RSS Disk State

[Chris Rasmussen [FAU]]

Hi all,

On one of our EVA 8000s here the RSS Disk State for more than 1 of the disk groups says 'None'. I know this isn't good.

I've had a look through ITRC and found a couple of excellent discussions - http://forums11.itrc.hp.com/service/forums/questionanswer.do?admit=109447626+1257804744109+28353475&threadId=1003376 and http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1002866.

These give me some info about what the various RSS Disk State levels mean but what I don't understand is how this can happen at all.

It was my understanding that the RSS sets are managed by VCS and not user-controllable without what basically amounts to a 'hack' e.g. ungrouping, moving, regrouping.

How can this happen, i.e. how the RSS Disk State be 'None'? How can I fix it? The groups aren't disk-type specific and contain both FC and SATA disks (in different groups of course).

Thanks! :)
Chris

[Uwe Zessin · Accepted solution]

> It was my understanding that the RSS sets are managed by VCS

Correct (on the EVA 4/6/8xxx it is called XCS), but unfortunately the controller software is far, far from perfect. I had seen one case (not the most recent firmware, but I don't believe XCS has made any progress in that area...) where I've added disk drives to an existing disk group without any virtual disks. The parity state went from parity to mirror on an 2C12D while there were enough disk drives left (and distributed across the enclosures) to keep the parity state.

Some years back there was some rumor that engineering was thinking about adding code to actively optimize RSS states, but it seems that this area has never been touched. Worse: XCS/CV-EVA have been degraded to hide this type of information.

Hm, that rings a bell: I don't know where you looked, but there are certain software combinations where CV-EVA claims the RSS state is none, but the real state is already masked by the controller firmware.

Can you create a report using SSSU and saving the output of "ls disk full xml" and "ls disk_group full xml", put it in a .TXT file, compress it to a .ZIP file and attach it to a response? Then I could take a look at it and tell you if/which disk combinations are less than optimal.

[Víctor Cespón]

RSS state information is no longer available to the customer and is shown as "None" always.

http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01062600

To know the RSS distribution you must extract the information with SSSU and see in what RSS is each disk.

[Chris Rasmussen [FAU]]

Hi all,

I'm new to this particular role so I'm only going from screenshots right now and don't have access to run SSSU on the SAN in question just yet. My access should be granted today I hope!

Why would HP stop users from quickly viewing the RSS state information? Isn't this a vital way of telling if your redundancy is ok for a disk group? I'm a bit confused by that decision.

If I run SSSU and find that the DG does in fact have an RSS Disk State of 'None', what can I do about it? Do I have to physically ungroup, move and regroup the disks so the RSS sets are split across shelves?

Another question - can an RSS Disk State of 'None' be caused by DG physical disk placement unknowingly?

Sorry if I'm not providing very good info - it's hard to do that without access to CV-EVA. :)

[Víctor Cespón]

RSS disk state was removed because it informs you about what would happen if an antire disk shelf of disks was lost, but as this is a very rare event (after 2500 EVA cases, I can confirm this is NOT something you should worry about), it was removed from XCS and CV.

Yes, you can manually move disks around, the EVA does not care about the physical position of the disks, it uses their WWID. But RSS distribution can change after a disk failure or adding disks, a RSS can merge with another or split in two.

[Chris Rasmussen [FAU]]

Great, now we're getting somewhere.

> after 2500 EVA cases, I can confirm this is NOT something you should worry about

Ok, that's good to know!

> Yes, you can manually move disks around, the EVA does not care about the physical position of the disks, it uses their WWID. But RSS distribution can change after a disk failure or adding disks, a RSS can merge with another or split in two.

Does the RSS set *only* relate to redundancy of disk *shelves*? Do they relate to disk groups at all?

As a very high-level question is there anything that really needs to be done re RSS sets or will XCS always configure them in the best way? I'm trying to work out if there's anything I need to do once I have the full picture of how the RSS sets are configured currently (again, once I have my login to the 8000 here).

[Chris Rasmussen [FAU]]

> Can you create a report using SSSU and saving the output of "ls disk full xml" and "ls disk_group full xml", put it in a .TXT file, compress it to a .ZIP file and attach it to a response? Then I could take a look at it and tell you if/which disk combinations are less than optimal.

