Re: oracle 10.2.0.4.3 RAC RHEL 5.4 x86_64 nic bonding + trunking (etherchannel)

  • From: Ujang Jaenudin <ujang.jaenudin@xxxxxxxxx>
  • To: Jakub Wartak <vnulllists@xxxxxxxxxxxx>
  • Date: Thu, 3 Jun 2010 19:09:09 +0700

hi jakub,

I have set MIITOL variable to non-existent-file, like this:
#MIITOOL=/sbin/mii-tool
MIITOOL=/sbin/mii-tool-non

but got this when one switch turn off (we do HA test)

rac 10.2.0.4.4
bond0 is bonding for virtual ip, switch is trunked
bond1 for heartbeat.


2010-05-29 06:33:32.803: [ RACG][51317200]
[5674][51317200][ora.radpadb2.vip]: clsrcep:evm post return 1
2010-06-03 18:19:23.210: [ RACG][51317200]
[27658][51317200][ora.radpadb2.vip]: ping to 10.33.200.1 via bond0 failed,
rc = 1 (host=radpadb1)
ping to 10.33.200.1 via bond0 failed, rc = 1 (host=radpadb1)
Interface bond0 checked failed (host=radpadb1)
Invalid parameters, or failed to bring up VIP (host=radpadb1)

so, even node1 still alive, but from client/app server cannot connect to it.
node2 was restarted, due to share switch for public and heartbeat :)
this is blade world...we had no flexible networking on it :(



-- 
thanks and regards
ujang jaenudin
jakarta - indonesia


On Thu, Apr 29, 2010 at 1:23 AM, Jakub Wartak <vnulllists@xxxxxxxxxxxx>wrote:

> Dnia środa, 28 kwietnia 2010 13:30, Ujang Jaenudin napisał:
> > dear lists,
> >
> > anyone have deployed rac 10.2.0.4.3 on rhel 5.4 x86_64 which using NIC
> > bonding + trunking + jumbo frames?
> > is there any issue, which bugs / patches should I have to consider?
> >
> > I plan to deploy RAC and should reach 2000+ tps
> > and should minimize single point of failure...
>
> Not sure about 10.2.0.4, but 11.1.0.6/11.2 still uses mii-tool in racgvip
> (VIP
> failover). Notes from racgvip/11.1.0.6 script:
>
> " - if mii-tool reports the link is down, use ifconfig to verify the link
>   status.
>  - use mii-tool to check if the interface is up. If mii-tool does not exist
>   or fails, "ping -r -I <interface> <default gw>" will be used instead."
>
> but mii-tool executed on bond interfaces seems to always to return "10
> Mbit,
> half duplex, link ok", not to mention that mii-tool is marked deprecated in
> it's own manpage (checked OEL 5.5 a few seconds ago).
>
> BTW: There is "Bug 8333262: VIP NOT FAILING OVER WHEN CABLE UNPLUGGED" with
> workaround "do not pull cable" :))))))))
>
> You can fallback to old code by setting MIITOOL variable
> to /some/not/existent/file.
>
> --
> Jakub Wartak
> http://jakub.wartak.pl/blog
>

Other related posts: