# IBM(c) 2007 EPL license http://www.eclipse.org/legal/epl-v10.html package xCAT::Schema; BEGIN { $::XCATROOT = $ENV{'XCATROOT'} ? $ENV{'XCATROOT'} : -d '/opt/xcat' ? '/opt/xcat' : '/usr'; } use lib "$::XCATROOT/lib/perl"; use xCAT::ExtTab; # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! # # When making additions or deletions to this file please be sure to # modify BOTH the tabspec and defspec definitions. This includes # adding descriptions for any new attributes. # # Make sure any new attributes are not SQL reserved words by checking # on this site:http://www.petefreitag.com/tools/sql_reserved_words_checker/ # For Postgresql: check the following site for names that cannot be used # as attributes under any conditions: # http://www.postgresql.org/docs/8.3/static/sql-keywords-appendix.html # # Current SQL reserved words being used in this Schema with special # processing are the # following: # #Word Table Databases that will not allow # key site,passwd,prodkey,monsetting MySQL, DB2,SQL Server 2000 # dump nimimage SQL Server 2000 (microsoft) # power nodehm SQL Server 2000 # host policy,ivm SQL Server Future Keywords # parameters policy DB2,SQL Server Future Keywords,ISO/ANSI,SQL99 # time policy DB2,SQL Server Future Keywords,ISO/ANSI,SQL99 # rule policy SQL Server 2000 # value site,monsetting ODBC, DB2, SQL Server # Future Keywords,ISO/ANSI,SQL99 # # !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! #Note that the SQL is far from imaginative. Fact of the matter is that #certain SQL backends don't ascribe meaning to the data types anyway. #New format, not sql statements, but info enough to describe xcat tables %tabspec = ( statelite => { cols => [qw(node image statemnt mntopts comments disable)], keys => [qw(node)], required => [qw(node statemnt)], table_desc => 'The location on an NFS server where a nodes persistent files are stored. Any file marked persistent in the litefile table will be stored in the location specified in this table for that node.', descriptions => { node => 'The name of the node or group that will use this location.', image => "Reserved for future development, not used. ", statemnt => "The persistant read/write area where a node's persistent files will be written to, e.g: 10.0.0.1/state/. The node name will be automatically added to the pathname, so 10.0.0.1:/state, will become 10.0.0.1:/state/.", mntopts => "A comma-separated list of options to use when mounting the persistent directory. (Ex. 'soft') The default is to do a 'hard' mount. ", comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, #If we support multiple domains managed in a single xCAT instance, the following #tables come into play. Given no explicit request to span domains and no effort to #seriously evaluate wider support of multi-domain environments, will leave them #commented rather than tempt people to try with an expectation that it could work. domain => { cols => [qw(node ou comments disable)], keys => ['node'], table_desc => 'Mapping of nodes to domain attributes', descriptions => { node => 'The node or group the entry applies to', # domain => 'The name of the domain it is a member of, such as "example.com". Defaults to domain value from the site table', # the above column is unimplemented by anything, so leave it out for this pass ou => 'For an LDAP described machine account (i.e. Active Directory), the orginaztional unit to place the system. If not set, defaults to cn=Computers,dc=your,dc=domain', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, ############################################################################### # The next two are for kvm plugin to use to maintain persistent config data # not feasibly determined from contextual data ############################################################################### kvm_nodedata => { cols => [qw(node xml comments disable)], keys => [qw(node)], required => [qw(node)], table_desc => 'Persistant store for KVM plugin, not intended for manual modification.', types => { xml => 'VARCHAR(16000)', }, descriptions => { node => 'The node corresponding to the virtual machine', xml => 'The XML description generated by xCAT, fleshed out by libvirt, and stored for reuse', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, kvm_masterdata => { cols => [qw(name xml comments disable)], keys => [qw(name)], nodecol => 'name', table_desc=>'Persistant store for KVM plugin for masters', types => { xml => 'VARCHAR(16000)', }, descriptions => { name => 'The name of the relevant master', xml => 'The XML description to be customized for clones of this master', disable => "Set to 'yes' or '1' to comment out this row.", }, }, #domains => { # cols => [qw(domain nameserver authserver realm comments disable)], # keys => ['domain'], # table_desc => 'Parameters concerning domain-wide management', # descriptions => { # domain => 'The name of the domain, such as "example.com"', # nameserver => 'The address of the server that is responsible for updating DNS records', # authserver => 'The provider of authentication and authorization data, which may be a generic LDAP server as well as a Kerberos KDC, or specifically an active directory domain controller'. # realm => 'The kerberos realm name associated with the domain. Defaults to uppercase of the domain name', # comments => 'Any user-written notes.', # disable => "Set to 'yes' or '1' to comment out this row.", # }, #}, litetree => { cols => [qw(priority image directory comments disable)], keys => [qw(priority)], required => [qw(priority directory)], table_desc => 'Directory hierarchy to traverse to get the initial contents of node files. The files that are specified in the litefile table are searched for in the directories specified in this table.', descriptions => { priority => 'This number controls what order the directories are searched. Directories are searched from smallest priority number to largest.', image => "The name of the image that will use this directory, as specified in the osimage table. If image is not supplied, the default is 'ALL'. 'ALL' means use it for all images.", directory => 'The location (hostname:path) of a directory that contains files specified in the litefile table. Variables are allowed. E.g: $noderes.nfsserver://xcatmasternode/install/$node/#CMD=uname-r#/', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, litefile => { cols => [qw(image file options comments disable)], keys => [qw(image file)], required => [qw(file)], # default type is rw nfsroot table_desc => 'The litefile table specifies the directories and files on the statelite nodes that should be readwrite, persistent, or readonly overlay. All other files in the statelite nodes come from the readonly statelite image.', descriptions => { image => "The name of the image that will use these files, as specified in the osimage table. 'ALL' means use it for all images.", file => "The full pathname of the file. e.g: /etc/hosts. If the path is a directory, then it should be terminated with a '/'. ", options => "Options for the file:\n\n". qq{ tmpfs - It is the default option if you leave the options column blank. It provides a file or directory for the node to use when booting, its permission will be the same as the original version on the server. In most cases, it is read-write; however, on the next statelite boot, the original version of the file or directory on the server will be used, it means it is non-persistent. This option can be performed on files and directories..\n\n}. qq{ rw - Same as Above.Its name "rw" does NOT mean it always be read-write, even in most cases it is read-write. Please do not confuse it with the "rw" permission in the file system. \n\n}. qq{ persistent - It provides a mounted file or directory that is copied to the xCAT persistent location and then over-mounted on the local file or directory. Anything written to that file or directory is preserved. It means, if the file/directory does not exist at first, it will be copied to the persistent location. Next time the file/directory in the persistent location will be used. The file/directory will be persistent across reboots. Its permission will be the same as the original one in the statelite location. It requires the statelite table to be filled out with a spot for persistent statelite. This option can be performed on files and directories. \n\n}. qq{ con - The contents of the pathname are concatenated to the contents of the existing file. For this directive the searching in the litetree hierarchy does not stop when the first match is found. All files found in the hierarchy will be concatenated to the file when found. The permission of the file will be "-rw-r--r--", which means it is read-write for the root user, but readonly for the others. It is non-persistent, when the node reboots, all changes to the file will be lost. It can only be performed on files. Please do not use it for one directory.\n\n}. qq{ ro - The file/directory will be overmounted read-only on the local file/directory. It will be located in the directory hierarchy specified in the litetree table. Changes made to this file or directory on the server will be immediately seen in this file/directory on the node. This option requires that the file/directory to be mounted must be available in one of the entries in the litetree table. This option can be performed on files and directories.