Level 3 kit to fix system crashers?

[Jerry Eckert]

The release notes for kit VMS732_CLIUTL-V0100 indicates that it fixes several system crashers:

Sec. 5.2.5 -- A SHOW PROCESS command on a NULL process may cause an INVEXCEPTN crash.

Sec. 5.2.18 -- SHOW PROCESS/CONTINUOUS results in PGFIPLHI crash

Sec. 5.2.19 -- SHOW CLUSTER command can crash system

Why are the details of system crashers documented in the release notes? And why is the kit rating 3 instead of 1?

[Hein van den Heuvel]

Hi Jerry,

Those are reasonable questions, but this is IMHO, an unreasonable place to ask them.

While this web-site is under 'itrc.hp/service' and you may have obtained that kit and release notes from an other itrc.hp site the forum here have NOTHING to do with the official support process.
The forums is at best an effort peer-to-peer environment for which HP provides the infrastructure and occasionally some moderators.

Please raise this issue diretly with support.

Regards,
Hein.

[Jerry Eckert]

Thanks for the response Hein.

The main reason I posted here was to make others aware that there are fixes for potentially serious problems hiding behind the rating 3 label. If someone at HP notices it before I have time to spend an hour dealing with Bangalor, so much the better.

[John Gillings]

Jerry,

Like Hein says, for an official answer, you need to ask HP.

That said, having been involved in rating patches in the past... the rating system is based on a (fairly loose) interpretation of both the seriousness and the likelyhood of experiencing problems that the patch corrects.

That this patch is given a low rating even though it fixes serious problems suggests to me that they have a very low probability of occurring. It's all a matter of balancing the cost of downtime to install the patch with the potential cost of suffering a crash because the patch wasn't installed.

In some cases a particular system crash may only have been seen on a single site, maybe even only once. The fix still gets into the remedial stream and may be included in a patch along with other fixes to the same components.

(I can't find the actual kit you're talking about, so can't check the wording of the release notes - where did you find it?)

[John Gillings]

Jerry,

Did you mean VMS82A_CLIUTL? It's not an exact match with your section numbers, but it fixes similar sounding problems. The details say:

>A SHOW PROCESS/CONTINUOUS command can
>result in a PGFIPLHI crash on memory
>constrained systems.

That potentially means *extremely* memory constrained. Enough to cause a pagefault at the wrong place and time. More than likely the extent of the constraint is such that you wouldn't expect to find it in any serious production system.

Similarly, the SHOW CLUSTER bug sounds very obscure:

>Errors such as mis-configuration of
>NPAG_BAP_* parameters, insufficient non-
>paged pool and a missing
>CLUSTER_AUTHORIZE.DAT can cause a SHOW
>CLUSTER command to crash the system when
>the PEdriver is partially initialized.

Again, these are not circumstances you'd expect to find in a production system.

OpenVMS customers sometimes report quite bizzare sets of circumstances which cause trouble. More often than not, engineering fix the underlying problem, but given the rarity of the circumstances, they don't think it's critical enough to ask all customers to spend downtime installing a patch for something they're extremely unlikely to experience.

The release notes are there for you to read to override the recommendation of OpenVMS engineering should you so wish.

You need to consider the cost of downtime, the real risk to you of the problem the patch solves and the risk of the patch itself introducing some other problem.

[Jerry Eckert]

Thanks for your responses, John.

Alas, we're still running V7.3-2, so I was looking at the release notes for that version of the kit. I received them via email from OpenVMS.org.

I agree with your assessment that the conditions under which the SHOW CLUSTER crash is documented to occur are sufficiently rare that it is probably not a significant risk.

The documentation for the PGFIPLHI crash does not quantify "memory constrained", so it is difficult to judge the impact of this problem.

My primary concern is the SHOW PROCESS NULL crash since the documentation does not specify that other conditions must be present. The complete problem description is:

A SHOW PROCESS command on a NULL process may cause an INVEXCEPTN crash at SHOW+030350.

There are no workarounds provided.

Jerry

[John Gillings]

Jerry,

>Alas, we're still running V7.3-2, so
>I was looking at the release notes
>for that version of the kit.

I didn't think patches were still being produced for V7.3-2? I couldn't find "VMS732_CLIUTL-V0100" on ITRC, so where is it supposed to come from?

>My primary concern is the SHOW PROCESS
>NULL crash since the documentation does
>not specify that other conditions must be
>present

When was the last time you saw a NULL process? I think it stopped being displayed in SHOW SYSTEM around the time of V5.0, and as of Alpha NULL isn't really a process at all, just a loop somewhere in the guts of the PROCESS_MANAGEMENT execlet.

It's not entirely clear what the description means, but I'm fairly sure it's a pathological case that you have to be fairly inventive to exercise.