733298 – cman and corosync RRP handling are not consistent and needs improvements

Bug 733298 - cman and corosync RRP handling are not consistent and needs improvements

Summary: cman and corosync RRP handling are not consistent and needs improvements

Keywords:
Status:	CLOSED ERRATA
Alias:	None
Product:	Red Hat Enterprise Linux 6
Classification:	Red Hat
Component:	cluster
Sub Component:
Version:	6.3
Hardware:	Unspecified
OS:	Unspecified
Priority:	high
Severity:	high
Target Milestone:	rc
Target Release:	6.3
Assignee:	Fabio Massimo Di Nitto
QA Contact:	Cluster QE
Docs Contact:
URL:
Whiteboard:
Depends On:	722469 735911 738390
Blocks:	756082 758821 758823
TreeView+	depends on / blocked

Reported:	2011-08-25 11:55 UTC by Fabio Massimo Di Nitto
Modified:	2018-11-29 22:01 UTC (History)
CC List:	12 users (show)
Fixed In Version:	cluster-3.0.12.1-28.el6
Doc Type:	Bug Fix
Doc Text:	Cause: cman was not able to configure Redundant Ring Protocol (RRP) correctly in corosync. Consequence: RRP deployments would not work properly. Fix: cman has been improved to both configure RRP properly and perform extra sanity checks on user configurations. Result: it is now easier to deploy cluster with RRP and error reports are more extensive.
Clone Of:
Clones:	758821 758823 (view as bug list)
Environment:
Last Closed:	2012-06-20 13:54:50 UTC

Attachments	(Terms of Use)
Add an attachment (proposed patch, testcase, etc.)

Links
System	ID	Priority	Status	Summary	Last Updated
Red Hat Product Errata	RHBA-2012:0861	normal	SHIPPED_LIVE	cluster and gfs2-utils bug fix and enhancement update	2013-05-30 19:58:20 UTC

Description Fabio Massimo Di Nitto 2011-08-25 11:55:49 UTC

RRP is still TP in 6.2 with planned full support for 6.3.

There are several bits and pieces that need to be integrated correctly:

1) automatic multiple multicast address´ (one per ring)
2) ipv4 vs ipv6 (one ring on one protocol the other on another?)
3) cman configuration for RRP is at best annoying (mcast address is not global but per node)
4) cman needs to improve error handling for invalid RRP configs (more than 2 rings for example)

Since the RRP code in cman has been barely tested, we will need some work there to make sure it is robust and easy to use (and I need a bug to do the work and commit it)

Comment 1 Steven Dake 2011-09-14 18:59:06 UTC

#2 wont work and seems like something we couldn't effectively support.

Comment 2 Fabio Massimo Di Nitto 2011-09-14 19:04:46 UTC

(In reply to comment #1)
> #2 wont work and seems like something we couldn't effectively support.

that is exactly why we need to track combinations. This BZ is not to increment what we support, but to make sure cman does the correct configuration checks based on what we want to support and keep features in sync between corosync and cman config.

Comment 3 Fabio Massimo Di Nitto 2011-11-30 12:56:17 UTC

https://www.redhat.com/archives/cluster-devel/2011-November/msg00136.html

On the assumption that the patch should introduce no regressions in standard default config (tested, but not reported here):

1) don't allow more than 2 rings config

  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2"/>
      <altname name="clusternet-node1-eth3"/>
    </clusternode>

[root@clusternet-node1 ~]# ccs_config_validate
Configuration of more than 2 rings is not supported
Unable to get the configuration

(forcefully disable ccs_config_validate)

[root@clusternet-node1 ~]# cman_tool -d join -DNONE
Starting /usr/sbin/corosync corosync -f
CMAN_DEBUG=255
COROSYNC_DEFAULT_CONFIG_IFACE=xmlconfig:cmanpreconfig:openaisserviceenablestable
CMAN_PIPE=4
corosync [MAIN  ] Corosync Cluster Engine ('1.4.2'): started and ready to provide service.
corosync [MAIN  ] Corosync built-in features: nss dbus rdma snmp
corosync [MAIN  ] Successfully read config from /etc/cluster/cluster.conf
corosync [MAIN  ] Configuration of more than 2 rings is not supported
corosync [MAIN  ] Corosync Cluster Engine exiting with status 8 at main.c:1659.
forked process ID is 4634
Configuration of more than 2 rings is not supported
cman_tool: corosync daemon didn't start

2) enforce either different mcast or different ports on each ring to avoid conflicts.

NOTE corosync uses specified port and port-1. In a 2 rings, none of the 4 ports should ever conflict in case same mcast address is in use.

