Using Oracle Solaris Unified Archives to Replicate an Oracle Solaris Cluster 4.2 Cluster

Oracle Solaris Automated Installation (AI) was first supported in Oracle Solaris Cluster 4.0 software to install and configure a new cluster from IPS repositories. With Oracle Solaris Unified Archive introduced in Oracle Solaris 11.2 software, the automated installation of Oracle Solaris Cluster 4.2 software with the Oracle Solaris 11.2 OS is expanded with the following added functionality:

  1. Install and configure a new cluster from archives. The cluster is in the initial state, just like one created using the standard method of running scinstall(1M) on the potential cluster nodes.

  2. Restore cluster nodes from recovery archives created for the same nodes, due to hardware failures on the nodes.

  3. Replicate a new cluster from archives created for the nodes in a source cluster. The software packages and the cluster configuration on the new nodes will remain the same as in the source cluster, but the new nodes and some cluster objects (such as zone clusters) can have different system identities.

This document shows how to replicate a new cluster from the Oracle Solaris Unified Archives.

Replicating clusters can greatly reduce the effort of installing the Oracle Solaris OS, installing Oracle Solaris Cluster packages, configuring the nodes to form a cluster, installing and configuring applications, and applying SRUs or patches for maintenance. All of this can be done in one installation.

At first, the source cluster needs to be set up. This effort cannot be omitted. However, for use cases such as engineered systems, archives can be created for the source cluster nodes as master images, and can be used to replicate multiple clusters, as many as one wants, using this feature in Oracle Solaris Cluster 4.2 software. The more clusters you replicate, the more effort it saves.

The procedure to replicate a cluster includes the following steps:

  • Set up the source cluster and create archives for each source node.
  • Set up the AI server and the DHCP server.
  • Run scinstall on the AI server to configure the installation of the new cluster nodes, and add the new nodes to the configuration of the DHCP server.
  • Boot net install the new cluster nodes.

These same steps also apply to configuring a new cluster and restoring cluster nodes. The only difference is that, when running scinstall on the AI server, the menu options and inputs are different.

Requirements of the new cluster

When replicating a new cluster from a source cluster, the new cluster needs to have the following similar hardware configuration as the source cluster:

  • Same number of nodes.
  • Same architecture.
  • Same private adapters for cluster transport.

As for the archives used to install the new cluster nodes, they must be created for the global zone, not from a non-global zone; do not mix using clone and recovery types of archives; and exclude datasets on shared storage when creating archives and migrate the data separately.

Currently, Oracle Solaris Unified Archive only supports ZFS, therefore other file systems and volume managers that are configured in the source cluster are not included in the archive. They can be migrated using the corresponding methods of those types.

If quorum servers or NAS devices are configured in the source cluster, the cluster configuration related to these objects are carried to the new cluster. However, the configuration on these hosts is not updated. Manual intervention is needed on these systems for them to function in the new cluster.

Set up the source cluster and create an archive for each node

The source cluster can be set up with any of the existing supported methods. Let’s use a two-node cluster (host names source-node1 and source-node2) with HA for NFS agent and a zone cluster as simple examples just for illustration purpose. This source cluster has a shared disk quorum device.

The zone cluster myzc is configured, installed, and booted online, and the two zone cluster nodes have host name source-zcnode1 and source-zcnode2.

# clzonecluster status

=== Zone Clusters ===

--- Zone Cluster Status ---

Name Brand Node Name Zone Host Name Status Zone Status

---- ----- --------- -------------- ------ -----------

myzc solaris source-node1 source-zcnode1 Online Running

source-node2 source-zcnode2 Online Running

The HA for NFS resource group (named nfs-rg) contains a logical-hostname resource nfs-lh, an HAStoragePlus resource hasp-rs, and a SUNW.nfs resource nfs-rs. The local mount point for the NFS file system is /local/ufs as shown using clresource show.