\n\n}. qq{ link - It provides one file/directory for the node to use when booting, it is copied from the server, and will be placed in tmpfs on the booted node. In the local file system of the booted node, it is one symbolic link to one file/directory in tmpfs. And the permission of the symbolic link is "lrwxrwxrwx", which is not the real permission of the file/directory on the node. So for some application sensitive to file permissions, it will be one issue to use "link" as its option, for example, "/root/.ssh/", which is used for SSH, should NOT use "link" as its option. It is non-persistent, when the node is rebooted, all changes to the file/directory will be lost. This option can be performed on files and directories. \n\n}. qq{ link,con - It works similar to the "con" option. All the files found in the litetree hierarchy will be concatenated to the file when found. The final file will be put to the tmpfs on the booted node. In the local file system of the booted node, it is one symbolic link to the file/directory in tmpfs. It is non-persistent, when the node is rebooted, all changes to the file will be lost. The option can only be performed on files. \n\n}. qq{ link,persistent - It provides a mounted file or directory that is copied to the xCAT persistent location and then over-mounted to the tmpfs on the booted node, and finally the symbolic link in the local file system will be linked to the over-mounted tmpfs file/directory on the booted node. The file/directory will be persistent across reboots. The permission of the file/directory where the symbolic link points to will be the same as the original one in the statelite location. It requires the statelite table to be filled out with a spot for persistent statelite. The option can be performed on files and directories.\n\n}. qq{ link,ro - The file is readonly, and will be placed in tmpfs on the booted node. In the local file system of the booted node, it is one symbolic link to the tmpfs. It is non-persistent, when the node is rebooted, all changes to the file/directory will be lost. This option requires that the file/directory to be mounted must be available in one of the entries in the litetree table. The option can be performed on files and directories.}, comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", } }, vmmaster => { #will add columns as approriate, for now: #os arch profile to populate the corresponding nodetype fields of a cloned vm #storage to indicate where the master data is actually stored (i.e. virtual disk images) #storagemodel to allow chvm on a clone to be consistent with the master by default #nics to track the network mapping that may not be preserved by the respective plugin's specific cfg info #nicmodel same as storagemodel, except omitting for now until chvm actually does nics... cols => [qw(name os arch profile storage storagemodel nics vintage originator virttype comments disable)], keys => [qw(name)], nodecol => 'name', #well what do you know, I used it... table_desc => 'Inventory of virtualization images for use with clonevm. Manual intervention in this table is not intended.', descriptions => { 'name' => 'The name of a master', 'os' => 'The value of nodetype.os at the time the master was captured', 'arch' => 'The value of nodetype.arch at the time of capture', 'profile' => 'The value of nodetype.profile at time of capture', 'storage' => 'The storage location of bulk master information', 'storagemodel' => 'The default storage style to use when modifying a vm cloned from this master', 'nics' => 'The nic configuration and relationship to vlans/bonds/etc', 'vintage' => "When this image was created", 'originator' => 'The user who created the image', 'virttype' => 'The type of virtualization this image pertains to (e.g. vmware, kvm, etc)', } }, vm => { cols => [qw(node host migrationdest storage storagemodel cfgstore memory cpus nics nicmodel bootorder clockoffset virtflags master vncport textconsole powerstate beacon datacenter guestostype othersettings vidmodel vidproto vidpassword comments disable)], keys => [qw(node)], table_desc => 'Virtualization parameters', descriptions => { 'node' => 'The node or static group name', 'host' => 'The system that currently hosts the VM', 'migrationdest' => 'A noderange representing candidate destinations for migration (i.e. similar systems, same SAN, or other criteria that xCAT can use', 'storage' => 'A list of storage files or devices to be used. i.e. /cluster/vm/ or nfs:///path/to/folder/', 'storagemodel' => 'Model of storage devices to provide to guest', 'cfgstore' => 'Optional location for persistant storage separate of emulated hard drives for virtualization solutions that require persistant store to place configuration data', 'memory' => 'Megabytes of memory the VM currently should be set to.', 'master' => 'The name of a master image, if any, this virtual machine is linked to. This is generally set by clonevm and indicates the deletion of a master that would invalidate the storage of this virtual machine', 'cpus' => 'Number of CPUs the node should see.', 'nics' => 'Network configuration parameters. Of the general form [physnet:]interface,.. Generally, interface describes the vlan entity (default for native, tagged for tagged, vl[number] for a specific vlan. physnet is a virtual switch name or port description that is used for some virtualization technologies to construct virtual switches. hypervisor.netmap can map names to hypervisor specific layouts, or the descriptions described there may be used directly here where possible.', 'nicmodel' => 'Model of NICs that will be provided to VMs (i.e. e1000, rtl8139, virtio, etc)', 'bootorder' => 'Boot sequence (i.e. net,hd)', 'clockoffset' => 'Whether to have guest RTC synced to "localtime" or "utc" If not populated, xCAT will guess based on the nodetype.os contents.', 'virtflags' => 'General flags used by the virtualization method. For example, in Xen it could, among other things, specify paravirtualized setup, or direct kernel boot. For a hypervisor/dom0 entry, it is the virtualization method (i.e. "xen"). For KVM, the following flag=value pairs are recognized: imageformat=[raw|fullraw|qcow2] raw is a generic sparse file that allocates storage on demand fullraw is a generic, non-sparse file that preallocates all space qcow2 is a sparse, copy-on-write capable format implemented at the virtualization layer rather than the filesystem level clonemethod=[qemu-img|reflink] qemu-img allows use of qcow2 to generate virtualization layer copy-on-write reflink uses a generic filesystem facility to clone the files on your behalf, but requires filesystem support such as btrfs ', 'vncport' => 'Tracks the current VNC display port (currently not meant to be set', 'textconsole' => 'Tracks the Psuedo-TTY that maps to the serial port or console of a VM', 'powerstate' => "This flag is used by xCAT to track the last known power state of the VM.", 'othersettings' => "This allows specifying a semicolon delimited list of key->value pairs to include in a vmx file of VMware.", 'guestostype' => "This allows administrator to specify an identifier for OS to pass through to virtualization stack. Normally this should be ignored as xCAT will translate from nodetype.os rather than requiring this field be used\n", 'beacon' => "This flag is used by xCAT to track the state of the identify LED with respect to the VM.", 'datacenter' => "Optionally specify a datacenter for the VM to exist in (only applicable to VMWare)", 'vidproto' => "Request a specific protocol for remote video access be set up. For example, spice in KVM.", 'vidmodel' => "Model of video adapter to provide to guest. For example, qxl in KVM", 'vidpassword' => "Password to use instead of temporary random tokens for VNC and SPICE access", } }, hypervisor => { cols => [qw(node type mgr netmap defaultnet cluster datacenter preferdirect comments disable)], keys => [qw(node)], table_desc => 'Hypervisor parameters', descriptions => { 'node' => 'The node or static group name', 'type' => 'The plugin associated with hypervisor specific commands such as revacuate', mgr => 'The virtualization specific manager of this hypervisor when applicable', 'netmap' => 'Optional mapping of useful names to relevant physical ports. For example, 10ge=vmnic_16.0&vmnic_16.1,ge=vmnic1 would be requesting two virtual switches to be created, one called 10ge with vmnic_16.0 and vmnic_16.1 bonded, and another simply connected to vmnic1. Use of this allows abstracting guests from network differences amongst hypervisors', 'defaultnet' => 'Optionally specify a default network entity for guests to join to if they do not specify.', 'cluster' => 'Specify to the underlying virtualization infrastructure a cluster membership for the hypervisor.', 'datacenter' => 'Optionally specify a datacenter for the hypervisor to exist in (only applicable to VMWare)', 'preferdirect' => 'If a mgr is declared for a hypervisor, xCAT will default to using the mgr for all operations. If this is field is set to yes or 1, xCAT will prefer to directly communicate with the hypervisor if possible' } }, websrv => { cols => [qw(node port username password comments disable)], keys => [qw(node)], table_desc => 'Web service parameters', descriptions => { 'node' => 'The web service hostname.', 'port' => 'The port of the web service.', 'username' => 'Userid to use to access the web service.', 'password' => 'Password to use to access the web service.', 'comments' => 'Any user-written notes.', 'disable' => "Set to 'yes' or '1' to comment out this row.", }, }, boottarget => { cols => [qw(bprofile kernel initrd kcmdline comments disable)], keys => [qw(bprofile)], table_desc => 'Specify non-standard initrd, kernel, and parameters that should be used for a given profile.', descriptions => { 'bprofile' => 'All nodes with a nodetype.profile value equal to this value and nodetype.os set to "boottarget", will use the associated kernel, initrd, and kcmdline.', 'kernel' => 'The kernel that network