Summary: IPMP on Boot comes up with Failed Interfaces (Sol9)...

From: Ryan Krenzischek <ryan_at_vaxpower.org>
Date: Thu Mar 17 2005 - 19:02:41 EST
Thanks to all who replied:

 	Michael Imrick (Provided documentation found after summary)
 	Jon Hudson
 	Darren Dunham
 	Paul Sagneri
 	Gary Chambers
 	Dirk Hamilton

[..much cut..] (scroll down for the whole doc)
Summary: Your default gateway must accept ICMP echo/echo-reply.

IPMP only use the defaultrouter as probe target. Each test interface of 
the IPMP group send ICMP requests only to the defaultrouter. To get the 
configuration, IPMP looks to the routing table and is independent from 
/etc/defaultrouter file. NO "all hosts" multicast (224.0.0.1) will be 
sent.

Advantages:
- Easiest configuration for IPMP.

Disadvantages:
- When the defaultrouter is down IPMP failover does not work anymore.
   The in.mpathd does NOT send out multicasts to get other probe
   targets, therefore all interfaces in the IPMP group get the state
   "failed". You can ignore this bug/feature when you have a
   defaultrouter which is 100% online! Please look to RFE 4431511 and
   4489960 for further information.
- If you have a lot of IPMP groups, the defaultrouter has to reply to
   a lot of ICMP requests. Take care of defaultrouter. Do not overload
   the defaultrouter.
[..lots more cut here..]

IPMP is now happily working.

All the way at the bottom of the message is my original post.  The 
document which pointed out my problem is attached for your review (sorry 
the document did not honor 80 columns--for those of us still reading our 
email via xterm):

Keyword(s):IPMP

Description:


Document Body:

Contents:

1.  Production and test interfaces in the same IP subnet
     1.1  With defaultrouter
     1.2  Without defaultrouter
     1.3  With dedicated hosts acting as test targets with "host-routes"
     1.4  Configuration examples for 1.1, 1.2 and 1.3

2.  Production and test interfaces in different IP subnets but the same
     physical network
     2.1  With defaultrouter in production subnet and test subnet
     2.2  With defaultrouter in production subnet but without
defaultrouter in
test subnet
     2.3  With dedicated hosts acting as test targets with "host-routes"
     2.4  Configuration examples for 2.1, 2.2 and 2.3

A.  Start script for adding static "host routes" permanently
     /etc/rc2.d/S70staticroutes

B.  Summary

Note: This document does not cover the basic knowledge of IP
Multipathing. If
you would need it, please check it in http://docs.sun.com/?q=IPMP

All IPMP patches mentioned below should be installed.


 		Sparc 	Intel 	Comments
OS 
Release

7 or earlier 	N/S 	N/S 	  IPMP was introduced with Solaris[TM] 8
Update 2
 				  for Solaris 8 FCS the Patches 108528,
 				  109898, 109900 are required


8 		108528   108529   Solaris Kernel Patch (in.mpathd, ip,
icmp, ifconfig)
 		109898   109899   ARP Patch
 		109900	 109901	  Startup Patch (/etc/init.d/network and
/sbin/ifparse)
 		109902   109903   Offlining capability (Update 3
 		110378   110396   Mipagent patch needed for offlining
capability
 		111802   111803   Introduce RCM suspend/resume (Update
6)
 		110839 	 110855	  Introduce RCM IPMP module (Update 6)




9 		112902		  (if rev >= 08 then all the patches
below are required:)
 		112911		  ifconfig (see SunAlert 57435)(1)
 		112912		  libinetcfg.so
 		113964    N/P     mpathd,routed
 		112854		  icmp
 		113464		  header files


Key
-----
N/S - Product not supported on this OS 
N/P - No patch applicable for this OS

Legend:
-------
IPMP    system with IPMP group
T<n>    target host <n>
p       network/host part of IP address of interface in production
subnet
t       network/host part of IP address of interface in test subnet
---     data link subnet (i.e broadcast domain) with one IP subnet 
===     data link subnet (i.e broadcast domain) with two (or more) IP
subnets



Good to know:
-------------
The operation of IP Multipathing (in.mpathd) depends on the routing 
configuration. Therefore in.mpathd always refers to the routing-table 
(IRE-cache) to distinguish which test partner(s) are going to be used.
Test partners are required for deciding if the interface is working
properly.

in.mpathd by default chooses the defaultrouter as single test-target
for probing. If no defaultrouter exists for the test-interface ip
address,
arbitrary hosts on the link are detected by sending out "all hosts"
multicast packets (224.0.0.1) on the wire to detect its test-partners. 
An "all routers" multicasts (224.0.0.2) will never be sent! The first
five 
hosts that are responding to the echo packets are chosen as targets
for probing. In such a non-defaultrouter environment, the in.mpathd 
always tries to find five probe targets via an "all hosts" multicast.