Yep I'll do that once my access is sorted. The scripting guide for SSSU documents "ls disk " command but doesn't mention the full or xml switches. Are they SSSU version specific?

[Víctor Cespón]

Edit attached SSSUscan.txt to put the IP address, user name, password and name of the EVA.

Run with SSSU.EXE "file SSSUscan.txt" > config.txt

Attach result here.

--------------------

RSS state = Parity means no two disks of a RSS are in the same disk shelf

RSS state = Miror means no two disks that are a mirror pair are in the same disk shelf

RDD state = None means two disks from the same RSS and that are a mirror pair are on the same disk shelf.

[Chris Rasmussen [FAU]]

Thanks, I'll do that shortly and post the results as soon as I can.

Oh and my mistake for saying the documentation for "ls disk" doesn't include info about "full" or "xml" ... it's in the command reference near the end of the document. Oops. :)

[Chris Rasmussen [FAU]]

While I wait for my access to be granted, is it possible for 2 disks in an RSS that are *not* a mirror pair to be in the same disk shelf? Or are RSS disks always part of a mirror pair? If a RSS has 11 disks the term 'pair' seems to be impossible for at least one of them. :)

[Chris Rasmussen [FAU]]

Hmmm and what if a vdisk's requested write cache is Write-back but the actual write cache is Write-through. What does that mean? Does it indicate a problem?

[Uwe Zessin]

I have seen this on virtual disks or mirror clones without a presentation to a host. Present a vdisk and the view immediately changes to "write-back".

[Uwe Zessin]

Both members of a mirror pair are always present in the same RSS, if that was the question.

If both members are on the same disk enclosure, the RSS state is "none" (if we ignore the function removal / wrong display in CV). It is possible to have both members in the same enclosure -- just build an EVA with 8 disk drives in one enclosure (yes, I've implemented a few ;-)

A disk that has no 'partner' cannot be used to store VRAID-1 data, but it is part of an RSS. In other words: create a disk group with an odd number of disk drives (e.g. 9,11,13,...) and fill it completely with VRAID-1 vdisks -- you will find out that on disk drive has no data.

[Chris Rasmussen [FAU]]

vcespon/Uwe: I'm not sure I should be attaching SSSU output for a client system to a public forum like this. Would you still be happy to look at the results of SSSU if I emailed them to you? I only collected vdisk, group and controller info - hope that's ok. I didn't get info about each individual disk as there are 200+ of them and that process takes a while.

I'm not sure how to get the RSS Disk State info if both CV-EVA and SSSU now report 'None' at all times since XCS 6.100.

Pls advise?

[Víctor Cespón]

The RSS state is no longer displayed on the public part of Command View, but it can be calculated from the SSSU output.

You can send the output of the SSSU script to vcespon@gmail.com for example.

[Matt Palmer_2]

Hi Chris,

I realise this reply is a little late, but thought it may help, as I have helped another user on the forum with a program/script I wrote to display a color map of the RSS state of the EVA, to help me with tracking RSS state changes, and keeping the shelves as protected as possible.

If you look here:
http://forums13.itrc.hp.com/service/forums/questionanswer.do?admit=109447627+1269268561102+28353475&threadId=1321854

you will see it. If its of any use to you, please award points as you see fit.

Hope that helps

regards

Matt

[Chris Rasmussen [FAU]]

Hi Matt,

Very interesting, thanks for posting.

I've already got all the SSSU output I need though and I see your script runs SSSU (for SuSE Linux?) and grabs the output as well ... I think. So, a couple of questions:

1. The SSSU output is in XML format as I've been using Excel to hack with it up until now - is that ok? Or does your script expect another format?
2. Which parts of the script would need to be removed so that it still produces the formatted output but doesn't do *any* of the EVA interrogation stuff? We have no Linux servers on this particular customer site but I do have 1 here that I can use for Linux scripts - it has no connectivity to the customer arrays though.

Thanks,
Chris

[Matt Palmer_2]

Hi Chris,

you could just modify the section containing PHP script and mysql DB if you already have output. Just make sure that the output is formatted in the same order that it is required in the array loop.

my script passes the output of a column list into a mysql DB, these columns are then queried and I run up a multi-dimensional array in PHP to draw a colour coded map of the array.

hope that helps

regards

Matt