(default enforces 2 different mcast addresses, same port)

2a) same mcast / same port

  <cman port="666">
   <multicast addr="239.192.99.74"/>
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2" mcast="239.192.99.74" port="666"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Alternate communication channel (mcast: 239.192.99.74 ports: 665,664) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 666,665)
Unable to get the configuration

[root@clusternet-node2 ~]# cman_tool join -DNONE
corosync [MAIN  ] Corosync Cluster Engine ('1.4.2'): started and ready to provide service.
corosync [MAIN  ] Corosync built-in features: nss dbus rdma snmp
corosync [MAIN  ] Successfully read config from /etc/cluster/cluster.conf
corosync [MAIN  ] Alternate communication channel (mcast: 239.192.99.74 ports: 666,665) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 666,665)
corosync [MAIN  ] Corosync Cluster Engine exiting with status 8 at main.c:1659.
Alternate communication channel (mcast: 239.192.99.74 ports: 666,665) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 666,665)
cman_tool: corosync daemon didn't start

2b) same mcast / ports overlap

  <cman port="666">
   <multicast addr="239.192.99.74"/>
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2" mcast="239.192.99.74" port="665"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Alternate communication channel (mcast: 239.192.99.74 ports: 665,664) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 666,665)
Unable to get the configuration

[root@clusternet-node2 ~]# cman_tool join -DNONE
corosync [MAIN  ] Corosync Cluster Engine ('1.4.2'): started and ready to provide service.
corosync [MAIN  ] Corosync built-in features: nss dbus rdma snmp
corosync [MAIN  ] Successfully read config from /etc/cluster/cluster.conf
corosync [MAIN  ] Alternate communication channel (mcast: 239.192.99.74 ports: 665,664) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 666,665)
corosync [MAIN  ] Corosync Cluster Engine exiting with status 8 at main.c:1659.
Alternate communication channel (mcast: 239.192.99.74 ports: 665,664) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 666,665)
cman_tool: corosync daemon didn't start

2c) same as 2b but backwards

  <cman port="665">
   <multicast addr="239.192.99.74"/>
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2" mcast="239.192.99.74" port="666"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Alternate communication channel (mcast: 239.192.99.74 ports: 666,665) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 665,664)
Unable to get the configuration

[root@clusternet-node2 ~]# cman_tool join -DNONE
corosync [MAIN  ] Corosync Cluster Engine ('1.4.2'): started and ready to provide service.
corosync [MAIN  ] Corosync built-in features: nss dbus rdma snmp
corosync [MAIN  ] Successfully read config from /etc/cluster/cluster.conf
corosync [MAIN  ] Alternate communication channel (mcast: 239.192.99.74 ports: 666,665) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 665,664)
corosync [MAIN  ] Corosync Cluster Engine exiting with status 8 at main.c:1659.
Alternate communication channel (mcast: 239.192.99.74 ports: 666,665) cannot use
same address and ports of primary channel (mcast: 239.192.99.74 ports: 665,664)
cman_tool: corosync daemon didn't start

2d) different mcast

  <cman port="666">
   <multicast addr="239.192.99.74"/>
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2" mcast="239.192.99.75" port="666"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Configuration validates

cman starts fine

2e) different ports

  <cman port="666">
   <multicast addr="239.192.99.74"/>
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2" mcast="239.192.99.74" port="777"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Configuration validates

cman starts fine

3) both rings must be on the same proto (v4 / v6)

setup both rings to support v4 and v6. For simplicity:

3ffe::1     clusternet-node1-eth2-v6
3ffe::2     clusternet-node2-eth2-v6
4000::1     clusternet-node1-eth1-v6
4000::2     clusternet-node2-eth1-v6

3a) both rings on v4:

<cluster name="fabbione" config_version="3" >
  <logging debug="on"/>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Configuration validates

[root@clusternet-node2 daemon]# cman_tool status
Version: 6.2.0
Config Version: 3
Cluster Name: fabbione
Cluster Id: 25573
Cluster Member: Yes
Cluster Generation: 88
Membership state: Cluster-Member
Nodes: 2
Expected votes: 2
Total votes: 2
Node votes: 1
Quorum: 2
Active subsystems: 1
Flags:
Ports Bound: 0
Node name: clusternet-node2-eth1
Node ID: 2
Multicast addresses: 239.192.99.73 239.192.99.74
Node addresses: 192.168.4.2 192.168.5.2

3b) ring 0 is on v6 ring 1 on v4

<cluster name="fabbione" config_version="3" >
  <logging debug="on"/>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1-v6" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Node address family does not match multicast address family
Unable to get the configuration

[root@clusternet-node2 ~]# cman_tool join
Node address family does not match multicast address family
Unable to get the configuration
cman_tool: Not joining, configuration is not valid

3c) ring 0 is on v4 ring 1 on v6

<cluster name="fabbione" config_version="3" >
  <logging debug="on"/>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2-v6"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Node address family does not match multicast address family
Unable to get the configuration