The in.mpathd detects failures and repairs by sending out 'icmp-echo' 
probes (like pinging) from all interfaces that are part of the IPMP
group. 
If there are no replies to five consecutive probes, the interface is 
considered to have failed and performs a failover of the network access
to another interface in the IPMP group. The probing rate depends on the 
failure detection time which is defined in /etc/default/mpathd. By
default,
failure detection time is 10 seconds. Thus the five probes will be sent
within the failure detection time.



1.  Production and test interfaces in the same IP subnet
======================================================

1.1 With defaultrouter
----------------------

                                        +-------------+
                                        |defaultrouter|
                                        +------o------+
                                               | p=t:172.20.20.1
                                               |
                                               |
          ----------+-+------------------------+-------------------------
                    | |                                  p=t:172.20.20/24
    p:172.20.20.10  | |
    t:172.20.20.210 | | t:172.20.20.220
                 +---o-o---+
                 |   IPMP  |
                 +---------+

IPMP only use the defaultrouter as probe target. Each test interface of
the
IPMP group send ICMP requests only to the defaultrouter. To get the
configuration, IPMP looks to the routing table and is independent from
/etc/defaultrouter file. NO "all hosts" multicast (224.0.0.1) will be
sent.

Advantages:
- Easiest configuration for IPMP.

Disadvantages: 
- When the defaultrouter is down IPMP failover does not work anymore.
   The in.mpathd does NOT send out multicasts to get other probe
   targets, therefore all interfaces in the IPMP group get the state
   "failed". You can ignore this bug/feature when you have a
   defaultrouter which is 100% online! Please look to RFE 4431511 and
   4489960 for further information.
- If you have a lot of IPMP groups, the defaultrouter has to reply to
   a lot of ICMP requests. Take care of defaultrouter. Do not overload
   the defaultrouter.
- The defaultrouter has to reliably answer ICMP echo requests. (e.g.
   firewalls sometimes do not)




1.2 Without defaultrouter
-------------------------

                             +---------+           +---------+
                             |    T1   |   ......  |    T5   |
                             +----o----+           +----o----+
                                  | p=t:172.20.20.110   |
p=t:172.20.20.150
                                  |                     |
                                  |                     |
      -----------+-+--------------+---------------------+-------
                 | |                            p=t:172.20.20/24
p:172.20.20.10  | |
t:172.20.20.210 | | t:172.20.20.220

             +---o-o---+
             |   IPMP  |
             +---------+

IPMP dynamically determines five arbitrary hosts on the link via "all
hosts" multicast address (224.0.0.1). At the least, you need one probe
target that IPMP will work. Beware that one probe target is not reliable
enough. If there are less than five targets available the in.mpathd sent
out the "all hosts" multicasts to get a complete list of five probe
targets.

Advantages:
- easiest configuration for IPMP in a subnet without a defaultrouter.
- is very reliable due to the five targets.


Disadvantages: 
- a subnet without an defaultrouter is very rare.




1.3 With dedicated hosts acting as test targets with "host-routes" 
----------------------------------------------------------------

             +-------------+  +---------+           +---------+
             |defaultrouter|  |    T1   |   ......  |    T5   |
             +------o------+  +----o----+           +----o----+
                    !              | p=t:172.20.20.110   |
p=t:172.20.20.150
                    !              |                     |
                    !              |                     |
     ------------+-+---------------+---------------------+-------
                 | |                              p=t:172.20.20/24
p:172.20.20.10  | |
t:172.20.20.210 | | t:172.20.20.220
              +---o-o---+
              |   IPMP  |
              +---------+

Some "host routes" will be defined with a startscript in
/etc/rc2.d/S70staticroutes.  (The script is attached to this document.)
When IPMP refer to the routing table it will choose the first five
defined
"host routes" as probe targets. This is due to the fact that normally
the
"host routes" are before the defaultrouter in the routing table. If you
have less than five "host routes" also the defaultrouter (when
available)
will be used as probe target as well.

Example:
a) Configuration with host1, host2 ... hostN (with N=5 or N>5),
defaultrouter :
==> The first five hosts (host1 ... host5) will be defined as target.

b) Configuration with less than 5 hosts : for instance, host1, host2,
defaultrouter :
==> The three systems (host1, host2, defaultrouter) will be defined as
target.

Also in this case the in.mpathd tries to get five probe targets all
the time from the routing table. Remember in this configuration the
in.mpathd does NOT send "all hosts" multicasts!


