• Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

Comments (20)

1 john1212 commented Permalink

This is wonderful! Are you sure? <br /> But what a young Administrators who previously worked on linux <br /> they wants rename all devices like at linux. <br /> Somebody renamed en0 to hdisk0, hdisk0 to en0? Funny? How do work? <br /> Always be able to make a link to the device. <br /> You always can change name devices, but it was difficult.

2 cggibbo commented Permalink

Thanks for the comment john1212. Yes, it is wonderful. <div>&nbsp;</div> Junior administrators can and will make mistakes. That's why they are juniors! :) They learn from their mistakes. It is the Senior admins job to provide them with the right "spiritual" guidance and discipline so that they make fewer and fewer mistakes. <div>&nbsp;</div> Common sense is also required/expected when working as a system administrator. Allocating device names that have some meaning is common sense and expected.

3 AnthonyEnglish commented Permalink

I wonder if there's an option to change a device that's in use by means of the -P flag so it would take effect after the next reboot. Not sure you'd want to do it anyway, but it could be helpful. <div>&nbsp;</div> This chdev renaming complements the ability to rename LVs, as you outlined in an earlier post, allows you to get around that multibos feature where the LV names for the standby innstance are suffixed with _bos. <div>&nbsp;</div> One last comment: renaming en0 to hdisk0 and vice versa is a bit like creating a link to the rm command and calling it "ls". You could do it once at someone else's expense, but probably not a great career move.

4 cggibbo commented Permalink

Hi Anthony, a -P flag is a nice idea. I'll mention it to the development folk. Will let you know.

5 john1212 commented Permalink

I agree with you. <br /> Administrators creates many virtual servers, for base servers, for clusters. Everywhere Disk Storage. The ability to change the name of devices was expected.

6 Steve_Snape commented Permalink

Something like the –P option would be great! <br /> I’m sure for this with OCD like me, like to see rootvg on hdisk0 :o) <br /> There used to be an “unsupported” way to update the odm to change a rootvg hdisk to hdisk0, but it was always a bit risky, and I never got it to work after 5.3. <br />

7 cggibbo commented Permalink

Here's some feedback I received from the AIX development team: <div>&nbsp;</div> "Is it possible? Yes, maybe. But it is not a simple change. The -P flag with the chdev command changes an attribute just in ODM with the result that the ODM is out of sync with how the device is actually running. The worst problem with this is that if one now runs lsattr to display the device's attribute settings, they see what is in ODM and not how the device is configured. Not the most desirable thing, but one that most people can live with. <div>&nbsp;</div> But now try to do the same thing with the rename command. If we just apply the name changes to ODM and nothing else, we will no longer have a record of the device under its previous name. I'm not going to go into the details here, but this could result in major inconsistency within a device driver that could result in a system crash. Sure, if the reboot is immediately after the rename, probably not a problem. But if it isn't, and other config commands are run, there is too much risk. <div>&nbsp;</div> We have looked at other approaches where we don't make the change in ODM but make a record of the name change request that we would then perform on the subsequent reboot. This is a lot more complicated and not something we can commit to." <br />

8 cggibbo commented Permalink

Some more interesting feedback from AIX development. <div>&nbsp;</div> "Hi Chris, <div>&nbsp;</div> I've figured this out. I was doing my testing on a system with the latest 61 TL. The same source code for the rendev command was supposed to be used for both that and the 710 release. But I just found out that the developer dropped the wrong version into the 71 release. This is very bad! I'll be working to get it fixed fixed this afternoon, but I have no idea when it will be available. <div>&nbsp;</div> In a nutshell, the problem is that when a parent device is renamed, the ODM for child devices is not being updated to reflect the new parent name. In your case, you see this in the path to the child device. In other cases, it will affect the child device itself and result in a new child being created with a new name, just as you are seeing a new path get created when cfgmgr is run. If a device, even an adapter, has no children, it can be renamed without any problems. This problem is ONLY in the AIX 71 release. <div>&nbsp;</div> Feel free to share on your blog. <div>&nbsp;</div> Thanks!"

9 cfrs_au commented Permalink

All it really works for is say changing hdisk1 to hdisk2 which is totally pointless. <br /> If you try to do something useful with rendev it will break your whole environment. <div>&nbsp;</div> A few quick examples but there are tons more: <br /> 1) Renaming a vscsi client adapter to something with a '-' in it causes a kernel panic. Pretty funny. <br /> 2) LPM doesn't work if you rename vscsi host adapters. <br /> 3) viosbr doesn't work if you rename vscsi host adapters. <div>&nbsp;</div> 'Some devices may have special requirements on their names in order for other devices or applications to use them.' -&gt; read as 'DO NOT USE THIS COMMAND IT WILL BREAK YOUR SYSTEM.' <div>&nbsp;</div> <div>&nbsp;</div> <div>&nbsp;</div> <div>&nbsp;</div>

10 cggibbo commented Permalink

Thanks for your comment cfrs_au! LOL! Sorry to hear you've been burnt so badly by one little command! :) <div>&nbsp;</div> I assume you've reported all of this to IBM support....so they can fix the issues and make AIX even better? <div>&nbsp;</div> Renaming hdisks is pointless hey? Perhaps you've not worked in clustered environments before? <div>&nbsp;</div> The rendev command has already helped me in renaming hdisks across several VIOS. Save me much time and energy. <div>&nbsp;</div>

11 cfrs_au commented Permalink

Just a further warning for anyone thinking of creative uses for rendev :). <div>&nbsp;</div> The command's purpose and limitations should be better documented IMO. <div>&nbsp;</div> I guess the reason I try not to care about hdisk names is that before making a change I need to verify what the pvid/path/lun/size/type is for each hdisk, and the only reliable way to do that is to query the os/san array for current info so I have scripts to do all that (otherwise not much work would get done :p) <div>&nbsp;</div> Trusting hdisk device names to match seems a bit sketchy to me. My preference would be to use mknod whenever possible for stuff like RAC that requires identical device names.

12 frenger commented Permalink

Hi <br /> We have a lot of very old VIOS server and recently we thought it might be a good idea to rename the vtscsi device to somthing like lpar_hdisk# to get a better overview. On our newer Vios I manged to use rendev as root (there is sadly no chvdev command present) to rename the device. But it feels a bit risky as someone stated above viosbr can get troubles and so on. I don't feel relly save at the moment renaming all vtscsi devices on a VIO without long term testing. Then we realized that pre AIX 6.1 vios (1.4 &amp;1.5) not even has the rendev command... <br /> Is there a "legal" way to rename vtscsi devices in a supported way. I mean even when I do it as root on more recent VIOS versions, its not really supported because it fells like circumventing the VIO ODMs. Any idea on this? <br /> Regards Sebastian

13 cggibbo commented Permalink

"Is there a "legal" way to rename vtscsi devices in a supported way?" <div>&nbsp;</div> Well I guess one, supported, method would be to remove the device mapping and then re-map the device with the new/desired vtscsi name. <div>&nbsp;</div>

14 frenger commented Permalink

I would have preffered a command that can change a device that is currently in use. Theoretically you can remove all devices from one VIO because in a dual VIO concept nothing should go wrong. Praktically its a whole differnt story. Oracle DBs for example are very sensiteve with the IO of their control-files. We had some default mpio path switches on Oracle DBs and in 90% of the time the DB coredumped. Besides this fact it's to complicated and work intensive to remove and reconfigure all devices of our VIO LPARS. So I have to hope a VIOS release in the Future deals with this "issue". <br /> Thanks, Sebastian

15 cggibbo commented Permalink

"I would have preffered a command that can change a device that is currently in use..................So I have to hope a VIOS release in the Future deals with this "issue" <div>&nbsp;</div> Couldn't agree more. <div>&nbsp;</div> One way to request this enhancement would be to contact your IBM rep and suggest a Design Change Request (DCR). <div>&nbsp;</div> Cheers. <div>&nbsp;</div> Chris

Add a Comment Add a Comment