[root@clusternet-node2 ~]# cman_tool join
Node address family does not match multicast address family
Unable to get the configuration
cman_tool: Not joining, configuration is not valid

3d) both rings are on v6

<cluster name="fabbione" config_version="3" >
  <logging debug="on"/>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1-v6" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2-v6"/>
    </clusternode>

[root@clusternet-node2 ~]# ccs_config_validate
Configuration validates

[root@clusternet-node2 daemon]# cman_tool status
Version: 6.2.0
Config Version: 3
Cluster Name: fabbione
Cluster Id: 25573
Cluster Member: Yes
Cluster Generation: 8
Membership state: Cluster-Member
Nodes: 2
Expected votes: 2
Total votes: 2
Node votes: 1
Quorum: 2
Active subsystems: 1
Flags:
Ports Bound: 0
Node name: clusternet-node2-eth1-v6
Node ID: 2
Multicast addresses: ff15::63e5 ff15::63e6
Node addresses: 4000::2 3ffe::2

4) simplify alt config

current config for full custom alt mcast address and port is rather insane.
Let's take the extreme case where both cman port/ttl/mcast address,
alt port/ttl/mcast address needs to be customized.

  <cman port="666">
   <multicast addr="239.192.99.73" ttl="2"/>
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <altname name="clusternet-node1-eth2" mcast="239.192.99.88" port="888" ttl="3"/>
    </clusternode>
    <clusternode name="clusternet-node2-eth1" votes="1" nodeid="2">
      <altname name="clusternet-node2-eth2" mcast="239.192.99.88" port="888" ttl="3"/>
    </clusternode>

first issue:

  <cman port="666">
   <multicast addr="239.192.99.73" ttl="2"/>
  </cman>

disconnection between network data for cman

second issue:

  <altname name="clusternet-node2-eth2" mcast="239.192.99.88" port="888" ttl="3"/>

needs to be repeated for every single node in the cluster.

totem.interface.ringnumber=0
totem.interface.bindnetaddr=192.168.4.2
totem.interface.mcastaddr=239.192.99.73
totem.interface.mcastport=666
totem.interface.ttl=2
totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.mcastaddr=239.192.99.88
totem.interface.mcastport=888
totem.interface.ttl=3

after patch:

  <cman>
   <multicast addr="239.192.99.73" port="666" ttl="2"/>
   <altmulticast addr="239.192.99.88" port="888" ttl="3"/>
  </cman>

  <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
   <altname name="clusternet-node1-eth2"/>

  <clusternode name="clusternet-node2-eth1" votes="1" nodeid="2">
   <altname name="clusternet-node2-eth2"/>

totem.interface.ringnumber=0
totem.interface.bindnetaddr=192.168.4.2
totem.interface.mcastaddr=239.192.99.73
totem.interface.mcastport=666
totem.interface.ttl=2
totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.mcastaddr=239.192.99.88
totem.interface.mcastport=888
totem.interface.ttl=3

backward compatibility (old config model overrides new config model)

  <cman port="777">
   <multicast addr="239.192.99.73" port="666" ttl="2"/>
   <altmulticast addr="239.192.99.88" port="888" ttl="3"/>
  </cman>

cman port has higher priority vs multicast port

totem.interface.ringnumber=0
totem.interface.bindnetaddr=192.168.4.2
totem.interface.mcastaddr=239.192.99.73
totem.interface.mcastport=777

altname multicast has higher priority vs altmulticast

  <cman port="777">
   <multicast addr="239.192.99.73" port="666" ttl="2"/>
   <altmulticast addr="239.192.99.88" port="888" ttl="3"/>
  </cman>

    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <altname name="clusternet-node1-eth2" mcast="239.192.99.99"/>

totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.mcastaddr=239.192.99.99
totem.interface.mcastport=888
totem.interface.ttl=3

altname port has higher priority vs altmulticast port

  <cman port="777">
   <multicast addr="239.192.99.73" port="666" ttl="2"/>
   <altmulticast addr="239.192.99.88" port="888" ttl="3"/>
  </cman>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <altname name="clusternet-node1-eth2" mcast="239.192.99.99" port="999"/>

totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.mcastaddr=239.192.99.99
totem.interface.mcastport=999
totem.interface.ttl=3

altname ttl has higher priority vs altmulticast ttl

  <cman port="777">
   <multicast addr="239.192.99.73" port="666" ttl="2"/>
   <altmulticast addr="239.192.99.88" port="888" ttl="3"/>
  </cman>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <altname name="clusternet-node1-eth2" mcast="239.192.99.99" port="999" ttl="20"/>

totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.mcastaddr=239.192.99.99
totem.interface.mcastport=999
totem.interface.ttl=20

5) NOT RELEVANT FOR QE - fix broadcast default ports

broadcast enforces address 255.255.255.255 on all interfaces
and therefor needs 2 different set of ports.

ring0 gets DEFAULT_PORT
ring1 gets DEFAULT_PORT + 2

unless manual override is in place.