boot actions should currently acquire and use. Note this could be a chained boot loader such as memdisk or a non-linux boot loader', 'initrd' => 'The initial ramdisk image that network boot actions should use (could be a DOS floppy or hard drive image if using memdisk as kernel)', 'kcmdline' => 'Arguments to be passed to the kernel', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row." } }, bootparams => { cols => [qw(node kernel initrd kcmdline addkcmdline dhcpstatements adddhcpstatements comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => 'Current boot settings to be sent to systems attempting network boot for deployment, stateless, or other reasons. Mostly automatically manipulated by xCAT.', descriptions => { 'node' => 'The node or group name', 'kernel' => 'The kernel that network boot actions should currently acquire and use. Note this could be a chained boot loader such as memdisk or a non-linux boot loader', 'initrd' => 'The initial ramdisk image that network boot actions should use (could be a DOS floppy or hard drive image if using memdisk as kernel)', 'kcmdline' => 'Arguments to be passed to the kernel', 'addkcmdline' => 'User specified one or more parameters to be passed to the kernel', 'dhcpstatements' => 'xCAT manipulated custom dhcp statements (not intended for user manipulation)', 'adddhcpstatements' => 'Custom dhcp statements for administrator use (not implemneted yet)', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", } }, prodkey => { cols => [qw(node product key comments disable)], keys => [qw(node product)], table_desc => 'Specify product keys for products that require them', descriptions => { node => "The node name or group name.", product => "A string to identify the product (for OSes, the osname would be used, i.e. wink28", key => "The product key relevant to the aforementioned node/group and product combination", comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", } }, chain => { cols => [qw(node currstate currchain chain ondiscover comments disable)], keys => [qw(node)], table_desc => 'Controls what operations are done (and it what order) when a node is discovered and deployed.', descriptions => { node => 'The node name or group name.', currstate => 'The current chain state for this node. Set by xCAT.', currchain => 'The current execution chain for this node. Set by xCAT. Initialized from chain and updated as chain is executed.', chain => 'A comma-delimited chain of actions to be performed automatically for this node. Valid values: discover, boot or reboot, install or netboot, runcmd=, runimage=, shell, standby. (Default - same as no chain). Example, for BMC machines use: runcmd=bmcsetup,standby.', ondiscover => 'What to do when a new node is discovered. Valid values: nodediscover.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, deps => { cols => [qw(node nodedep msdelay cmd comments disable)], keys => [qw(node cmd)], required => [qw(node cmd)], table_desc => 'Describes dependencies some nodes have on others. This can be used, e.g., by rpower -d to power nodes on or off in the correct order.', descriptions => { node => 'The node name or group name.', nodedep => 'Comma-separated list of nodes or node groups it is dependent on.', msdelay => 'How long to wait between operating on the dependent nodes and the primary nodes.', cmd => 'Comma-seperated list of which operation this dependency applies to.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, hosts => { cols => [qw(node ip hostnames otherinterfaces comments disable)], keys => [qw(node)], table_desc => 'IP address and hostnames of nodes. This info can be used to populate /etc/hosts or DNS.', descriptions => { node => 'The node name or group name.', ip => 'The IP address of the node.', hostnames => 'Hostname aliases added to /etc/hosts for this node.', otherinterfaces => 'Other IP addresses to add for this node. Format: -:,:,...', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, ipmi => { cols => [qw(node bmc bmcport username password comments disable )], keys => [qw(node)], table_desc => 'Settings for nodes that are controlled by an on-board BMC via IPMI.', descriptions => { node => 'The node name or group name.', bmc => 'The hostname of the BMC adapater.', bmcport => 'In systems with selectable shared/dedicated ethernet ports, this parameter can be used to specify the preferred port. 0 means use the shared port, 1 means dedicated, blank is to not assign', username => 'The BMC userid. If not specified, the key=ipmi row in the passwd table is used as the default.', password => 'The BMC password. If not specified, the key=ipmi row in the passwd table is used as the default.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, iscsi => { cols => [qw(node server target lun iname file userid passwd kernel kcmdline initrd comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => 'Contains settings that control how to boot a node from an iSCSI target', descriptions => { node => 'The node name or group name.', server => 'The server containing the iscsi boot device for this node.', target => 'The iscsi disk used for the boot device for this node. Filled in by xCAT.', lun => 'LUN of boot device. Per RFC-4173, this is presumed to be 0 if unset. tgtd often requires this to be 1', iname => 'Initiator name. Currently unused.', file => 'The path on the server of the OS image the node should boot from.', userid => 'The userid of the iscsi server containing the boot device for this node.', passwd => 'The password for the iscsi server containing the boot device for this node.', kernel => 'The path of the linux kernel to boot from.', kcmdline => 'The kernel command line to use with iSCSI for this node.', initrd => 'The initial ramdisk to use when network booting this node.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, mac => { cols => [qw(node interface mac comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => "The MAC address of the node's install adapter. Normally this table is populated by getmacs or node discovery, but you can also add entries to it manually.", descriptions => { node => 'The node name or group name.', interface => 'The adapter interface name that will be used to install and manage the node. E.g. eth0 (for linux) or en0 (for AIX).)', mac => 'The mac address or addresses for which xCAT will manage static bindings for this node. This may be simply a mac address, which would be bound to the node name (such as "01:02:03:04:05:0E"). This may also be a "|" delimited string of "mac address!hostname" format (such as "01:02:03:04:05:0E!node5|01:02:03:05:0F!node6-eth1").', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, monitoring => { cols => [qw(name nodestatmon comments disable)], keys => [qw(name)], required => [qw(name)], table_desc => 'Controls what external monitoring tools xCAT sets up and uses. Entries should be added and removed from this table using the provided xCAT commands monstart and monstop.', descriptions => { name => "The name of the mornitoring plug-in module. The plug-in must be put in $ENV{XCATROOT}/lib/perl/xCAT_monitoring/. See the man page for monstart for details.", nodestatmon => 'Specifies if the monitoring plug-in is used to feed the node status to the xCAT cluster. Any one of the following values indicates "yes": y, Y, yes, Yes, YES, 1. Any other value or blank (default), indicates "no".', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, monsetting => { cols => [qw(name key value comments disable)], keys => [qw(name key)], required => [qw(name key)], table_desc => 'Specifies the monitoring plug-in specific settings. These settings will be used by the monitoring plug-in to customize the behavior such as event filter, sample interval, responses etc. Entries should be added, removed or modified by chtab command. Entries can also be added or modified by the monstart command when a monitoring plug-in is brought up.', descriptions => { name => "The name of the mornitoring plug-in module. The plug-in must be put in $ENV{XCATROOT}/lib/perl/xCAT_monitoring/. See the man page for monstart for details.", key => 'Specifies the name of the attribute. The valid values are specified by each monitoring plug-in. Use "monls name -d" to get a list of valid keys.', value => 'Specifies the value of the attribute.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, mp => { cols => [qw(node mpa id nodetype comments disable)], keys => [qw(node)], table_desc => 'Contains the hardware control info specific to blades. This table also refers to the mpa table, which contains info about each Management Module.', descriptions => { node => 'The blade node name or group name.', mpa => 'The managment module used to control this blade.', id => 'The slot number of this blade in the BladeCenter chassis.', nodetype => 'The hardware type for mp node. Valid values: mm,cmm, blade.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, mpa => { cols => [qw(mpa username password comments disable)], keys => [qw(mpa)], nodecol => "mpa", table_desc => 'Contains info about each Management Module and how to access it.', descriptions => { mpa => 'Hostname of the management module.', username => 'Userid to use to access the management module. If not specified, the key=blade row in the passwd table is used as the default.', password => 'Password to use to access the management module. If not specified, the key=blade row in the passwd table is used as the default.