Advantages:
- The defaultrouter is not important for the IPMP configuration because
   if the defaultrouter is not available you have still some "host
routes"
   for probing. 
- IPMP is always high available due to independency to the defaultrouter

Disadvantages:
- More administrative work to do. 
- Due to static configuration you should check that some of the probe
targets   are always available.
- Bug#4685978 IPMP does not detect NIC repair when only one of the two
   targets is up. Should not happen if you have more than 2 target hosts.
   Fixed in Solaris 9 HW8/03 Update4.




1.4 Configuration examples for 1.1, 1.2 and 1.3
-------------------------------------------------

/etc/hosts
----------
127.0.0.1        localhost 
172.20.20.10     host10       loghost
172.20.20.210    host10-test-qfe0
172.20.20.220    host10-test-qfe4

/etc/hostname.qfe0
------------------
host10 netmask + broadcast + group ipmp0 up \
addif host10-test-qfe0 deprecated -failover netmask + broadcast + up

/etc/hostname.qfe4
------------------
host10-test-qfe4 deprecated -failover netmask + broadcast + group ipmp0
up

ifconfig output:
----------------
qfe0: flags=9040843<UP,BROADCAST,RUNNING,MULTICAST,IPv4> mtu 1500 index
3
         inet 172.20.20.10 netmask ffffff00 broadcast 172.20.20.255
         groupname ipmp0
         ether 8:0:20:e8:88:dc 
qfe0:1:
flags=1000843<UP,BROADCAST,RUNNING,MULTICAST,DEPRECATED,IPv4,NOFAILOVER>
mtu 1500 index 3
         inet 172.20.20.210 netmask ffffff00 broadcast 172.20.20.255
qfe4:
flags=9040843<UP,BROADCAST,RUNNING,MULTICAST,DEPRECATED,IPv4,NOFAILOVER>
mtu 1500 index 4
         inet 172.20.20.220 netmask ffffff00 broadcast 172.20.20.255
         groupname ipmp0
         ether 8:0:20:e8:89:34





2.  Production and test interfaces in different IP subnets
==========================================================

If you have not enough additional ip-addresses on hand for setting up 
IPMP, you can configure the ipmp-test-interfaces in a different 
ip-network than your production network (e.g. 192.168., 10. ..). But you

must make sure that there are enough test-partners (also in the new 
test-network) who are responding to the ipmp-test-interfaces. You may
also configure a defaultrouter in the new test-network in case you have 
an existing 100.1% reliable test-partner which should act as a single 
test-partner. In such a configuration in.mpathd will only use its 
test-subnet IP addresses as source address for outgoing probe packets.

Note: The in.mpathd only looks to the test subnet. Therefore if you
       have no IP addresses available in the test subnet the IPMP
       group will fail although if the production subnet is available.




2.1 With defaultrouter in production subnet and test subnet
-----------------------------------------------------------

                                        +-------------+
                                        |defaultrouter|
                                        +------o------+
                                               | p: 172.20.20.1
                                               | t: 192.168.1.1
                                               |
                                               |           t:
192.168.1/24

===========+=+========================+==========================
                    | |                                    p:
172.20.20/24
   p: 172.20.20.10  | |
   t: 192.168.1.210 | | t: 192.168.1.220
                +---o-o---+
                |   IPMP  |
                +---------+

IPMP only use the defaultrouter as probe target. Each test interface of
the
IPMP group send ICMP requests only to the defaultrouter. To get the
configuration IPMP looks to the routing table and is independent from
/etc/defaultrouter file. NO "all hosts" multicast (224.0.0.1) will be
sent.

Advantages:
- Test interfaces don't need IP addresses of the production subnet

Disadvantages:
- The interface of the defaultrouter has to reside in both the
   production AND test subnet.
- Exceptional configuration of defaultrouter.
- All which are mentioned in section 1.1




2.2 With defaultrouter in production subnet net but without
defaultrouter
in test subnet 
------------------------------------------------------------------------
---------
-------

         +-------------+      +---------+       +---------+
         |defaultrouter|      |    T1   |  ...  |    T5   |
         +------o------+      +----o----+       +----o----+
                | p:172.20.20.1    | p:172.20.20.110 | p:172.20.20.150
                |                  | t:192.168.1.110 | t:192.168.1.150
                |                  |                 |
                |                  |                 |
t:192.168.1/24

============+=+=+==============+=================+=====================
                  | |
p:172.20.20/24
p:172.20.20.10   | |
t:192.168.1.210  | | t:192.168.1.220
              +---o-o---+
              |   IPMP  |
              +---------+

IPMP dynamically determines five arbitrary hosts in the test subnet via
"all hosts" multicast  address (224.0.0.1). At least you need one probe
target that IPMP will work. Beware that one probe target is not reliable
enough. If there are less than five targets available the in.mpathd sent
out the "all hosts" multicasts to get a complete list of five probe
targets.

