xCAT/xcat-core
2
0
Fork 0
mirror of https://github.com/xcat2/xcat-core.git synced 2025-05-22 11:42:05 +00:00
Code Issues Releases Wiki Activity

Add hierarchy doc

Browse Source 
Add hierarchy documents including introduction, database
management,defining node, service node setup, test service
node and appendix.

TODO: resolve the fake_todo link reference in the following patch
This commit is contained in:
chenglch 2015-09-02 02:47:30 -04:00
parent 4bbde50119
commit 57c7dca6da
16 changed files with 1158 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

167
docs/source/advanced/hierarchy/appendix_a_setup_backup_service_nodes.rst Normal file
View File

@ -0,0 +1,167 @@
Appendix A: Setup backup Service Nodes
======================================
For reliability, availability, and serviceability purposes you may wish to
designate backup service nodes in your hierarchical cluster. The backup
service node will be another active service node that is set up to easily
take over from the original service node if a problem occurs. This is not an
automatic failover feature. You will have to initiate the switch from the
primary service node to the backup manually. The xCAT support will handle most
of the setup and transfer of the nodes to the new service node. This
procedure can also be used to simply switch some compute nodes to a new
service node, for example, for planned maintenance.
Abbreviations used below:
* MN - management node.
* SN - service node.
* CN - compute node.
Initial deployment
------------------
Integrate the following steps into the hierarchical deployment process
described above.
#. Make sure both the primary and backup service nodes are installed,
configured, and can access the MN database.
#. When defining the CNs add the necessary service node values to the
"servicenode" and "xcatmaster" attributes of the `node definitions
<http://localhost/fake_todo>`_.
#. (Optional) Create an xCAT group for the nodes that are assigned to each SN.
This will be useful when setting node attributes as well as providing an
easy way to switch a set of nodes back to their original server.
To specify a backup service node you must specify a comma-separated list of
two **service nodes** for the servicenode value of the compute node. The first
one is the primary and the second is the backup (or new SN) for that node.
Use the hostnames of the SNs as known by the MN.
For the **xcatmaster** value you should only include the primary SN, as known
by the compute node.
In most hierarchical clusters, the networking is such that the name of the
SN as known by the MN is different than the name as known by the CN. (If
they are on different networks.)
The following example assume the SN interface to the MN is on the "a"
network and the interface to the CN is on the "b" network. To set the
attributes you would run a command similar to the following. ::
chdef <noderange> servicenode="xcatsn1a,xcatsn2a" xcatmaster="xcatsn1b"
The process can be simplified by creating xCAT node groups to use as the
<noderange> in the `chdef <http://localhost/fake_todo>`_ command. To create an
xCAT node group containing all the nodes that have the service node "SN27"
you could run a command similar to the following. ::
mkdef -t group sn1group members=node[01-20]
**Note: Normally backup service nodes are the primary SNs for other compute
nodes. So, for example, if you have 2 SNs, configure half of the CNs to use
the 1st SN as their primary SN, and the other half of CNs to use the 2nd SN
as their primary SN. Then each SN would be configured to be the backup SN
for the other half of CNs.**
When you run `makedhcp <http://localhost/fake_todo>`_, it will configure dhcp
and tftp on both the primary and backup SNs, assuming they both have network
access to the CNs. This will make it possible to do a quick SN takeover
without having to wait for replication when you need to switch.
xdcp Behaviour with backup servicenodes
---------------------------------------
The xdcp command in a hierarchical environment must first copy (scp) the
files to the service nodes for them to be available to scp to the node from
the service node that is it's master. The files are placed in
``/var/xcat/syncfiles`` directory by default, or what is set in site table
SNsyncfiledir attribute. If the node has multiple service nodes assigned,
then xdcp will copy the file to each of the service nodes assigned to the
node. For example, here the files will be copied (scp) to both service1 and
rhsn. lsdef cn4 | grep servicenode. ::
servicenode=service1,rhsn
f a service node is offline ( e.g. service1), then you will see errors on
your xdcp command, and yet if rhsn is online then the xdcp will actually
work. This may be a little confusing. For example, here service1 is offline,
but we are able to use rhsn to complete the xdcp. ::
xdcp cn4 /tmp/lissa/file1 /tmp/file1
service1: Permission denied (publickey,password,keyboard-interactive).
service1: Permission denied (publickey,password,keyboard-interactive).
service1: lost connection
The following servicenodes: service1, have errors and cannot be updated
Until the error is fixed, xdcp will not work to nodes serviced by these service nodes.
xdsh cn4 ls /tmp/file1
cn4: /tmp/file1
Switch to the backup SN
-----------------------
When an SN fails, or you want to bring it down for maintenance, use this
procedure to move its CNs over to the backup SN.
Move the nodes to the new service nodes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Use the xCAT `snmove <http://localhost/fake_todo>`_ to make the database
updates necessary to move a set of nodes from one service node to another, and
to make configuration modifications to the nodes.
For example, if you want to switch all the compute nodes that use service
node "sn1" to the backup SN (sn2), run: ::
snmove -s sn1
Modified database attributes
""""""""""""""""""""""""""""
The **snmove** command will check and set several node attribute values.
**servicenode**: : This will be set to either the second server name in the
servicenode attribute list or the value provided on the command line.
**xcatmaster**: : Set with either the value provided on the command line or it
will be automatically determined from the servicenode attribute.
**nfsserver**: : If the value is set with the source service node then it will
be set to the destination service node.
**tftpserver**: : If the value is set with the source service node then it will
be reset to the destination service node.
**monserver**: : If set to the source service node then reset it to the
destination servicenode and xcatmaster values.
**conserver**: : If set to the source service node then reset it to the
destination servicenode and run **makeconservercf**
Run postscripts on the nodes
""""""""""""""""""""""""""""
If the CNs are up at the time the snmove command is run then snmove will run
postscripts on the CNs to reconfigure them for the new SN. The "syslog"
postscript is always run. The "mkresolvconf" and "setupntp" scripts will be
run IF they were included in the nodes postscript list.
You can also specify an additional list of postscripts to run.
Modify system configuration on the nodes
""""""""""""""""""""""""""""""""""""""""
If the CNs are up the snmove command will also perform some configuration on
the nodes such as setting the default gateway and modifying some
configuration files used by xCAT.
Switching back
--------------
The process for switching nodes back will depend on what must be done to
recover the original service node. If the SN needed to be reinstalled, you
need to set it up as an SN again and make sure the CN images are replicated
to it. Once you've done this, or if the SN's configuration was not lost,
then follow these steps to move the CNs back to their original SN:
* Use snmove: ::
snmove sn1group -d sn1