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, networks => { cols => [qw(netname net mask mgtifname gateway dhcpserver tftpserver nameservers ntpservers logservers dynamicrange nodehostname ddnsdomain vlanid domain comments disable)], keys => [qw(net mask)], table_desc => 'Describes the networks in the cluster and info necessary to set up nodes on that network.', descriptions => { netname => 'Name used to identify this network definition.', net => 'The network address.', mask => 'The network mask.', mgtifname => 'The interface name of the management/service node facing this network. !remote! indicates a non-local network for relay DHCP.', gateway => 'The network gateway. It can be set to an ip address or the keyword , the keyword indicates the cluster-facing ip address configured on this management node or service node. Leaving this field blank means that there is no gateway for this network.', dhcpserver => 'The DHCP server that is servicing this network. Required to be explicitly set for pooled service node operation.', tftpserver => 'The TFTP server that is servicing this network. If not set, the DHCP server is assumed.', nameservers => 'A comma delimited list of DNS servers that each node in this network should use. This value will end up in the nameserver settings of the /etc/resolv.conf on each node in this network. If this attribute value is set to the IP address of an xCAT node, make sure DNS is running on it. In a hierarchical cluster, you can also set this attribute to "" to mean the DNS server for each node in this network should be the node that is managing it (either its service node or the management node). Used in creating the DHCP network definition, and DNS configuration.', ntpservers => 'The ntp servers for this network. Used in creating the DHCP network definition. Assumed to be the DHCP server if not set.', logservers => 'The log servers for this network. Used in creating the DHCP network definition. Assumed to be the DHCP server if not set.', dynamicrange => 'The IP address range used by DHCP to assign dynamic IP addresses for requests on this network. This should not overlap with entities expected to be configured with static host declarations, i.e. anything ever expected to be a node with an address registered in the mac table.', nodehostname => 'A regular expression used to specify node name to network-specific hostname. i.e. "/\z/-secondary/" would mean that the hostname of "n1" would be n1-secondary on this network. By default, the nodename is assumed to equal the hostname, followed by nodename-interfacename.', ddnsdomain => 'A domain to be combined with nodename to construct FQDN for DDNS updates induced by DHCP. This is not passed down to the client as "domain"', vlanid => 'The vlan ID if this network is within a vlan.', domain => 'The DNS domain name (ex. cluster.com).', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, nodegroup => { cols => [qw(groupname grouptype members wherevals comments disable)], keys => [qw(groupname)], table_desc => 'Contains group definitions, whose membership is dynamic depending on characteristics of the node.', descriptions => { groupname => 'Name of the group.', grouptype => 'The only current valid value is dynamic. We will be looking at having the object def commands working with static group definitions in the nodelist table.', members => 'The value of the attribute is not used, but the attribute is necessary as a place holder for the object def commands. (The membership for static groups is stored in the nodelist table.)', wherevals => 'A list of "attr*val" pairs that can be used to determine the members of a dynamic group, the delimiter is "::" and the operator * can be ==, =~, != or !~.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, nodehm => { cols => [qw(node power mgt cons termserver termport conserver serialport serialspeed serialflow getmac comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => "Settings that control how each node's hardware is managed. Typically, an additional table that is specific to the hardware type of the node contains additional info. E.g. the ipmi, mp, and ppc tables.", descriptions => { node => 'The node name or group name.', power => 'The method to use to control the power of the node. If not set, the mgt attribute will be used. Valid values: ipmi, blade, hmc, ivm, fsp. If "ipmi", xCAT will search for this node in the ipmi table for more info. If "blade", xCAT will search for this node in the mp table. If "hmc", "ivm", or "fsp", xCAT will search for this node in the ppc table.', mgt => 'The method to use to do general hardware management of the node. This attribute is used as the default if power or getmac is not set. Valid values: ipmi, blade, hmc, ivm, fsp, bpa. See the power attribute for more details.', cons => 'The console method. If nodehm.serialport is set, this will default to the nodehm.mgt setting, otherwise it defaults to unused. Valid values: cyclades, mrv, or the values valid for the mgt attribute.', termserver => 'The hostname of the terminal server.', termport => 'The port number on the terminal server that this node is connected to.', conserver => 'The hostname of the machine where the conserver daemon is running. If not set, the default is the xCAT management node.', serialport => 'The serial port for this node, in the linux numbering style (0=COM1/ttyS0, 1=COM2/ttyS1). For SOL on IBM blades, this is typically 1. For rackmount IBM servers, this is typically 0.', serialspeed => 'The speed of the serial port for this node. For SOL this is typically 19200.', serialflow => "The flow control value of the serial port for this node. For SOL this is typically 'hard'.", getmac => 'The method to use to get MAC address of the node with the getmac command. If not set, the mgt attribute will be used. Valid values: same as values for mgmt attribute.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, nodelist => { cols => [qw(node groups status statustime appstatus appstatustime primarysn hidden comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => "The list of all the nodes in the cluster, including each node's current status and what groups it is in.", descriptions => { node => 'The hostname of a node in the cluster.', groups => "A comma-delimited list of groups this node is a member of. Group names are arbitrary, except all nodes should be part of the 'all' group.", status => 'The current status of this node. This attribute will be set by xCAT software. Valid values: defined, booting, netbooting, booted, discovering, configuring, installing, alive, standingby, powering-off, unreachable. If blank, defined is assumed. The possible status change sequenses are: For installaton: defined->[discovering]->[configuring]->[standingby]->installing->booting->booted->[alive], For diskless deployment: defined->[discovering]->[configuring]->[standingby]->netbooting->booted->[alive], For booting: [alive/unreachable]->booting->[alive], For powering off: [alive]->powering-off->[unreachable], For monitoring: alive->unreachable. Discovering and configuring are for x Series dicovery process. Alive and unreachable are set only when there is a monitoring plug-in start monitor the node status for xCAT. Please note that the status values will not reflect the real node status if you change the state of the node from outside of xCAT (i.e. power off the node using HMC GUI).', statustime => "The data and time when the status was updated.", appstatus => "A comma-delimited list of application status. For example: 'sshd=up,ftp=down,ll=down'", appstatustime =>'The date and time when appstatus was updated.', primarysn => 'Not used currently. The primary servicenode, used by this node.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", hidden => "Used to hide fsp and bpa definitions, 1 means not show them when running lsdef and nodels", }, }, nodepos => { cols => [qw(node rack u chassis slot room comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => 'Contains info about the physical location of each node. Currently, this info is not used by xCAT, and therefore can be in whatevery format you want. It will likely be used in xCAT in the future.', descriptions => { node => 'The node name or group name.', rack => 'The frame the node is in.', u => 'The vertical position of the node in the frame', chassis => 'The BladeCenter chassis the blade is in.', slot => 'The slot number of the blade in the chassis.', room => 'The room the node is in.