Advantages:
- easiest configuration for IPMP if you have too less IP addresses
   available in the production subnet.
- is very reliable due to the five targets.

Disadvantages:
- the probe targets must be available before you can setup the
   IPMP host.
- more administrative work because you have to setup some probe
   targets with an additional interface in the test subnet.
   (Recommendation: Also use IPMP on the target hosts. Then
   you are save that the IP address for the test subnet on the
   target host are always available after an reboot. It's enough
   to have only one interface for IPMP.)




2.3 with dedicated hosts acting as test targets with "host-routes" 
-----------------------------------------------------------------

        +-------------+       +---------+        +---------+
        |defaultrouter|       |    T1   |  ....  |    T5   |
        +------o------+       +----o----+        +----o----+
               !                   | p:172.20.20.110 | p:172.20.20.150
               !                   | t:192.168.1.110 | t:192.168.1.150
               !                   |                 |
               !                   |                 |
t:192.168.1/24

=============+=+===============+=================+=====================
                 | |
p:172.20.20/24
p:172.20.20.10  | |
t:192.168.1.210 | | t:192.168.1.220
             +---o-o---+
             |   IPMP  |
             +---------+

Some "host routes" will be defined in the test subnet with a startscript
in
/etc/rc2.d/S70staticroutes. (The script is attached to this document.)
When
IPMP refer to the routing table it will choose the first five defined
"host
routes" as probe targets in the test subnet. This is due to the fact
that
normally the "host routes" are before the defaultrouter in the routing
table. If you have less than five "host routes" also the defaultrouter
(when available in the test subnet) will be used as probe target as
well.


Example:
- please look to the example of section 1.3

Advantages:
- test interfaces don't need IP addresses of the production subnet
- all which are mentioned in section 1.3

Disadvantages:
- all which are mentioned in section 1.3
- all which are mentioned in section 2.2



2.4 Configuration examples for 2.1, 2.2 and 2.3
------------------------------------------------

/etc/hosts
----------
127.0.0.1        localhost 
172.20.20.10     host10       loghost
192.168.1.210    host10-test-qfe0
192.168.1.220    host10-test-qfe4

/etc/hostname.qfe0
------------------
host10 netmask + broadcast + group ipmp0 up \
addif host10-test-qfe0 deprecated -failover netmask + broadcast + up

/etc/hostname.qfe4
------------------
host10-test-qfe4 deprecated -failover netmask + broadcast + group ipmp0
up

ifconfig output:
----------------
qfe0: flags=9040843<UP,BROADCAST,RUNNING,MULTICAST,IPv4> mtu 1500 index
3
         inet 172.20.20.10 netmask ffffff00 broadcast 172.20.20.255
         groupname ipmp0
         ether 8:0:20:e8:88:dc 
qfe0:1:
flags=1000843<UP,BROADCAST,RUNNING,MULTICAST,DEPRECATED,IPv4,NOFAILOVER>
mtu 1500 index 3
         inet 192.168.1.210 netmask ffffff00 broadcast 172.20.20.255
qfe4:
flags=9040843<UP,BROADCAST,RUNNING,MULTICAST,DEPRECATED,IPv4,NOFAILOVER>
mtu 1500 index 4
         inet 192.168.1.220 netmask ffffff00 broadcast 172.20.20.255
         groupname ipmp0
         ether 8:0:20:e8:89:34


A.
----------- Begin of start script /etc/rc2.d/S70staticroutes
--------------

#!/sbin/sh
# /etc/rc2.d/S70staticroutes /etc/init.d/staticroutes
# Copyright (c) 2003 by Sun Microsystems, Inc.
# All rights reserved.
#
#ident  "@(#)staticroutes      1.0.1 
#
# Edit the following IPMP test  TARGETS to suit your needs.
# To install:
# 1) cp S70staticroutes to /etc/rc2.d
# 2) perform edits on the script as required
# 3) chmod 744 /etc/rc2.d/S70staticroutes
# 4) chown root:sys /etc/rc2.d/S70staticroutes
# 5) ln /etc/rc2.d/S70staticroutes /etc/init.d/staticroutes
#
TARGET1=172.20.20.110
TARGET2=172.20.20.120
TARGET3=172.20.20.130
TARGET4=172.20.20.140
TARGET5=172.20.20.150
TARGET6=172.20.20.160
TARGET7=172.20.20.170
TARGET8=172.20.20.180
TARGET9=172.20.20.190