# clresource list -v

Resource Name Resource Type Resource Group

------------- ------------- --------------

nfs-rs SUNW.nfs:3.3 nfs-rg

hasp-rs SUNW.HAStoragePlus:10 nfs-rg

nfs-lh SUNW.LogicalHostname:5 nfs-rg

# clresource show -v -p HostnameList nfs-lh

=== Resources ===

Resource: nfs-lh

--- Standard and extension properties ---

HostnameList: source-lh-hostname

Class: extension

Description: List of hostnames this resource manages

Per-node: False

Type: stringarray

# clresource show -p FilesystemMountPoints hasp-rs

=== Resources ===

Resource: hasp-rs

--- Standard and extension properties ---

FilesystemMountPoints: /local/ufs

Class: extension

Description: The list of file system mountpoints

Per-node: False

Type: stringarray

On each node in this source cluster, create an unified archive for the global zone. The archive can be of clone type or recovery type. A clone archive created for the global zone contains multiple deployable systems. Non-global zones are excluded from the global zone, and each non-global zone is a single deployable system. A recovery archive created for the global zone consists just one single deployable system. Installing from a global zone recovery archive installs the global zone as well as the non-global zone that it contains. Check the Unified Archive introduction blog for more details.

Since there is a zone cluster in the source cluster, create a recovery archive for each node so that the zones will get installed.

As an example, we use archive-host as the name of the host that is mounted under /net and exports a file system to store the archives.

source-node1# archiveadm create -r /net/archive-host/export/source-node1.sua

source-node2# archiveadm create -r /net/archive-host/export/source-node2.sua

Note even though the recovery archive contains multiple boot environment (BE), only the current active BE is updated to function in the new cluster.

Set Up the AI server and the DHCP server

The new cluster nodes must be networked with a designated AI server, a DHCP server, and the host for storing the archive files. The AI server must have static IP and be installed with Oracle Solaris 11.2. The archive files created for the source cluster nodes must be accessible from the AI server as well. The archive location can be an autofs mount point mounted via /net/host, an http, or an https URL.

On the AI server, install the Oracle Solaris Cluster installation package ha-cluster/system/install. Do not install other Oracle Solaris Cluster packages. Installing this package also installs the Oracle Solaris installadm package and the Internet Systems Consortium (ISC) DHCP package service/network/dhcp/isc-dhcp, if they are not yet installed.

# pkg publisher

PUBLISHER TYPE STATUS P LOCATION

solaris origin online F http://ipkg.us.oracle.com/solaris11/release/

ha-cluster origin online F http://ipkg.us.oracle.com/ha-cluster/release

# pkg install ha-cluster/system/install

Run scinstall on the AI server to configure the installation

The tool to configure the AI installation of the new cluster nodes is /usr/cluster/bin/scinstall. As the same tool to configure a new cluster in the non-AI method, it gives a different menu when it runs at the AI server. Use the interactive method (running the scinstall command without any options) instead of the command line options method, as it gives help texts and prompts.

An archive must be created for the global zone of each source cluster node, and one archive file can only be specified to install just one node in the new cluster. This 1:1 mapping relationship must be maintained.

To avoid using the same host identities in both the source and the new clusters, the tool prompts for a host identity mapping file. It is a text file that contains 1:1 mapping from host identities used in the source cluster to the new host identities in the new cluster. The host names of the physical nodes do not need to be included in this file. The host names or IPs configured for zone clusters, non-global zones, and logical-hostname or shared-address resources can be included. Hosts used in name services for zones can be included too.

The file can contain multiple lines, and each line has two columns. The first column is the host name or IP used in the source cluster, and the second column is the corresponding new host name or IP address that will be used in the new cluster. “#” at the beginning of a line marks the comment lines.

# cat /net/archive-host/export/mapping_new_cluster.txt

# zone cluster host names

source-zcnode1 new-zcnode1

source-zcnode2 new-zcnode2

# ha-nfs logicalhost resource host name

source-lh-hostname new-lh-hostname

The scinstall tool provides menus and options, and prompts for the following user inputs, as shown with examples in the following table.

Item

Description

Example Values

Root password

Password for root account to access the nodes upon installation completes


Cluster name

Name of the new cluster

new_cluster

Node names and MACs

Node names in the new cluster and their MACs

new-node1 00:14:4F:FA:42:42

new-node2 00:14:4F:FA:EF:E8

Archive locations

Archive location for each node in the new cluster

new-node1: /net/archive-host/export/source-node1.sua

new-node2: /net/archive-host/export/source-node2.sua

Network address and netmask (optional)

Network address for private network and netmask

Network address: 172.17.0.0

netmask: 255.255.240.0

Host ID mapping file (optional)

A text file for 1:1 mapping from the old host identities in the source cluster to new host identities

/net/archive-host/export/mapping_new_cluster.txt

After confirming all the inputs, scinstall creates the install service for the new cluster, named as cluster-name-{sparc|i386}, creates manifest files and sysconfig profiles for each new node, and associates each node as clients to this install service.