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, noderes => { cols => [qw(node servicenode netboot tftpserver tftpdir nfsserver monserver nfsdir installnic primarynic discoverynics cmdinterface xcatmaster current_osimage next_osimage nimserver routenames comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => 'Resources and settings to use when installing nodes.', descriptions => { node => 'The node name or group name.', servicenode => 'A comma separated list of node names (as known by the management node) that provides most services for this node. The first service node on the list that is accessible will be used. The 2nd node on the list is generally considered to be the backup service node for this node when running commands like snmove.', netboot => 'The type of network booting to use for this node. Valid values: pxe or xnba for x86* architecture, yaboot for POWER architecture.', tftpserver => 'The TFTP server for this node (as known by this node).', tftpdir => 'The directory that roots this nodes contents from a tftp and related perspective. Used for NAS offload by using different mountpoints.', nfsserver => 'The NFS or HTTP server for this node (as known by this node).', monserver => 'The monitoring aggregation point for this node. The format is "x,y" where x is the ip address as known by the management node and y is the ip address as known by the node.', nfsdir => 'The path that should be mounted from the NFS server.', installnic => 'The network adapter on the node that will be used for OS deployment, the installnic can be set to the network adapter name or the mac address or the keyword "mac" which means that the network interface specified by the mac address in the mac table will be used. If not set, primarynic will be used.', primarynic => 'The network adapter on the node that will be used for xCAT management, the primarynic can be set to the network adapter name or the mac address or the keyword "mac" which means that the network interface specified by the mac address in the mac table will be used. Default is eth0.', discoverynics => 'If specified, force discovery to occur on specific network adapters only, regardless of detected connectivity. Syntax can be simply "eth2,eth3" to restrict discovery to whatever happens to come up as eth2 and eth3, or by driver name such as "bnx2:0,bnx2:1" to specify the first two adapters managed by the bnx2 driver', cmdinterface => 'Not currently used.', #defnetname => 'The host (or ip) by which a node should be addressed (i.e. in psh/pscp). By default, nodename is assumed to be equal to this', xcatmaster => 'The hostname of the xCAT service node (as known by this node). This acts as the default value for nfsserver and tftpserver, if they are not set. If xcatmaster is not set, the node will use whoever responds to its boot request as its master. For the directed bootp case for POWER, it will use the management node if xcatmaster is not set.', current_osimage => 'Not currently used. The name of the osimage data object that represents the OS image currently deployed on this node.', next_osimage => 'Not currently used. The name of the osimage data object that represents the OS image that will be installed on the node the next time it is deployed.', nimserver => 'Not used for now. The NIM server for this node (as known by this node).', routenames => 'A comma separated list of route names that refer to rows in the routes table. These are the routes that should be defined on this node when it is deployed.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, switches => { cols => [qw(switch snmpversion username password privacy auth linkports sshusername sshpassword switchtype comments disable)], keys => [qw(switch)], nodecol => "switch", table_desc => 'Parameters to use when interrogating switches', descriptions => { switch => 'The hostname/address of the switch to which the settings apply', snmpversion => 'The version to use to communicate with switch. SNMPv1 is assumed by default.', username => 'The username to use for SNMPv3 communication, ignored for SNMPv1', password => 'The password or community string to use for SNMPv3 or SNMPv1 respectively. Falls back to passwd table, and site snmpc value if using SNMPv1', privacy => 'The privacy protocol to use for v3. DES is assumed if v3 enabled, as it is the most readily available.', auth => 'The authentication protocol to use for SNMPv3. SHA is assumed if v3 enabled and this is unspecified', linkports => 'The ports that connect to other switches. Currently, this column is only used by vlan configuration. The format is: "port_number:switch,port_number:switch...". Please refer to the switch table for details on how to specify the port numbers.', sshusername => 'The user name for ssh.', sshpassword => 'The password for ssh.', switchtype => 'The type of switch. It is used to identify the file name that implements the functions for this swithc. The valid values are: MellanoxIB etc.', }, }, nodetype => { cols => [qw(node os arch profile provmethod supportedarchs nodetype comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => 'A few hardware and software characteristics of the nodes.', descriptions => { node => 'The node name or group name.', os => 'The operating system deployed on this node. Valid values: AIX, rhels*,rhelc*, rhas*,centos*,SL*, fedora*, sles* (where * is the version #). As a special case, if this is set to "boottarget", then it will use the initrd/kernel/parameters specified in the row in the boottarget table in which boottarget.bprofile equals nodetype.profile.', arch => 'The hardware architecture of this node. Valid values: x86_64, ppc64, x86, ia64.', profile => 'The string to use to locate a kickstart or autoyast template to use for OS deployment of this node. If the provmethod attribute is set to an osimage name, that takes precendence. Otherwise, the os, profile, and arch are used to search for the files in /install/custom first, and then in /opt/xcat/share/xcat.', provmethod => 'The provisioning method for node deployment. The valid values are install, netboot, statelite or an os image name from the osimage table. If an image name is specified, the osimage definition stored in the osimage table and the linuximage table (for Linux) or nimimage table (for AIX) are used to locate the files for templates, pkglists, syncfiles, etc. On Linux, if install, netboot or statelite is specified, the os, profile, and arch are used to search for the files in /install/custom first, and then in /opt/xcat/share/xcat.', supportedarchs => 'Comma delimited list of architectures this node can execute.', nodetype => 'A comma-delimited list of characteristics of this node. Valid values: ppc, blade, vm (virtual machine), osi (OS image), mm, rsa, switch.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, notification => { cols => [qw(filename tables tableops comments disable)], keys => [qw(filename)], required => [qw(tables filename)], table_desc => 'Contains registrations to be notified when a table in the xCAT database changes. Users can add entries to have additional software notified of changes. Add and remove entries using the provided xCAT commands regnotif and unregnotif.', descriptions => { filename => 'The path name of a file that implements the callback routine when the monitored table changes. Can be a perl module or a command. See the regnotif man page for details.', tables => 'Comma-separated list of xCAT database tables to monitor.', tableops => 'Specifies the table operation to monitor for. Valid values: "d" (rows deleted), "a" (rows added), "u" (rows updated).', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, osimage => { cols => [qw(imagename profile imagetype provmethod rootfstype osname osvers osdistro osarch synclists postscripts postbootscripts comments disable)], keys => [qw(imagename)], table_desc => 'Basic information about an operating system image that can be used to deploy cluster nodes.', descriptions => { imagename => 'The name of this xCAT OS image definition.', imagetype => 'The type of operating system image this definition represents (linux,AIX).', provmethod => 'The provisioning method for node deployment. The valid values are install, netboot or statelite. It is not used by AIX.', rootfstype => 'The filesystem type for the rootfs is used when the provmethod is statelite. The valid values are nfs or ramdisk. The default value is nfs', profile => 'The node usage category. For example compute, service.', osname => 'Operating system name- AIX or Linux.', osvers => 'The Linux operating system deployed on this node. Valid values: rhels*,rhelc*, rhas*,centos*,SL*, fedora*, sles* (where * is the version #).', osdistro => 'Not used.', osarch => 'The hardware architecture of this node. Valid values: x86_64, ppc64, x86, ia64.', synclists => 'The fully qualified name of a file containing a list of files to synchronize on the nodes.', postscripts => 'Comma separated list of scripts that should be run on this image after diskfull installation or diskless boot. For installation of RedHat, CentOS, Fedora, the scripts will be run before the reboot. For installation of SLES, the scripts will be run after the reboot but before the init.d process. For diskless deployment, the scripts will be run at the init.d time, and xCAT will automatically add the list of scripts from the postbootscripts attribute to run after postscripts list. For installation of AIX, the scripts will run after the reboot and acts the same as the postbootscripts attribute. For AIX, use the postbootscripts attribute. Support will be added in the future for the postscripts attribute to run the scripts before the reboot in AIX. ', postbootscripts => 'Comma separated list of scripts that should be run on this after diskfull installation or diskless boot. On AIX these scripts are run during the processing of /etc/inittab. On Linux they are run at the init.d time. xCAT automatically adds the scripts in the xcatdefaults.postbootscripts attri bute to run first in the list.