case "$1" in
         'start')
                 /usr/bin/echo "Adding static routes for IPMP ..."
                 /usr/sbin/route add host $TARGET1 $TARGET1
                 /usr/sbin/route add host $TARGET2 $TARGET2
                 /usr/sbin/route add host $TARGET3 $TARGET3
                 /usr/sbin/route add host $TARGET4 $TARGET4
                 /usr/sbin/route add host $TARGET5 $TARGET5
                 /usr/sbin/route add host $TARGET6 $TARGET6
                 /usr/sbin/route add host $TARGET7 $TARGET7
                 /usr/sbin/route add host $TARGET8 $TARGET8
                 /usr/sbin/route add host $TARGET9 $TARGET9
                 ;;
         'stop')
                 /usr/bin/echo "Deleting static routes for IPMP ..."
                 /usr/sbin/route delete host $TARGET1 $TARGET1
                 /usr/sbin/route delete host $TARGET2 $TARGET2
                 /usr/sbin/route delete host $TARGET3 $TARGET3
                 /usr/sbin/route delete host $TARGET4 $TARGET4
                 /usr/sbin/route delete host $TARGET5 $TARGET5
                 /usr/sbin/route delete host $TARGET6 $TARGET6
                 /usr/sbin/route delete host $TARGET7 $TARGET7
                 /usr/sbin/route delete host $TARGET8 $TARGET8
                 /usr/sbin/route delete host $TARGET9 $TARGET9
                 ;;
esac

----------- End of start script /etc/rc2.d/S70staticroutes
--------------



B. Summary

It's not easy to give a general recommendation because it depends on the
network infrastructure which the customers have. Therefore you have to
discuss the various possibilities with your customer. Maybe the most
used
setups are 1.1 and 2.2.

---------- Original message ----------
Date: Thu, 17 Mar 2005 15:26:12 -0500 (EST)
From: Ryan Krenzischek
To: sunmanagers@sunmanagers.org
Subject: IPMP on Boot comes up with Failed Interfaces (Sol9)...

Hi All,

I seem to be having a problem.  I have configured IPMP per the blueprint dated 
November 2002 on a V120 running Solaris 9.

The interfaces come up failed on boot.  I can un-fail the interface then see 
the network.  Any idea why I am seeing this behavior?

...For obvious reasons, I have masked the IP, Broadcast, and MAC (except the 
last part)

/etc/hostname.eri0
box1-eth0 netmask + broadcast + failover up \
group production \
addif box1-eth0test netmask + broadcast + \
deprecated -failover up

/etc/hostname.eri1
box1-eth1 netmask + broadcast + failover up \
group production \
addif box1-eth1test netmask + broadcast + \
deprecated -failover up

Output from ifconfig -a
lo0: flags=1000849<UP,LOOPBACK,RUNNING,MULTICAST,IPv4> mtu 8232 index 1
         inet 127.0.0.1 netmask ff000000
eri0: flags=11000843<UP,BROADCAST,RUNNING,MULTICAST,IPv4,FAILED> mtu 1500 index 
2
         inet 1.1.1.70 netmask ffffff00 broadcast 1.1.1.255
         groupname production
         ether X:X:XX:XX:XX:5
eri0:1: 
flags=19040843<UP,BROADCAST,RUNNING,MULTICAST,DEPRECATED,IPv4,NOFAILOVER,FAILED> 
mtu 1500 index 2
         inet 1.1.1.71 netmask ffffff00 broadcast 1.1.1.255
eri1: flags=11000803<UP,BROADCAST,MULTICAST,IPv4,FAILED> mtu 1500 index 3
         inet 1.1.1.72 netmask ffffff00 broadcast 1.1.1.255
         groupname production
         ether X:X:XX:XX:XX:6
eri1:1: 
flags=19040803<UP,BROADCAST,MULTICAST,DEPRECATED,IPv4,NOFAILOVER,FAILED> mtu 
1500 index 3
         inet 1.1.1.73 netmask ffffff00 broadcast 1.1.1.255

and /etc/hosts:

1.1.1.70   box1-eth0 #Primary intf
1.1.1.71   box1-eth0test # IPMP Test address for eth0
1.1.1.72   box1-eth1 # Second Address for ethernet controller
1.1.1.73   box1-eth1test # IPMP Test address for eth1

My interfaces do show different MAC addresses.  I have set eeprom 
local-mac-address?=true and rebooted the box.
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Thu Mar 17 19:02:52 2005

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:44 EST