66
docs/source/advanced/hierarchy/appendix_b_diagnostics.rst Normal file
View File

@ -0,0 +1,66 @@
Appendix B: Diagnostics
=======================
* **root ssh keys not setup** -- If you are prompted for a password when ssh to
the service node, then check to see if /root/.ssh has authorized_keys. If
the directory does not exist or no keys, on the MN, run xdsh service -K,
to exchange the ssh keys for root. You will be prompted for the root
password, which should be the password you set for the key=system in the
passwd table.
* **XCAT rpms not on SN** --On the SN, run rpm -qa | grep xCAT and make sure
the appropriate xCAT rpms are installed on the servicenode. See the list of
xCAT rpms in :ref:`setup_service_node_stateful_label`. If rpms
missing check your install setup as outlined in Build the Service Node
Stateless Image for diskless or :ref:`setup_service_node_stateful_label` for
diskfull installs.
* **otherpkgs(including xCAT rpms) installation failed on the SN** --The OS
repository is not created on the SN. When the "yum" command is processing
the dependency, the rpm packages (including expect, nmap, and httpd, etc)
required by xCATsn can't be found. In this case, please check whether the
``/install/postscripts/repos/<osver>/<arch>/`` directory exists on the MN.
If it is not on the MN, you need to re-run the "copycds" command, and there
will be some file created under the
``/install/postscripts/repos/<osver>/<arch>`` directory on the MN. Then, you
need to re-install the SN, and this issue should be gone.
* **Error finding the database/starting xcatd** -- If on the Service node when
you run tabdump site, you get "Connection failure: IO::Socket::SSL:
connect: Connection refused at ``/opt/xcat/lib/perl/xCAT/Client.pm``". Then
restart the xcatd daemon and see if it passes by running the command:
service xcatd restart. If it fails with the same error, then check to see
if ``/etc/xcat/cfgloc`` file exists. It should exist and be the same as
``/etc/xcat/cfgloc`` on the MN. If it is not there, copy it from the MN to
the SN. The run service xcatd restart. This indicates the servicenode
postscripts did not complete successfully. Check to see your postscripts
table was setup correctly in :ref:`add_service_node_postscripts_label` to the
postscripts table.
* **Error accessing database/starting xcatd credential failure**-- If you run
tabdump site on the servicenode and you get "Connection failure:
IO::Socket::SSL: SSL connect attempt failed because of handshake
problemserror:14094418:SSL routines:SSL3_READ_BYTES:tlsv1 alert unknown ca
at ``/opt/xcat/lib/perl/xCAT/Client.pm``", check ``/etc/xcat/cert``. The
directory should contain the files ca.pem and server-cred.pem. These were
suppose to transfer from the MN ``/etc/xcat/cert`` directory during the
install. Also check the ``/etc/xcat/ca`` directory. This directory should
contain most files from the ``/etc/xcat/ca`` directory on the MN. You can
manually copy them from the MN to the SN, recursively. This indicates the
the servicenode postscripts did not complete successfully. Check to see
your postscripts table was setup correctly in
:ref:`add_service_node_postscripts_label` to the postscripts table. Again
service xcatd restart and try the tabdump site again.
* **Missing ssh hostkeys** -- Check to see if ``/etc/xcat/hostkeys`` on the SN,
has the same files as ``/etc/xcat/hostkeys`` on the MN. These are the ssh
keys that will be installed on the compute nodes, so root can ssh between
compute nodes without password prompting. If they are not there copy them
from the MN to the SN. Again, these should have been setup by the
servicenode postscripts.
* **Errors running hierarchical commands such as xdsh** -- xCAT has a number of
commands that run hierarchically. That is, the commands are sent from xcatd
on the management node to the correct service node xcatd, which in turn
processes the command and sends the results back to xcatd on the management
node. If a hierarchical command such as xcatd fails with something like
"Error: Permission denied for request", check ``/var/log/messages`` on the
management node for errors. One error might be "Request matched no policy
rule". This may mean you will need to add policy table entries for your
xCAT management node and service node:

14
docs/source/advanced/hierarchy/appendix_c_migrating_mn_to_sn.rst Normal file
View File

@ -0,0 +1,14 @@
Appendix C: Migrating a Management Node to a Service Node
=========================================================
If you find you want to convert an existing Management Node to a Service
Node you need to work with the xCAT team. It is recommended for now, to
backup your database, setup your new Management Server, and restore your
database into it. Take the old Management Node and remove xCAT and all xCAT
directories, and your database. See ``Uninstalling_xCAT
<http://localhost/fake_todo>`_ and then follow the process for setting up a
SN as if it is a new node.