  <cman broadcast="yes">
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2"/>
    </clusternode>
    <clusternode name="clusternet-node2-eth1" votes="1" nodeid="2">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node2"/>
        </method>
      </fence>
      <altname name="clusternet-node2-eth2"/>
    </clusternode>
  </clusternodes>

totem.interface.ringnumber=0
totem.interface.bindnetaddr=192.168.4.2
totem.interface.broadcast=yes
totem.interface.mcastport=5405
totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.broadcast=yes
totem.interface.mcastport=5407

  <cman broadcast="yes" port="999">
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2"/>
    </clusternode>

totem.interface.ringnumber=0
totem.interface.bindnetaddr=192.168.4.2
totem.interface.broadcast=yes
totem.interface.mcastport=999
totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.broadcast=yes
totem.interface.mcastport=5407

  <cman broadcast="yes" port="999">
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2" port="1000"/>
    </clusternode>

[root@clusternet-node1 ~]# cman_tool join
Alternate communication channel (mcast: 255.255.255.255 ports: 1000,999) cannot use
same address and ports of primary channel (mcast: 255.255.255.255 ports: 999,998)
Unable to get the configuration
cman_tool: Not joining, configuration is not valid


  <cman broadcast="yes" port="999">
  </cman>
  <clusternodes>
    <clusternode name="clusternet-node1-eth1" votes="1" nodeid="1">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node1"/>
        </method>
      </fence>
      <altname name="clusternet-node1-eth2" port="1001"/>
    </clusternode>
    <clusternode name="clusternet-node2-eth1" votes="1" nodeid="2">
      <fence>
        <method name="single">
          <device name="xvm" domain="clusternet-node2"/>
        </method>
      </fence>
      <altname name="clusternet-node2-eth2" port="1001"/>
    </clusternode>
  </clusternodes>

totem.interface.ringnumber=0
totem.interface.bindnetaddr=192.168.4.2
totem.interface.broadcast=yes
totem.interface.mcastport=999
totem.interface.ringnumber=1
totem.interface.bindnetaddr=192.168.5.2
totem.interface.broadcast=yes
totem.interface.mcastport=1001

Comment 5 Steven Dake 2011-11-30 17:19:58 UTC

Fabio,

Great work!.  I have general concerns about mixing ipv4 and ipv6 in RRP.  This is not something we should support.

Regards
-steve

Comment 6 Fabio Massimo Di Nitto 2011-11-30 18:18:10 UTC

(In reply to comment #5)
> Fabio,
> 
> Great work!.

Thanks :)

>I have general concerns about mixing ipv4 and ipv6 in RRP.  This
> is not something we should support.

The patch protects against it already.

Test case #3 shows all 4 possible combinations and 2 of them (mixed) will fail to start with error.

Comment 7 Lon Hohberger 2011-11-30 18:47:29 UTC

This looks great, Fabio -- the only thing that we'll need to work out is which configuration(s) are we going to support from luci/ccs.

Comment 8 Fabio Massimo Di Nitto 2011-11-30 18:55:50 UTC

(In reply to comment #7)
> This looks great, Fabio -- the only thing that we'll need to work out is which
> configuration(s) are we going to support from luci/ccs.

I think luci/ccs should create the config in the new format, but it will still need to be able to understand the old one and eventually convert it.

The old format is heavy to read and maintain.

Comment 10 Fabio Massimo Di Nitto 2012-03-07 14:29:13 UTC

respin with correct patch

Comment 11 Robert Peterson 2012-03-07 14:33:57 UTC

I tested this on my mixed-version cluster,
gfs-a16c-0{1,2,3,4}.mpc.lab.eng.bos.redhat.com and it works.

Comment 13 Fabio Massimo Di Nitto 2012-05-02 07:18:53 UTC

    Technical note added. If any revisions are required, please edit the "Technical Notes" field
    accordingly. All revisions will be proofread by the Engineering Content Services team.
    
    New Contents:
Cause: cman was not able to configure Redundant Ring Protocol (RRP) correctly in corosync.
Consequence: RRP deployments would not work properly.
Fix: cman has been improved to both configure RRP properly and perform extra sanity checks on user configurations.
Result: it is now easier to deploy cluster with RRP and error reports are more extensive.

Comment 16 errata-xmlrpc 2012-06-20 13:54:50 UTC

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

http://rhn.redhat.com/errata/RHBA-2012-0861.html

Note You need to log in before you can comment on or make changes to this bug.