write access, or via a local modification to something like LDDRIVER to simulate errors. Sources to ZDEC and LD are available.

[tqs_reuters]

Thanks all for answering

To labadie
Yes, error logging is enabled
We'll check with the reply/enable part

To Jur
Controller is
3X-KZPEA-DB Two Chan Ultr3 Cap SCSI Adptr
We'll see if ana/disk helps

To Hoff
We'll try to use LDDRIVER and/or ZDEC

========================================

From our experience, pulling out the disk is very different from really having a faulty disk.
I think VMS can determine if disk is being pulled out or is really having problems.

[John Gillings · Accepted solution]

>I think VMS can determine if disk is being
>pulled out or is really having problems.

Not until you do something that requires an interaction with the disk. If there are no I/O operations on the disk, there will be no errors. If the "access the disk" you're talking about are reads, they can be satisfied from cache, and not need to touch the drive.

True story... a large cluster accidentally had a system disk popped out of a storage shelf. It ran for a whole week before anyone noticed. There was enough memory that the working pieces of the system disk were held in cache, and all the log files had been moved to other drives. No errors, no problems. Just reseat the drive and all fixed.

If a WRITE operation is attempted on a missing disk, it will go into MOUNTVERIFY state, until MVTIMEOUT is exceeded.

You can argue either way about this behaviour. On one hand, why report an error unless it's "real"? On the other hand, a missing disk is likely to be an error, so it makes sense to report it. Each opinion has its merits and weaknesses.

Perhaps you could create a procedure that scans all disks, and performs some operation which requires physically interacting with the disk every N minutes to detect missing drives? (Perhaps rewrite a 1 block file in the root directory?).

[tqs_reuters]

To John

Thanks, we'll try that out too