13
docs/source/advanced/hierarchy/appendix_d_set_up_hierarchical_conserver.rst Normal file
View File

@ -0,0 +1,13 @@
Appendix D: Set up Hierarchical Conserver
=========================================
To allow you to open the rcons from the Management Node using the
conserver daemon on the Service Nodes, do the following:
* Set nodehm.conserver to be the service node (using the ip that faces the
management node) ::
chdef -t <noderange> conserver=<servicenodeasknownbytheMN>
makeconservercf
service conserver stop
service conserver start

11
docs/source/advanced/hierarchy/configure_dhcp.rst Normal file
View File

@ -0,0 +1,11 @@
Configure DHCP
==============
Add the relevant networks into the DHCP configuration, refer to:
`XCAT_pLinux_Clusters/#setup-dhcp <http://localhost/fake_todo>`_
Add the defined nodes into the DHCP configuration, refer to:
`XCAT_pLinux_Clusters/#configure-dhcp <http://localhost/fake_todo>`_

39
docs/source/advanced/hierarchy/define_and_install_compute_node.rst Normal file
View File

@ -0,0 +1,39 @@
Define and install your Compute Nodes
=====================================
Make /install available on the Service Nodes
--------------------------------------------
Note that all of the files and directories pointed to by your osimages should
be placed under the directory referred to in site.installdir (usually
/install), so they will be available to the service nodes. The installdir
directory is mounted or copied to the service nodes during the hierarchical
installation.
If you are not using the NFS-based statelite method of booting your compute
nodes and you are not using service node pools, set the installloc attribute
to "/install". This instructs the service node to mount /install from the
management node. (If you don't do this, you have to manually sync /install
between the management node and the service nodes.)
::
chdef -t site clustersite installloc="/install"
Make compute node syncfiles available on the servicenodes
---------------------------------------------------------
If you are not using the NFS-based statelite method of booting your compute
nodes, and you plan to use the syncfiles postscript to update files on the
nodes during install, you must ensure that those files are sync'd to the
servicenodes before the install of the compute nodes. To do this after your
nodes are defined, you will need to run the following whenever the files in
your synclist change on the Management Node:
::
updatenode <computenoderange> -f
At this point you can return to the documentation for your cluster environment
to define and deploy your compute nodes.

