iscsi breaks ipv6

[Ryan Kogelheide_1]

Hello. I've arrived at a customer site and they said that their installation of a NetApp on 11.23 broke all the ipv6 apps due to the installation of iscsi.

All I know is that now I can't use any X apps, which makes Oracle installs impossible. How do I start trying to resolve this problem? What do they mean by this?

[Steven E. Protter]

I'm confused. ipv6 is the new ip address system. I don't recall it having anything to do with X Windows.

Quick X check.

vi /etc/rc.config.d/xfs
# first variable needs to be one.

/sbin/init.d/xfs start

What is the DISPLAY variable.

See what happens. If you get an error report back please.

I think you and I both need more information to adequately deal with this issue.

If iscsi broke anything, the short term fix is to see about backing it out.

Steven

[Florian Heigl (new acc)]

Unfortunately my ipv6 box is 32bit only so there was no iscsi available when I wanted to test it, otherwise I could tell You if I experienced the same problem.

Probably You should simply open a call with HP to have them cross-check if this combination of addon-products is incompatible.

First You could check if any of the files that came with the TOUR bundle were replaced by the iSCSI driver.
In that case I'd ask them to reapply the TOUR2.x bundle and see if that helps.

[Florian Heigl (new acc)]

To cross-check the X-App problem.

It appears this company is very complete with
it's IPv6 setup, which is wonderful.

On the other hand the result is that a gethostbyname returns the IPv6 address first, then the IPv4.

So if You export DISPLAY=workstation1:0.0 the applications/X Libraries will lookup the ip address and use the IPv6 returned by the nameserver to access workstation1.

If I remember it correctly a resolver can be told to work in IPv4 mode which would be a nice temporary fix to keep things working.

You can cross-check to find out if this error is the cause by simply trying export DISPLAY=192.168.0.0:0.0 or whatever Your workstation IPv4 address is.

[Ryan Kogelheide_1]

Cool. Thanks very much for your responses.

I'll give you some examples of the errors I get with IP tools.

If I try to ftp to another host, I get
"ftp: socket: Not enough space".

Installing Oracles Management Agent, I get
"rresvport_af: socket: Not enough space : ".

running xclock with DISPLAY set to a numeric ip, I get
"Error: Can't open display: 196.68.23.45:0.0".

[marie-noelle jeanson_1]

Ryan,

Do they have the HP iscsi initiator software installed to work with Netapp? The hp-ux iscsi software initiator does not support ipv6. See the release note http://docs.hp.com/en/T1452-90009/index.html

Regards,

Marie-Noelle

[Florian Heigl (new acc)]

ryan, please see if a netstat -m also breaks, if yes, the iSCSI initiator broke something in their kernel.

I don't think this will be resolvable without hp's help, or someone else running ipv6 and a netapp.
if this is not a total testing environment You should really try booting vmunix.prev or better completely removing the iSCSI driver.
in between You could take an ignite backup and try reapplying TOUR after iSCSI.
If I happen to be very enthusiastic this weekend I could do some trying with a C3600 at my employers site - just they don't have a netapp and I'll end up doing nasty things with ipnat.

there also is an iSCSI initiator available from cisco, if they have a fairly good support contract, You might be able to access and try it.

still, if this is no testing environment, get rid of iSCSI, otherwise call in hp.

@marie-noelle - i think there a thin line between doesn't support ipv6 (cannot work over ipv6 transport) and breaks ipv6 transport. :)

-> off to bed

[Jeff Schussele]

HI,

It might be useful to check the ndd value of
tcp_conn_request_max
with iscsi I can certainly see a much higher likelyhood of having a ton of outstanding connection requests coming back from the netapp. 'Cause we all know disks *never* want much I/O, eh?
And IF you're not running a completely Gigabit path...well.....

My 2 cents,
Jeff

[Ryan Kogelheide_1]

Thanks again for thinking about this.

The guys upstairs got back to me that there is a call in with HP now and it has gone to level 3. They suspect some sort of resource contention and are installing something this weekend that collects statistics.

The netstat installed with 11.23 doesn't have a -n. Nonetheless, all the netstat options I tried seemed to work.

tcp_conn_request_max is 4096.

This is a 7x24 four nines system, and they don't have a test bed for trying your suggestions (I know, I know - it's the first thing I complained about).

I'm not sure who's iSCSI they are using. How do I find out the version of the drivers?

[Ryan Kogelheide_1]

Also, I hate to think that any of you are making any more effort than necessary for my sake. I just was hoping either somebody had run across this before or could at least give me some insight into what was happening. I'll update this on Monday and hopefully I can have more info on what is happening.

Thanks again.

[Angela Lycett]

I realize this is a forum for IT resource info but Ryan if you graduated from PGSS could you e-mail me at angela@happyplanet.com.