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, linuximage => { cols => [qw(imagename template pkglist pkgdir otherpkglist otherpkgdir exlist postinstall rootimgdir kerneldir nodebootif otherifce netdrivers kernelver krpmver permission dump crashkernelsize comments disable)], keys => [qw(imagename)], table_desc => 'Information about a Linux operating system image that can be used to deploy cluster nodes.', descriptions => { imagename => 'The name of this xCAT OS image definition.', template => 'The fully qualified name of the template file that is used to create the kick start file for diskful installation.', pkglist => 'The fully qualified name of the file that stores the distro packages list that will be included in the image.', pkgdir => 'The name of the directory where the distro packages are stored.', otherpkglist => 'The fully qualified name of the file that stores non-distro package lists that will be included in the image.', otherpkgdir => 'The base directory where the non-distro packages are stored.', exlist => 'The fully qualified name of the file that stores the file names and directory names that will be excluded from the image during packimage command. It is used for diskless image only.', postinstall => 'The fully qualified name of the script file that will be run at the end of the packimage command. It is used for diskless image only.', rootimgdir => 'The directory name where the image is stored. It is used for diskless image only.', kerneldir => 'The directory name where the 3rd-party kernel is stored. It is used for diskless image only.', nodebootif => 'The network interface the stateless/statelite node will boot over (e.g. eth0)', otherifce => 'Other network interfaces (e.g. eth1) in the image that should be configured via DHCP', netdrivers => 'The ethernet device drivers of the nodes which will use this linux image, at least the device driver for the nodes\' installnic should be included', kernelver => 'The version of linux kernel used in the linux image. If the kernel version is not set, the default kernel in rootimgdir will be used', krpmver => 'The rpm version of kernel packages (for SLES only). If it is not set, the default rpm version of kernel packages will be used.', permission => 'The mount permission of /.statelite directory is used, its default value is 755', dump => qq{The NFS directory to hold the Linux kernel dump file (vmcore) when the node with this image crashes, its format is "nfs:///". If you want to use the node's "xcatmaster" (its SN or MN), can be left blank. For example, "nfs:///" means the NFS directory to hold the kernel dump file is on the node's SN, or MN if there's no SN.}, crashkernelsize => 'the size that assigned to the kdump kernel. If the kernel size is not set, 256M will be the default value.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, passwd => { cols => [qw(key username password cryptmethod comments disable)], keys => [qw(key username)], table_desc => 'Contains default userids and passwords for xCAT to access cluster components. In most cases, xCAT will also actually set the userid/password in the relevant component when it is being configured or installed. Userids/passwords for specific cluster components can be overidden in other tables, e.g. mpa, ipmi, ppchcp, etc.', descriptions => { key => 'The type of component this user/pw is for. Valid values: blade (management module), ipmi (BMC), system (nodes), omapi (DHCP), hmc, ivm, cec, frame, switch.', username => 'The default userid for this type of component', password => 'The default password for this type of component', cryptmethod => 'Indicates the method that was used to encrypt the password attribute. On AIX systems, if a value is provided for this attribute it indicates that the passwword attribute is encrypted. If the cryptmethod value is not set it indicates the password is a simple string value. On Linux systems, the cryyptmethod is not supported however the code attempts to auto-discover MD5 encrypted passowrds.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, policy => { cols => [qw(priority name host commands noderange parameters time rule comments disable)], keys => [qw(priority)], table_desc => 'The policy table in the xCAT database controls who has authority to run specific xCAT operations. It is basically the Access Control List (ACL) for xCAT.', descriptions => { priority => 'The priority value for this rule. This value is used to identify this policy data object (i.e. this rule).', name => 'The username that is allowed to perform the commands specified by this rule. Default is "*" (all users).', host => 'The host from which users may issue the commands specified by this rule. Default is "*" (all hosts).', commands => 'The list of commands that this rule applies to. Default is "*" (all commands).', noderange => 'The Noderange that this rule applies to. Default is "*" (all nodes).', parameters => 'A regular expression that matches the command parameters (everything except the noderange) that this rule applies to. Default is "*" (all parameters).', time => 'Time ranges that this command may be executed in. This is not supported.', rule => 'Specifies how this rule should be applied. Valid values are: allow, accept, trusted. Allow or accept will allow the user to run the commands. Any other value will deny the user access to the commands. Trusted means that once this client has been authenticated via the certificate, all other information that is sent (e.g. the username) is believed without question. This authorization should only be given to the xcatd on the management node at this time.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, postscripts => { cols => [qw(node postscripts postbootscripts comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => ' The scripts that should be run on each node after installation or diskless boot.', descriptions => { node => 'The node name or group name.', postscripts => 'Comma separated list of scripts that should be run on this node after diskfull installation or diskless boot. Each script can take zero or more parameters. For example: "script1 p1 p2,script2,...". xCAT automatically adds the postscripts from the xcatdefaults.postscripts attribute of the table to run first on the nodes after install or diskless boot. For installation of RedHat, CentOS, Fedora, the scripts will be run before the reboot. For installation of SLES, the scripts will be run after the reboot but before the init.d process. For diskless deployment, the scripts will be run at the init.d time, and xCAT will automatically add the list of scripts from the postbootscripts attribute to run after postscripts list. For installation of AIX, the scripts will run after the reboot and acts the same as the postbootscripts attribute. For AIX, use the postbootscripts attribute. Support will be added in the future for the postscripts attribute to run the scripts before the reboot in AIX. ', postbootscripts => 'Comma separated list of scripts that should be run on thisnode after diskfull installation or diskless boot. Each script can take zero or more parameters. For example: "script1 p1 p2,script2,...". On AIX these scripts are run during the processing of /etc/inittab. On Linux they are run at the init.d time. xCAT automatically adds the scripts in the xcatdefaults.postbootscripts attri bute to run first in the list.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, ppc => { cols => [qw(node hcp id pprofile parent nodetype supernode sfp comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => 'List of system p hardware: HMCs, IVMs, FSPs, BPCs, CECs, Frames.', descriptions => { node => 'The node name or group name.', hcp => 'The hardware control point for this node (HMC, IVM, Frame or CEC). Do not need to set for BPAs and FSPs.', id => 'For LPARs: the LPAR numeric id; for CECs: the cage number; for Frames: the frame number.', pprofile => 'The LPAR profile that will be used the next time the LPAR is powered on with rpower. For DFM, the pprofile attribute should be set to blank ', parent => 'For LPARs: the CEC; for FSPs: the CEC; for CEC: the frame (if one exists); for BPA: the frame; for frame: the building block number (which consists 1 or more service nodes and compute/storage nodes that are serviced by them - optional).', nodetype => 'The hardware type of the node. Only can be one of fsp, bpa, cec, frame, ivm, hmc and lpar', supernode => 'Indicates the connectivity of this CEC in the HFI network. A comma separated list of 2 ids. The first one is the supernode number the CEC is part of. The second one is the logical location number (0-3) of this CEC within the supernode.', sfp => 'The Service Focal Point of this Frame. This is the name of the HMC that is responsible for collecting hardware service events for this frame and all of the CECs within this frame.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, ppcdirect => { cols => [qw(hcp username password comments disable)], keys => [qw(hcp username)], nodecol => "hcp", table_desc => 'Info necessary to use FSPs/BPAs to control system p CECs/Frames.', descriptions => { hcp => 'Hostname of the FSPs/BPAs(for ASMI) and CECs/Frames(for DFM).', username => 'Userid of the FSP/BPA(for ASMI) and CEC/Frame(for DFM). If not filled in, xCAT will look in the passwd table for key=fsp. If not in the passwd table, the default used is admin.', password => 'Password of the FSP/BPA(for ASMI) and CEC/Frame(for DFM). If not filled in, xCAT will look in the passwd table for key=fsp. If not in the passwd table, the default used is admin.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, ppchcp => { cols => [qw(hcp username password comments disable)], keys => [qw(hcp)], nodecol => "hcp", table_desc => 'Info necessary to use HMCs and IVMs as hardware control points for LPARs.', descriptions => { hcp => 'Hostname of the HMC or IVM.', username => 'Userid of the HMC or IVM. If not filled in, xCAT will look in the passwd table for key=hmc or key=ivm. If not in the passwd table, the default used is hscroot for HMCs and padmin for IVMs.', password => 'Password of the HMC or IVM. If not filled in, xCAT will look in the passwd table for key=hmc or key=ivm. If not in the passwd table, the default used is abc123 for HMCs and padmin for IVMs.