345
docs/source/advanced/hierarchy/define_service_node.rst Normal file
View File

@ -0,0 +1,345 @@
Define the service nodes in the database
========================================
This document assumes that you have previously **defined** your compute nodes
in the database. It is also possible at this point that you have generic
entries in your db for the nodes you will use as service nodes as a result of
the node discovery process. We are now going to show you how to add all the
relevant database data for the service nodes (SN) such that the SN can be
installed and managed from the Management Node (MN). In addition, you will
be adding the information to the database that will tell xCAT which service
nodes (SN) will service which compute nodes (CN).
For this example, we have two service nodes: **sn1** and **sn2**. We will call
our Management Node: **mn1**. Note: service nodes are, by convention, in a
group called **service**. Some of the commands in this document will use the
group **service** to update all service nodes.
Note: a Service Node's service node is the Management Node; so a service node
must have a direct connection to the management node. The compute nodes do not
have to be directly attached to the Management Node, only to their service
node. This will all have to be defined in your networks table.
Add Service Nodes to the nodelist Table
---------------------------------------
Define your service nodes (if not defined already), and by convention we put
them in a **service** group. We usually have a group compute for our compute
nodes, to distinguish between the two types of nodes. (If you want to use your
own group name for service nodes, rather than service, you need to change some
defaults in the xCAT db that use the group name service. For example, in the
postscripts table there is by default a group entry for service, with the
appropriate postscripts to run when installing a service node. Also, the
default ``kickstart/autoyast`` template, pkglist, etc that will be used have
files names based on the profile name service.) ::
mkdef sn1,sn2 groups=service,ipmi,all
Add OS and Hardware Attributes to Service Nodes
-----------------------------------------------
When you ran copycds, it creates several osimage definitions, including some
appropriate for SNs. Display the list of osimages and choose one with
"service" in the name: ::
lsdef -t osimage
For this example, let's assume you chose the stateful osimage definition for
rhels 6.3: rhels6.3-x86_64-install-service . If you want to modify any of the
osimage attributes (e.g. ``kickstart/autoyast`` template, pkglist, etc),
make a copy of the osimage definition and also copy to ``/install/custom``
any files it points to that you are modifying.
Now set some of the common attributes for the SNs at the group level: ::
chdef -t group service arch=x86_64 \
os=rhels6.3 \
nodetype=osi
profile=service \
netboot=xnba installnic=mac \
primarynic=mac \
provmethod=rhels6.3-x86_64-install-service
Add Service Nodes to the servicenode Table
------------------------------------------
An entry must be created in the servicenode table for each service node or the
service group. This table describes all the services you would like xcat to
setup on the service nodes. (Even if you don't want xCAT to set up any
services - unlikely - you must define the service nodes in the servicenode
table with at least one attribute set (you can set it to 0), otherwise it will
not be recognized as a service node.)
When the xcatd daemon is started or restarted on the service node, it will
make sure all of the requested services are configured and started. (To
temporarily avoid this when restarting xcatd, use "service xcatd reload"
instead.)
To set up the minimum recommended services on the service nodes: ::
chdef -t group -o service setupnfs=1 \
setupdhcp=1 setuptftp=1 \
setupnameserver=1 \
setupconserver=1
.. TODO
See the setup* attributes in the `node object definition man page
<http://localhost/fake_todo>`_ for the services available. (The HTTP server
is also started when setupnfs is set.)
If you are using the setupntp postscript on the compute nodes, you should also
set setupntp=1. For clusters with subnetted management networks (i.e. the
network between the SN and its compute nodes is separate from the network
between the MN and the SNs) you might want to also set setupipforward=1.
.. _add_service_node_postscripts_label:
Add Service Node Postscripts
----------------------------
By default, xCAT defines the service node group to have the "servicenode"
postscript run when the SNs are installed or diskless booted. This
postscript sets up the xcatd credentials and installs the xCAT software on
the service nodes. If you have your own postscript that you want run on the
SN during deployment of the SN, put it in ``/install/postscripts`` on the MN
and add it to the service node postscripts or postbootscripts. For example: ::
chdef -t group -p service postscripts=<mypostscript>
Notes:
* For Red Hat type distros, the postscripts will be run before the reboot
of a kickstart install, and the postbootscripts will be run after the
reboot.
* Make sure that the servicenode postscript is set to run before the
otherpkgs postscript or you will see errors during the service node
deployment.
* The -p flag automatically adds the specified postscript at the end of the
comma-separated list of postscripts (or postbootscripts).
If you are running additional software on the service nodes that need **ODBC**
to access the database (e.g. LoadLeveler or TEAL), use this command to add
the xCAT supplied postbootscript called "odbcsetup". ::
chdef -t group -p service postbootscripts=odbcsetup
Assigning Nodes to their Service Nodes
--------------------------------------
The node attributes **servicenode** and **xcatmaster** define which SN
services this particular node. The servicenode attribute for a compute node
defines which SN the MN should send a command to (e.g. xdsh), and should be
set to the hostname or IP address of the service node that the management
node contacts it by. The xcatmaster attribute of the compute node defines
which SN the compute node should boot from, and should be set to the
hostname or IP address of the service node that the compute node contacts it
by. Unless you are using service node pools, you must set the xcatmaster
attribute for a node when using service nodes, even if it contains the same
value as the node's servicenode attribute.
Host name resolution must have been setup in advance, with ``/etc/hosts``, DNS
or dhcp to ensure that the names put in this table can be resolved on the
Management Node, Service nodes, and the compute nodes. It is easiest to have a
node group of the compute nodes for each service node. For example, if all the
nodes in node group compute1 are serviced by sn1 and all the nodes in node
group compute2 are serviced by sn2:
::
chdef -t group compute1 servicenode=sn1 xcatmaster=sn1-c
chdef -t group compute2 servicenode=sn2 xcatmaster=sn2-c
Note: in this example, sn1 and sn2 are the node names of the service nodes
(and therefore the hostnames associated with the NICs that the MN talks to).
The hostnames sn1-c and sn2-c are associated with the SN NICs that communicate
with their compute nodes.
Note: if not set, the attribute tftpserver's default value is xcatmaster,
but in some releases of xCAT it has not defaulted correctly, so it is safer
to set the tftpserver to the value of xcatmaster.
These attributes will allow you to specify which service node should run the
conserver (console) and monserver (monitoring) daemon for the nodes in the
group specified in the command. In this example, we are having each node's
primary SN also act as its conserver and monserver (the most typical setup).
::
chdef -t group compute1 conserver=sn1 monserver=sn1,sn1-c
chdef -t group compute2 conserver=sn2 monserver=sn2,sn2-c
Service Node Pools
^^^^^^^^^^^^^^^^^^
Service Node Pools are multiple service nodes that service the same set of
compute nodes. Having multiple service nodes allows backup service node(s) for
a compute node when the primary service node is unavailable, or can be used
for work-load balancing on the service nodes. But note that the selection of
which SN will service which compute node is made at compute node boot time.
After that, the selection of the SN for this compute node is fixed until the
compute node is rebooted or the compute node is explicitly moved to another SN
using the `snmove <http://localhost/fake_todo>`_ command.
To use Service Node pools, you need to architect your network such that all of
the compute nodes and service nodes in a partcular pool are on the same flat
network. If you don't want the management node to respond to manage some of
the compute nodes, it shouldn't be on that same flat network. The
site, dhcpinterfaces attribute should be set such that the SNs' DHCP daemon
only listens on the NIC that faces the compute nodes, not the NIC that faces
the MN. This avoids some timing issues when the SNs are being deployed (so
that they don't respond to each other before they are completely ready). You
also need to make sure the `networks <http://localhost/fake_todo>`_ table
accurately reflects the physical network structure.
To define a list of service nodes that support a set of compute nodes, set the
servicenode attribute to a comma-delimited list of the service nodes. When
running an xCAT command like xdsh or updatenode for compute nodes, the list
will be processed left to right, picking the first service node on the list to
run the command. If that service node is not available, then the next service
node on the list will be chosen until the command is successful. Errors will
be logged. If no service node on the list can process the command, then the
error will be returned. You can provide some load-balancing by assigning your
service nodes as we do below.
When using service node pools, the intent is to have the service node that
responds first to the compute node's DHCP request during boot also be the
xcatmaster, the tftpserver, and the NFS/http server for that node. Therefore,
the xcatmaster and nfsserver attributes for nodes should not be set. When
nodeset is run for the compute nodes, the service node interface on the
network to the compute nodes should be defined and active, so that nodeset
will default those attribute values to the "node ip facing" interface on that
service node.
For example: ::
chdef -t node compute1 servicenode=sn1,sn2 xcatmaster="" nfsserver=""
chdef -t node compute2 servicenode=sn2,sn1 xcatmaster="" nfsserver=""
You need to set the sharedtftp site attribute to 0 so that the SNs will not
automatically mount the ``/tftpboot`` directory from the management node:
::
chdef -t site clustersite sharedtftp=0
For statefull (full-disk) node installs, you will need to use a local
``/install`` directory on each service node. The ``/install/autoinst/node``
files generated by nodeset will contain values specific to that service node
for correctly installing the nodes.
::
chdef -t site clustersite installloc=""
With this setting, you will need to remember to rsync your ``/install``
directory from the xCAT management node to the service nodes anytime you
change your ``/install/postscripts``, custom osimage files, os repositories,
or other directories. It is best to exclude the ``/install/autoinst`` directory
from this rsync.
::
rsync -auv --exclude 'autoinst' /install sn1:/
Note: If your service nodes are stateless and site.sharedtftp=0, if you reboot
any service node when using servicenode pools, any data written to the local
``/tftpboot`` directory of that SN is lost. You will need to run nodeset for
all of the compute nodes serviced by that SN again.
For additional information about service node pool related settings in the
networks table, see ref: networks table, see :ref:`setup_networks_table_label`.
Conserver and Monserver and Pools
"""""""""""""""""""""""""""""""""
The support of conserver and monserver with Service Node Pools is still not
supported. You must explicitly assign these functions to a service node using
the nodehm.conserver and noderes.monserver attribute as above.
Setup Site Table
----------------
If you are not using the NFS-based statelite method of booting your compute
nodes, set the installloc attribute to ``/install``. This instructs the
service node to mount ``/install`` from the management node. (If you don't do
this, you have to manually sync ``/install`` between the management node and
the service nodes.) ::
chdef -t site clustersite installloc="/install"
For IPMI controlled nodes, if you want the out-of-band IPMI operations to be
done directly from the management node (instead of being sent to the
appropriate service node), set site.ipmidispatch=n.
If you want to throttle the rate at which nodes are booted up, you can set the
following site attributes:
* syspowerinterval
* syspowermaxnodes
* powerinterval (system p only)
See the `site table man page <http://localhost/fack_todo>`_ for details.
.. _setup_networks_table_label:
Setup networks Table
--------------------
All networks in the cluster must be defined in the networks table. When xCAT
is installed, it runs makenetworks, which creates an entry in the networks
table for each of the networks the management node is on. You need to add
entries for each network the service nodes use to communicate to the compute
nodes.
For example: ::
mkdef -t network net1 net=10.5.1.0 mask=255.255.255.224 gateway=10.5.1.1
If you want to set the nodes' xcatmaster as the default gateway for the nodes,
the gateway attribute can be set to keyword "<xcatmaster>". In this case, xCAT
code will automatically substitute the IP address of the node's xcatmaster for
the keyword. Here is an example:
::
mkdef -t network net1 net=10.5.1.0 mask=255.255.255.224 gateway=<xcatmaster>
The ipforward attribute should be enabled on all the xcatmaster nodes that
will be acting as default gateways. You can set ipforward to 1 in the
servicenode table or add the line "net.ipv4.ip_forward = 1" in file
``/etc/sysctl``.conf and then run "sysctl -p /etc/sysctl.conf" manually to
enable the ipforwarding.
Note:If using service node pools, the networks table dhcpserver attribute can
be set to any single service node in your pool. The networks tftpserver, and
nameserver attributes should be left blank.
Verify the Tables
--------------------
To verify that the tables are set correctly, run lsdef on the service nodes,
compute1, compute2: ::
lsdef service,compute1,compute2
Add additional adapters configuration script (optional)
------------------------------------------------------------
It is possible to have additional adapter interfaces automatically configured
when the nodes are booted. XCAT provides sample configuration scripts for
ethernet, IB, and HFI adapters. These scripts can be used as-is or they can be
modified to suit your particular environment. The ethernet sample is
``/install/postscript/configeth``. When you have the configuration script that
you want you can add it to the "postscripts" attribute as mentioned above. Make
sure your script is in the ``/install/postscripts`` directory and that it is
executable.
Note: For system p servers, if you plan to have your service node perform the
hardware control functions for its compute nodes, it is necessary that the SN
ethernet network adapters connected to the HW service VLAN be configured.
Configuring Secondary Adapters
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
o configure secondary adapters, see `Configuring_Secondary_Adapters
<http://localhost/fake_todo>`_