The output from running scinstall also contains the instructions for adding the new cluster nodes as clients to the DHCP server. Installing Oracle Solaris 11.2 Systems has every detail about setting up the ISC DHCP server and adding clients to the DHCP configuration.

Boot net install the new cluster nodes

Boot the nodes from network to start the installation. For SPARC nodes,

Ok boot net:dhcp – install

For x86 nodes, press the proper function key to boot from the network.

After the installation completes, the nodes are automatically rebooted three times before joining the new cluster. Check the cluster nodes status with the /usr/clutser/bin/clquorum command. The shared disk quorum device is re-created using a DID in the freshly populated device name spaces.

# clquorum status

=== Cluster Quorum ===

--- Quorum Votes Summary from (latest node reconfiguration) ---

Needed Present Possible

------ ------- --------

2 3 3

--- Quorum Votes by Node (current status) ---

Node Name Present Possible Status

--------- ------- -------- ------

new-node1 1 1 Online

new-node2 1 1 Online

--- Quorum Votes by Device (current status) ---

Device Name Present Possible Status

----------- ------- -------- ------

d1 1 1 Online

The zone cluster has updated host names. Its zone status is Running but its cluster status is Offline. Check its configuration for any manual updates to perform in the new environment. If the new configuration looks fine, using the clzonecluster reboot command to bring the zone cluster to Online cluster status.

# clzonecluster status

=== Zone Clusters ===

--- Zone Cluster Status ---

Name Brand Node Name Zone Host Name Status Zone Status

---- ----- --------- -------------- ------ -----------

myzc solaris source-node1 new-zcnode1 Offline Running

source-node2 new-zcnode2 Offline Running

# clzonecluster reboot myzc

The private network address is changed to the one specified during running scinstall.

# cluster show -t global | grep private_

private_netaddr: 172.17.0.0

private_netmask: 255.255.240.0

The nfs-rg and its resources are in offline state as well with updated host name for the logical-hostname resource.

# clresource show -v -p HostnameList nfs-lh

=== Resources ===

Resource: nfs-lh

--- Standard and extension properties ---

HostnameList: new-lh-hostname

Class: extension

Description: List of hostnames this resource manages

Per-node: False

Type: stringarray

Since the archive does not contain the UFS file systems used in resource group nfs-rg, the mount entry (I.e., /local/ufs) for this UFS file system is commented out in the /etc/vfstab file. Update /etc/vfstab to bring it back on all the nodes. Then on one node, create the file system on that shared disk. Finally bring the resource group nfs-rg online using the clresourcegroup online command. The files in this file system in the source cluster can be copied over, or just start with the new file system.

# grep '/local/ufs' /etc/vfstab

/dev/global/dsk/d4s6 /dev/global/rdsk/d4s6 /local/ufs ufs 2 no logging

# cldevice list -v d4

DID Device Full Device Path

---------- ----------------

d4 new-node2:/dev/rdsk/c1d4

d4 new-node1:/dev/rdsk/c1d4

# format /dev/rdsk/c1d4s2

# newfs /dev/did/rdsk/d4s6

newfs: construct a new file system /dev/did/rdsk/d4s6: (y/n)? y

# clresourcegroup online nfs-rg

# clresource status

=== Cluster Resources ===

Resource Name Node Name State Status Message

------------- --------- ----- --------------

nfs-rs new-node1 Online Online - Service is online.

new-node2 Offline Offline

hasp-rs new-node1 Online Online

new-node2 Offline Offline

nfs-lh new-node1 Online Online - LogicalHostname online.

new-node2 Offline Offline

At this stage, this newly replicated cluster and the agent are fully functional.

- Lucia Lai <yue.lai@oracle.com>
Oracle Solaris Cluster 
Comments:

Post a Comment:
  • HTML Syntax: NOT allowed
About

Oracle Solaris Cluster Engineering Blog

Search

Archives
« May 2015
SunMonTueWedThuFriSat
     
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
18
19
20
21
22
23
24
25
26
27
28
29
30
31
      
Today