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, servicenode => { cols => [qw(node nameserver dhcpserver tftpserver nfsserver conserver monserver ldapserver ntpserver ftpserver nimserver ipforward comments disable)], keys => [qw(node)], table_desc => 'List of all Service Nodes and services that will be set up on the Service Node.', descriptions => { node => 'The hostname of the service node as known by the Management Node.', nameserver => 'Do we set up DNS on this service node? Valid values:yes or 1, no or 0. If yes, creates named.conf file with forwarding to the management node and starts named. If no or 0, it does not change the current state of the service. ', dhcpserver => 'Do we set up DHCP on this service node? Not supported on AIX. Valid values:yes or 1, no or 0. If yes, runs makedhcp -n. If no or 0, it does not change the current state of the service. ', tftpserver => 'Do we set up TFTP on this service node? Not supported on AIX. Valid values:yes or 1, no or 0. If yes, configures and starts atftp. If no or 0, it does not change the current state of the service. ', nfsserver => 'Do we set up file services (HTTP,FTP,or NFS) on this service node? For AIX will only setup NFS, not HTTP or FTP. Valid values:yes or 1, no or 0.If no or 0, it does not change the current state of the service. ', conserver => 'Do we set up Conserver on this service node? Valid values:yes or 1, no or 0. If yes, configures and starts conserver daemon. If no or 0, it does not change the current state of the service.', monserver => 'Is this a monitoring event collection point? Valid values:yes or 1, no or 0. If no or 0, it does not change the current state of the service.', ldapserver => 'Do we set up ldap caching proxy on this service node? Not supported on AIX. Valid values:yes or 1, no or 0. If no or 0, it does not change the current state of the service.', ntpserver => 'Not used. Use setupntp postscript to setup a ntp server on this service node? Valid values:yes or 1, no or 0. If no or 0, it does not change the current state of the service.', ftpserver => 'Do we set up a ftp server on this service node? Not supported on AIX Valid values:yes or 1, no or 0. If yes, configure and start vsftpd. If no or 0, it does not change the current state of the service. xCAT is not using ftp for compute nodes provisioning or any other xCAT features, so this attribute can be set to 0 if the ftp service will not be used for other purposes', nimserver => 'Not used. Do we set up a NIM server on this service node? Valid values:yes or 1, no or 0. If no or 0, it does not change the current state of the service.', ipforward => 'Do we set up ip forwarding on this service node? Valid values:yes or 1, no or 0. If no or 0, it does not change the current state of the service.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, site => { cols => [qw(key value comments disable)], keys => [qw(key)], table_desc => "Global settings for the whole cluster. This table is different from the \nother tables in that each attribute is just named in the key column, rather \nthan having a separate column for each attribute. The following is a list of \nthe attributes currently used by xCAT.\n", descriptions => { # Do not put description text past column 88, so it displays well in a 100 char wide window. # ----------------------------------------------------------------------------------|---------- key => "Attribute Name: Description\n\n". " auditskipcmds: List of commands that will not be written to the auditlog table.\n". " 'ALL' means all cmds will be skipped. If attribute is null, all\n". " commands will be written.\n". " For example: tabdump,nodels,lsdef will not log those cmds.\n\n". " blademaxp: The maximum number of concurrent processes for blade hardware control.\n\n". " cleanupxcatpost: (yes/1 or no/0). Set to 'yes' or '1' to clean up the /xcatpost\n". " directory on the stateless and statelite nodes after the\n". " postscripts are run. Default is no.\n\n". " consoleondemand: When set to 'yes', conserver connects and creates the console\n". " output only when the user opens the console. Default is no on\n". " Linux, yes on AIX.\n\n". " databaseloc: Directory where we create the db instance directory.\n". " Default is /var/lib. Only DB2 is currently supported.\n". " Do not use the directory in the site.installloc or\n". " installdir attribute. This attribute must not be changed\n". " once db2sqlsetup script has been run and DB2 has been setup.\n\n". " db2installloc: The location which the service nodes should mount for\n". " the db2 code to install. Format is hostname:/path. If hostname is\n". " omitted, it defaults to the management node. Default is /mntdb2.\n\n". " defserialflow: The default serial flow - currently only used by the mknb command.\n\n". " defserialport: The default serial port - currently only used by mknb.\n\n". " defserialspeed: The default serial speed - currently only used by mknb.\n\n". " dhcpinterfaces: The network interfaces DHCP should listen on. If it is the same\n". " for all nodes, use simple comma-separated list of NICs. To\n". " specify different NICs for different nodes:\n". " mn|eth1,eth2;service|bond0.\n\n". " dhcpsetup: If set to 'n', it will skip the dhcp setup process in the nodeset cmd.\n\n". " disjointdhcps: If set to '1', the .leases file on a service node only contains\n". " the nodes it manages. The default value is '0'.\n". " '0' value means include all the nodes in the subnet.\n\n". " dnshandler: Name of plugin that handles DNS setup for makedns.\n". " domain: The DNS domain name used for the cluster.\n\n". " ea_primary_hmc: The hostname of the HMC that the Integrated Switch Network\n". " Management Event Analysis should send hardware serviceable\n". " events to for processing and potentially sending to IBM.\n\n". " ea_backup_hmc: The hostname of the HMC that the Integrated Switch Network\n". " Management Event Analysis should send hardware serviceable\n". " events to if the primary HMC is down.\n\n". " enableASMI: (yes/1 or no/0). If yes, ASMI method will be used after fsp-api. If no,\n". " when fsp-api is used, ASMI method will not be used. Default is no.\n\n". " excludenodes: A set of comma separated nodes and/or groups that would automatically\n". " be subtracted from any noderange, it can be used for excluding some\n". " failed nodes for any xCAT commands. See the 'noderange' manpage for\n". " details on supported formats.\n\n". " forwarders: The DNS servers at your site that can provide names outside of the\n". " cluster. The makedns command will configuire the DNS on the management\n". " node to forward requests it does not know to these servers.\n". " Note that the DNS servers on the service nodes will ignore this value\n". " and always be configured to forward requests to the management node.\n\n". " fsptimeout: The timeout, in milliseconds, to use when communicating with FSPs.\n\n". " genmacprefix: When generating mac addresses automatically, use this manufacturing\n". " prefix (e.g. 00:11:aa)\n\n". " genpasswords: Automatically generate random passwords for BMCs when configuring\n". " them.\n\n". " installdir: The local directory name used to hold the node deployment packages.\n\n". " installloc: The location from which the service nodes should mount the \n". " deployment packages in the format hostname:/path. If hostname is\n". " omitted, it defaults to the management node. The path must\n". " match the path in the installdir attribute.\n\n". " ipmidispatch: Whether or not to send ipmi hw control operations to the service\n". " node of the target compute nodes. Default is 'y'.\n\n". " ipmimaxp: The max # of processes for ipmi hw ctrl. The default is 64. Currently,\n". " this is only used for HP hw control.\n\n". " ipmiretries: The # of retries to use when communicating with BMCs. Default is 3.\n\n". " ipmisdrcache: If set to 'no', then the xCAT IPMI support will not cache locally\n". " the target node's SDR cache to improve performance.\n\n". " ipmitimeout: The timeout to use when communicating with BMCs. Default is 2.\n". " This attribute is currently not used.\n\n". " iscsidir: The path to put the iscsi disks in on the mgmt node.\n\n". " master: The hostname of the xCAT management node, as known by the nodes.\n\n". " maxssh: The max # of SSH connections at any one time to the hw ctrl point for PPC\n". " This parameter doesn't take effect on the rpower command.\n". " It takes effects on other PPC hardware control command\n". " getmacs/rnetboot/rbootseq and so on. Default is 8.\n\n". " mnroutenames: The name of the routes to be setup on the management node.\n". " It is a comma separated list of route names that are defined in the\n". " routes table.\n\n". " nameservers: A comma delimited list of DNS servers that each node in the cluster\n". " should use. This value will end up in the nameserver settings of the\n". " /etc/resolv.conf on each node. It is common (but not required) to set\n". " this attribute value to the IP addr of the xCAT management node, if\n". " you have set up the DNS on the management node by running makedns.\n". " In a hierarchical cluster, you can also set this attribute to\n". " \"\" to mean the DNS server for each node should be the\n". " node that is managing it (either its service node or the management\n". " node).\n\n". " nimprime : The name of NIM server, if not set default is the AIX MN. If Linux MN, then must be set for support of mixed cluster (TBD).\n\n". " nodestatus: If set to 'n', the nodelist.status column will not be updated during\n". " the node deployment, node discovery and power operations.\n\n". " ntpservers: A comma delimited list of NTP servers for the cluster - often the\n". " xCAT management node.\n\n". " syspowerinterval: The number of seconds the rpower command to servers will wait\n". " between performing the action for each server. Currently supported\n". " for system p CECs and system x IPMI servers. This is used for\n". " controlling the power on speed in large clusters. Default is 0.