2
docs/source/advanced/hierarchy/hierarchy_cluster.rst
View File

@ -1,2 +0,0 @@
Setting Up a Linux Hierarchical Cluster
=======================================

11
docs/source/advanced/hierarchy/index.rst
View File

@ -5,3 +5,14 @@ Hierarchical Clusters
:maxdepth: 2
introduction.rst
setup_mn_hierachical_database.rst
define_service_node.rst
configure_dhcp.rst
setup_service_node.rst
service_node_for_diskfull.rst
service_node_for_diskless.rst
test_service_node_installation.rst
appendix_a_setup_backup_service_nodes.rst
appendix_b_diagnostics.rst
appendix_c_migrating_mn_to_sn.rst
appendix_d_set_up_hierarchical_conserver.rst

48
docs/source/advanced/hierarchy/introduction.rst
View File

@ -1,10 +1,50 @@
Introduction
============
In supporting large clusters, it is desirable to have more than a single management node handling the installation and management of compute nodes.
In xCAT, these additional nodes are referred to as *Service Nodes (SN)*. The management node can delegate all management operations for a compute node to the Service node that is managing them. You can have one of more service nodes configured to install and manage a group of compute nodes.
In large clusters, it is desirable to have more than one node (the Management
Node - MN) handle the installation and management of the compute nodes. We
call these additional nodes **service nodes (SN)**. The management node can
delegate all management operations needed by a compute node to the SN that is
managing that compute node. You can have one or more service nodes setting up
to install and manage groups of compute nodes.
Service Nodes
-------------
With xCAT, you have the choice of either having each service node
install/manage a specific set of compute nodes, or having a pool of service
nodes, any of which can respond to an installation request from a compute
node. (Service node pools must be aligned with the network broadcast domains,
because the way a compute node choose its SN for that boot is by whoever
responds to the DHCP request broadcast first.) You can also have a hybrid of
the 2 approaches, in which for each specific set of compute nodes you have 2
or more SNs in a pool.
Each SN runs an instance of xcatd, just like the MN does. The xcatd daemons
communicate with each other using the same XML/SSL protocol that the xCAT
client uses to communicate with xcatd on the MN.
Daemon-based Databases
----------------------
The service nodes need to communicate with the xCAT database on the Management
Node. They do this by using the remote client capability of the database (i.e.
they don't go through xcatd for that). Therefore the Management Node must be
running one of the daemon-based databases supported by xCAT (PostgreSQL,
MySQL).
The default SQLite database does not support remote clients and cannot be used
in hierarchical clusters. This document includes instructions for migrating
your cluster from SQLite to one of the other databases. Since the initial
install of xCAT will always set up SQLite, you must migrate to a database that
supports remote clients before installing your service nodes.
Setup
-----
xCAT will help you install your service nodes as well as install on the SNs
xCAT software and other required rpms such as perl, the database client, and
other pre-reqs. Service nodes require all the same software as the MN
(because it can do all of the same functions), except that there is a special
top level xCAT rpm for SNs called xCATsn vs. the xCAT rpm that is on the
Management Node. The xCATsn rpm tells the SN that the xcatd on it should
behave as an SN, not the MN.

156
docs/source/advanced/hierarchy/service_node_for_diskfull.rst Normal file
View File

@ -0,0 +1,156 @@
.. _setup_service_node_stateful_label:
Set Up the Service Nodes for Stateful (Diskful) Installation (optional)
=======================================================================
Any cluster using statelite compute nodes must use a stateful (diskful) service
nodes.
Note: If you are using diskless service nodes, go to
:ref:`setup_service_node_stateless_label`.
First, go to the `Download_xCAT <http://localhost/fake_todo>`_ site and
download the level of the xCAT tarball you desire. Then go to
http://localhost/fake_todo and get the latest xCAT dependency tarball.
**Note: All xCAT service nodes must be at the exact same xCAT version as the
xCAT Management Node**. Copy the files to the Management Node (MN) and untar
them in the appropriate sub-directory of ``/install/post/otherpkgs``
**Note for the appropriate directory below, check the
``otherpkgdir=/install/post/otherpkgs/rhels6.4/ppc64`` attribute of the
osimage defined for the servicenode.**
For example ubuntu14.04.1-ppc64el-install-service **** ::
mkdir -p /install/post/otherpkgs/ubuntu14.04.1/ppc64el/
cd /install/post/otherpkgs/ubuntu14.04.1/ppc64el/
tar jxvf core-rpms-snap.tar.bz2
tar jxvf xcat-dep-ubuntu*.tar.bz2
Next, add rpm names into your own version of
service.<osver>.<arch>.otherpkgs.pkglist file. In most cases, you can find an
initial copy of this file under /opt/xcat/share/xcat/install/<platform> . If
not, copy one from a similar platform.
::
mkdir -p /install/custom/install/ubuntu/
cp /opt/xcat/share/xcat/install/ubuntu/service.ubuntu.otherpkgs.pkglist/\
install/custom/install/ubuntu/service.ubuntu.otherpkgs.pkglist
vi /install/custom/install/ubuntu/service.ubuntu.otherpkgs.pkglist
Make sure the following entries are included in the
``/install/custom/install/ubuntu/service.ubuntu.otherpkgs.pkglist``:
::
mariadb-client
mariadb-common
xcatsn
conserver-xcat
The "pkgdir" should include the online/local ubuntu official mirror with the
following command:
::
chdef -t osimage -o ubuntu14.04.1-ppc64el-install-service \
-p pkgdir="http://ports.ubuntu.com/ubuntu-ports trusty main, \
http://ports.ubuntu.com/ubuntu-ports trusty-updates main, \
http://ports.ubuntu.com/ubuntu-ports trusty universe, \
http://ports.ubuntu.com/ubuntu-ports trusty-updates universe"
plus the "otherpkgdir" should include the mirror under otherpkgdir on MN, this
can be done with: ::
chdef -t osimage -o ubuntu14.04.1-ppc64el-install-service -p \
otherpkgdir="http:// < Name or ip of Management Node > \
/install/post/otherpkgs/ubuntu14.04.1/ppc64el/xcat-core/ \
trusty main, http://< Name or ip of Management Node > \
/install/post/otherpkgs/ubuntu14.04.1/ppc64el/xcat-dep/ trusty main"
**Note: you will be installing the xCAT Service Node rpm xCATsn meta-package
on the Service Node, not the xCAT Management Node meta-package. Do not install
both.**
Update the rhels6 RPM repository (rhels6 only)
----------------------------------------------
* This section could be removed after the powerpc-utils-1.2.2-18.el6.ppc64.rpm
is built in the base rhels6 ISO.
* The direct rpm download link is:
ftp://linuxpatch.ncsa.uiuc.edu/PERCS/powerpc-utils-1.2.2-18.el6.ppc64.rpm
* The update steps are as following: ::
put the new rpm in the base OS packages
cd /install/rhels6/ppc64/Server/Packages
mv powerpc-utils-1.2.2-17.el6.ppc64.rpm /tmp
cp /tmp/powerpc-utils-1.2.2-18.el6.ppc64.rpm
# make sure that the rpm is be readable by other users
chmod +r powerpc-utils-1.2.2-18.el6.ppc64.rpm
* create the repodata
::
cd /install/rhels6/ppc64/Server
ls -al repodata/
total 14316
dr-xr-xr-x 2 root root 4096 Jul 20 09:34 .
dr-xr-xr-x 3 root root 4096 Jul 20 09:34 ..
-r--r--r-- 1 root root 1305862 Sep 22 2010 20dfb74c144014854d3b16313907ebcf30c9ef63346d632369a19a4add8388e7-other.sqlite.bz2
-r--r--r-- 1 root root 1521372 Sep 22 2010 57b3c81512224bbb5cebbfcb6c7fd1f7eb99cca746c6c6a76fb64c64f47de102-primary.xml.gz
-r--r--r-- 1 root root 2823613 Sep 22 2010 5f664ea798d1714d67f66910a6c92777ecbbe0bf3068d3026e6e90cc646153e4-primary.sqlite.bz2
-r--r--r-- 1 root root 1418180 Sep 22 2010 7cec82d8ed95b8b60b3e1254f14ee8e0a479df002f98bb557c6ccad5724ae2c8-other.xml.gz
-r--r--r-- 1 root root 194113 Sep 22 2010 90cbb67096e81821a2150d2b0a4f3776ab1a0161b54072a0bd33d5cadd1c234a-comps-rhel6-Server.xml.gz
**-r--r--r-- 1 root root 1054944 Sep 22 2010 98462d05248098ef1724eddb2c0a127954aade64d4bb7d4e693cff32ab1e463c-comps-rhel6-Server.xml**
-r--r--r-- 1 root root 3341671 Sep 22 2010 bb3456b3482596ec3aa34d517affc42543e2db3f4f2856c0827d88477073aa45-filelists.sqlite.bz2
-r--r--r-- 1 root root 2965960 Sep 22 2010 eb991fd2bb9af16a24a066d840ce76365d396b364d3cdc81577e4cf6e03a15ae-filelists.xml.gz
-r--r--r-- 1 root root 3829 Sep 22 2010 repomd.xml
-r--r--r-- 1 root root 2581 Sep 22 2010 TRANS.TBL
createrepo -g repodata \
/98462d05248098ef1724eddb2c0a127954aade64d4bb7d4e693cff32ab1e463c-comps-rhel6-Server.xml
Note: you should use comps-rhel6-Server.xml with its key as the group file.
Set the node status to ready for installation
---------------------------------------------
Run nodeset to the osimage name defined in the provmethod attribute on your
service node. ::
nodeset service osimage="<osimagename>"
For example ::
nodeset service osimage="ubuntu14.04.1-ppc64el-install-service"
Initialize network boot to install Service Nodes
------------------------------------------------
::
rnetboot service
Monitor the Installation
------------------------
Watch the installation progress using either wcons or rcons: ::
wcons service # make sure DISPLAY is set to your X server/VNC or
rcons <one-node-at-a-time>
tail -f /var/log/messages
Note: We have experienced one problem while trying to install RHEL6 diskful
service node working with SAS disks. The service node cannot reboots from SAS
disk after the RHEL6 operating system has been installed. We are waiting for
the build with fixes from RHEL6 team, once meet this problem, you need to
manually select the SAS disk to be the first boot device and boots from the
SAS disk.
Update Service Node Diskfull Image
----------------------------------
If you need to update the service nodes later on with a new version of xCAT
and its dependencies, obtain the new xCAT and xCAT dependencies rpms.
(Follow the same steps that were followed in
:ref:`setup_service_node_stateful_label`.

221
docs/source/advanced/hierarchy/service_node_for_diskless.rst Normal file
View File

@ -0,0 +1,221 @@
.. _setup_service_node_stateless_label:
Setup the Service Node for Stateless Deployment (optional)
==========================================================
**Note: The stateless service node is not supported in ubuntu hierarchy
cluster. For ubuntu, please skip this section.**
If you want, your service nodes can be stateless (diskless). The service node
must contain not only the OS, but also the xCAT software and its dependencies.
In addition, a number of files are added to the service node to support the
PostgreSQL, or MySQL database access from the service node to the Management
node, and ssh access to the nodes that the service nodes services.
The following sections explain how to accomplish this.
Build the Service Node Diksless Image
--------------------------------------
This section assumes you can build the stateless image on the management node
because the service nodes are the same OS and architecture as the management
node. If this is not the case, you need to build the image on a machine that
matches the service node's OS architecture.
* Create an osimage definition. When you run copycds, xCAT will create a
service node osimage definitions for that distribution. For a stateless
service node, use the *-netboot-service definition.
::
lsdef -t osimage | grep -i service
rhels6.4-ppc64-install-service (osimage)
rhels6.4-ppc64-netboot-service (osimage)
rhels6.4-ppc64-statelite-service (osimage)
lsdef -t osimage -l rhels6.3-ppc64-netboot-service
Object name: rhels6.3-ppc64-netboot-service
exlist=/opt/xcat/share/xcat/netboot/rh/service.exlist
imagetype=linux
osarch=ppc64
osdistroname=rhels6.3-ppc64
osname=Linux
osvers=rhels6.3
otherpkgdir=/install/post/otherpkgs/rhels6.3/ppc64
otherpkglist=/opt/xcat/share/xcat/netboot/rh/service.rhels6.ppc64.otherpkgs.pkglist
pkgdir=/install/rhels6.3/ppc64
pkglist=/opt/xcat/share/xcat/netboot/rh/service.rhels6.ppc64.pkglist
postinstall=/opt/xcat/share/xcat/netboot/rh/service.rhels6.ppc64.postinstall
profile=service
provmethod=netboot
rootimgdir=/install/netboot/rhels6.3/ppc64/service
* You can check the service node packaging to see if it has all the rpms you
require. We ship a basic requirements lists that will create a fully
functional service node. However, you may want to customize your service
node by adding additional operating system packages or modifying the files
excluded by the exclude list. View the files referenced by the osimage
pkglist, otherpkglist and exlist attributes:
::
cd /opt/xcat/share/xcat/netboot/rh/
view service.rhels6.ppc64.pkglist
view service.rhels6.ppc64.otherpkgs.pkglist
view service.exlist
If you would like to change any of these files, copy them to a custom
directory. This can be any directory you choose, but we recommend that you
keep it /install somewhere. A good location is something like
``/install/custom/netboot/<os>/service``. Make sure that your
``otherpkgs.pkglist`` file as an entry for
::
xcat/xcat-core/xCATsn
This is required to install the xCAT service node function into your image.
You may also choose to create an appropriate /etc/fstab file in your
service node image. Copy the script referenced by the postinstall
attribute to your directory and modify it as you would like:
::
cp /opt/xcat/share/xcat/netboot/rh/service.rhels6.ppc64.postinstall
/install/custom/netboot/rh
vi /install/custom/netboot/rh
# uncomment the sample fstab lines and change as needed:
proc /proc proc rw 0 0
sysfs /sys sysfs rw 0 0
devpts /dev/pts devpts rw,gid=5,mode=620 0 0
service_x86_64 / tmpfs rw 0 1
none /tmp tmpfs defaults,size=10m 0 2
none /var/tmp tmpfs defaults,size=10m 0 2
After modifying the files, you will need to update the osimage definition to
reference these files. We recommend creating a new osimage definition for
your custom image: ::
lsdef -t osimage -l rhels6.3-ppc64-netboot-service -z > /tmp/myservice.def
vi /tmp/myservice.def
# change the name of the osimage definition
# change any attributes that now need to reference your custom files
# change the rootimgdir attribute replacing 'service'
with a name to match your new osimage definition
cat /tmp/msyservice.def | mkdef -z
While you are here, if you'd like, you can do the same for your compute node
images, creating custom files and new custom osimage definitions as you need
to.
For more information on the use and syntax of otherpkgs and pkglist files,
see `Update Service Node Stateless Image <http://localhost/fake_todo>`_
* Make your xCAT software available for otherpkgs processing
* If you downloaded xCAT to your management node for installation, place a
copy of your xcat-core and xcat-dep in your otherpkgdir directory ::
lsdef -t osimage -o rhels6.3-ppc64-netboot-service -i otherpkgdir
Object name: rhels6.3-ppc64-netboot-service
otherpkgdir=/install/post/otherpkgs/rhels6.3/ppc64
cd /install/post/otherpkgs/rhels6.3/ppc64
mkdir xcat
cd xcat
cp -Rp <current location of xcat-core>/xcat-core
cp -Rp <current location of xcat-dep>/xcat-dep
* If you installed your management node directly from the Linux online
repository, you will need to download the xcat-core and xcat-dep tarballs
- Go to the `Download xCAT page <http://localhost/fake_todo>`_ and download
the level of xCAT tarball you desire.
- Go to the `Download xCAT Dependencies <http://localhost/fake_todo>`_ page
and download the latest xCAT dependency tarball. Place these into your
otherpkdir directory:
::
lsdef -t osimage -o rhels6.3-ppc64-netboot-service -i otherpkgdir
Object name: rhels6.3-ppc64-netboot-service
otherpkgdir=/install/post/otherpkgs/rhels6.3/ppc64
cd /install/post/otherpkgs/rhels6.3/ppc64
mkdir xcat
cd xcat
mv <xcat-core tarball> .
tar -jxvf <xcat-core tarball>
mv <xcat-dep tarball> .
tar -jxvf <xcat-dep tarball>
* Run image generation for your osimage definition:
::
genimage rhels6.3-ppc64-netboot-service
* Prevent DHCP from starting up until xcatd has had a chance to configure it:
::
chroot /install/netboot/rhels6.3/ppc64/service/rootimg chkconfig dhcpd off
chroot /install/netboot/rhels6.3/ppc64/service/rootimg chkconfig dhcrelay off
* IF using NFS hybrid mode, export /install read-only in service node image:
::
cd /install/netboot/rhels6.3/ppc64/service/rootimg/etc
echo '/install *(ro,no_root_squash,sync,fsid=13)' >exports
* Pack the image for your osimage definition:
::
packimage rhels6.3-ppc64-netboot-service
* Set the node status to ready for netboot using your osimage definition and
your 'service' nodegroup:
::
nodeset service osimage=rhels6.3-ppc64-netboot-service
* To diskless boot the service nodes
::
rnetboot service
Update Service Node Stateless Image
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
To update the xCAT software in the image at a later time:
* Download the updated xcat-core and xcat-dep tarballs and place them in
your osimage's otherpkgdir xcat directory as you did above.
* Generate and repack the image and reboot your service node.
* Run image generation for your osimage definition.
::
genimage rhels6.3-ppc64-netboot-service
packimage rhels6.3-ppc64-netboot-service
nodeset service osimage=rhels6.3-ppc64-netboot-service
rnetboot service
Note: The service nodes are set up as NFS-root servers for the compute nodes.
Any time changes are made to any compute image on the mgmt node it will be
necessary to sync all changes to all service nodes. In our case the
``/install`` directory is mounted on the servicenodes, so the update to the
compute node image is automatically available.
Monitor install and boot
------------------------
::
wcons service # make sure DISPLAY is set to your X server/VNC or
rcons <one-node-at-a-time> # or do rcons for each node
tail -f /var/log/messages

27
docs/source/advanced/hierarchy/setup_mn_hierachical_database.rst Normal file
View File

@ -0,0 +1,27 @@
Setup the MN Hierarchical Database
==================================
Before setting up service nodes, you need to set up either MySQL, PostgreSQL,
as the xCAT Database on the Management Node. The database client on the
Service Nodes will be set up later when the SNs are installed. MySQL and
PostgreSQL are available with the Linux OS.
Follow the instructions in one of these documents for setting up the
Management node to use the selected database:
MySQL or MariaDB
----------------
* Follow this documentation and be sure to use the xCAT provided mysqlsetup
command to setup the database for xCAT:
.. TODO http link
- `Setting_Up_MySQL_as_the_xCAT_DB <http://localhost/fake_todo>`_
PostgreSQL:
-----------
* Follow this documentation and be sure and use the xCAT provided pgsqlsetup
command to setup the database for xCAT:
.. TODO http link
- `Setting_Up_PostgreSQL_as_the_xCAT_DB <http://localhost/fake_todo>`_

27
docs/source/advanced/hierarchy/setup_mn_hierarchical_database.rst Normal file
View File

@ -0,0 +1,27 @@
Setup the MN Hierarchical Database
==================================
Before setting up service nodes, you need to set up either MySQL, PostgreSQL,
as the xCAT Database on the Management Node. The database client on the
Service Nodes will be set up later when the SNs are installed. MySQL and
PostgreSQL are available with the Linux OS.
Follow the instructions in one of these documents for setting up the
Management node to use the selected database:
MySQL or MariaDB
----------------
* Follow this documentation and be sure to use the xCAT provided mysqlsetup
command to setup the database for xCAT:
.. TODO http link
- `Setting_Up_MySQL_as_the_xCAT_DB <http://localhost/fake_todo>`_
PostgreSQL:
-----------
* Follow this documentation and be sure and use the xCAT provided pgsqlsetup
command to setup the database for xCAT:
.. TODO http link
- `Setting_Up_PostgreSQL_as_the_xCAT_DB <http://localhost/fake_todo>`_

6
docs/source/advanced/hierarchy/setup_service_node.rst Normal file
View File

@ -0,0 +1,6 @@
Setup Service Node
==================
* Follow this documentation to :ref:`setup_service_node_stateful_label`.
* Follow this documentation to :ref:`setup_service_node_stateless_label`.

11
docs/source/advanced/hierarchy/test_service_node_installation.rst Normal file
View File

@ -0,0 +1,11 @@
Test Service Node installation
==============================
* ssh to the service nodes. You should not be prompted for a password.
* Check to see that the xcat daemon xcatd is running.
* Run some database command on the service node, e.g tabdump site, or nodels,
and see that the database can be accessed from the service node.
* Check that ``/install`` and ``/tftpboot`` are mounted on the service node
from the Management Node, if appropriate.
* Make sure that the Service Node has Name resolution for all nodes, it will
service.