\n\n". " syspowermaxnodes: The number of servers to power on at one time before waiting\n". " 'syspowerinterval' seconds to continue on to the next set of\n". " nodes. Currently only used for IPMI servers and must be set if\n". " 'syspowerinterval' is set.\n\n". " powerinterval: The number of seconds the rpower command to LPARs will wait between\n". " performing the action for each LPAR. LPARs of different HCPs\n". " (HMCs or FSPs) are done in parallel. This is used to limit the\n". " cluster boot up speed in large clusters. Default is 0. This is\n". " currently only used for system p hardware.\n\n". " ppcmaxp: The max # of processes for PPC hw ctrl. If there are more than ppcmaxp hcps,\n". " this parameter will take effect. It will control the max number of processes \n". " for PPC hardware control commands. Default is 64.\n\n". " ppcretry: The max # of PPC hw connection attempts to HMC before failing.\n". " It only takes effect on the hardware control commands through HMC. \n". " Default is 3.\n\n". " ppctimeout: The timeout, in milliseconds, to use when communicating with PPC hw\n". " through HMC. It only takes effect on the hardware control commands\n". " through HMC. Default is 0.\n\n". " pruneservices: Whether to enable service pruning when noderm is run (i.e.\n". " removing DHCP entries when noderm is executed)\n\n". " rsh: This is no longer used. path to remote shell command for xdsh.\n\n". " rcp: This is no longer used. path to remote copy command for xdcp.\n\n". " sharedtftp: Set to 0 or no, xCAT should not assume the directory\n". " in tftpdir is mounted on all on Service Nodes. Default is 1/yes.\n". " If value is set to a hostname, the directory in tftpdir\n". " will be mounted from that hostname on the SN\n\n". " sharedinstall: Indicates if a shared file system will be used for installation\n". " resources. Possible values are: 'no', 'sns', or 'all'. 'no' \n". " means a shared file system is not being used. 'sns' means a\n". " shared filesystem is being used across all service nodes.\n". " 'all' means that the management as well as the service nodes\n". " are all using a common shared filesystem. The default is 'no'.\n". " skiptables: Comma separated list of tables to be skipped by dumpxCATdb\n". " SNsyncfiledir: The directory on the Service Node, where xdcp will copy the files\n". " from the MN that will eventually be copied to the compute nodes.\n\n". " snmpc: The snmp community string that xcat should use when communicating with the\n". " switches.\n\n". " sshbetweennodes: Comma separated list of groups to enable passwordless root \n". " ssh during install, or xdsh -K. Default is ALLGROUPS.\n". " Set to NOGROUPS,if you do not wish to enabled any groups.\n". " Service Nodes are not affected by this attribute\n". " they are always setup with\n". " passwordless root access to nodes and other SN.\n\n". " svloglocal: if set to 1, syslog on the service node will not get forwarded to the\n". " mgmt node.\n\n". " timezone: (e.g. America/New_York)\n\n". " tftpdir: tftp directory path. Default is /tftpboot\n\n". " useNmapfromMN: When set to yes, nodestat command should obtain the node status\n". " using nmap (if available) from the management node instead of the\n". " service node. This will improve the performance in a flat network.\n\n". " useSSHonAIX: (yes/1 or no/0). If yes, ssh/scp will be setup and used. If no,\n". " rsh/rcp will be setup and used on AIX. Default is yes.\n\n". " useNFSv4onAIX: (yes/1 or no/0). If yes, NFSv4 will be used with NIM. If no,\n". " NFSv3 will be used with NIM. Default is no.\n\n". " vcenterautojoin: When set to no, the VMWare plugin will not attempt to auto remove\n". " and add hypervisors while trying to perform operations. If users\n". " or tasks outside of xCAT perform the joining this assures xCAT\n". " will not interfere.\n\n". " vmwarereconfigonpower: When set to no, the VMWare plugin will make no effort to\n". " push vm.cpus/vm.memory updates from xCAT to VMWare.\n\n". " vsftp: Default is 'n'. If set to 'y', the xcatd on the mn will automatically\n". " bring up vsftpd. This setting does not apply to the sn, for sn\n". " you need to set servicenode.ftpserver=1 if you want xcatd to\n". " bring up vsftpd.\n\n". " xcatconfdir: Where xCAT config data is (default /etc/xcat).\n\n". " xcatmaxconnections: Number of concurrent xCAT protocol requests before requests\n". " begin queueing. This applies to both client command requests\n". " and node requests, e.g. to get postscripts. Default is 64.\n\n". " xcatdport: The port used by the xcatd daemon for client/server communication.\n\n". " xcatiport: The port used by xcatd to receive install status updates from nodes.\n\n", value => 'The value of the attribute specified in the "key" column.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, switch => { cols => [qw(node switch port vlan interface comments disable)], keys => [qw(node switch port)], table_desc => 'Contains what switch port numbers each node is connected to.', descriptions => { node => 'The node name or group name.', switch => 'The switch hostname.', port => 'The port number in the switch that this node is connected to. On a simple 1U switch, an administrator can generally enter the number as printed next to the ports, and xCAT will understand switch representation differences. On stacked switches or switches with line cards, administrators should usually use the CLI representation (i.e. 2/0/1 or 5/8). One notable exception is stacked SMC 8848M switches, in which you must add 56 for the proceeding switch, then the port number. For example, port 3 on the second switch in an SMC8848M stack would be 59', vlan => 'xCAT currently does not make use of this field, however it may do so in the future. For now, it can be used by administrators for their own purposes, but keep in mind some xCAT feature later may try to enforce this if set', interface => 'The interface name from the node perspective. This is not currently used by xCAT, but administrators may wish to use this for their own purposes', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, vpd => { cols => [qw(node serial mtm side asset uuid comments disable)], keys => [qw(node)], table_desc => 'The Machine type, Model, and Serial numbers of each node.', descriptions => { node => 'The node name or group name.', serial => 'The serial number of the node.', mtm => 'The machine type and model number of the node. E.g. 7984-6BU', side => '- or -. The side information for the BPA/FSP. The side attribute refers to which BPA/FSP, A or B, which is determined by the slot value returned from lsslp command. It also lists the physical port within each BPA/FSP which is determined by the IP address order from the lsslp response. This information is used internally when communicating with the BPAs/FSPs', asset => 'A field for administators to use to correlate inventory numbers they may have to accomodate', uuid => 'The UUID applicable to the node', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, nimimage => { cols => [qw(imagename nimtype lpp_source spot root dump paging resolv_conf tmp home shared_home res_group nimmethod script bosinst_data installp_bundle mksysb fb_script shared_root otherpkgs image_data comments disable)], keys => [qw(imagename)], table_desc => 'All the info that specifies a particular AIX operating system image that can be used to deploy AIX nodes.', descriptions => { imagename => 'User provided name of this xCAT OS image definition.', nimtype => 'The NIM client type- standalone, diskless, or dataless.', lpp_source => 'The name of the NIM lpp_source resource.', spot => 'The name of the NIM SPOT resource.', root => 'The name of the NIM root resource.', dump => 'The name of the NIM dump resource.', paging => 'The name of the NIM paging resource.', resolv_conf => 'The name of the NIM resolv_conf resource.', tmp => 'The name of the NIM tmp resource.', home => 'The name of the NIM home resource.', shared_home => 'The name of the NIM shared_home resource.', res_group => 'The name of a NIM resource group.', nimmethod => 'The NIM install method to use, (ex. rte, mksysb).', script => 'The name of a NIM script resource.', fb_script => 'The name of a NIM fb_script resource.', bosinst_data => 'The name of a NIM bosinst_data resource.', otherpkgs => "One or more comma separated installp or rpm packages. The rpm packages must have a prefix of 'R:', (ex. R:foo.rpm)", installp_bundle => 'One or more comma separated NIM installp_bundle resources.', mksysb => 'The name of a NIM mksysb resource.', shared_root => 'A shared_root resource represents a directory that can be used as a / (root) directory by one or more diskless clients.', image_data => 'The name of a NIM image_data resource.', comments => 'Any user-provided notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, performance => { cols => [qw(timestamp node attrname attrvalue comments disable)], keys => [qw(timestamp node attrname)], table_desc => 'Describes the system performance every interval unit of time.', descriptions => { timestamp => 'The time at which the metric was captured.', node => 'The node name.', attrname => 'The metric name.', attrvalue => 'The metric value.', comments => 'Any user-provided notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, eventlog => { cols => [qw(recid eventtime eventtype monitor monnode node application component id severity message rawdata comments disable)], keys => [qw(recid)], types => { recid => 'INTEGER AUTO_INCREMENT', }, tablespace =>'XCATTBS32K', table_desc => 'Stores the events occurred.', descriptions => { recid => 'The record id.', eventtime => 'The timestamp for the event.', eventtype => 'The type of the event.', # for RMC it's either "Event" or "Rearm event". monitor => 'The name of the monitor that monitors this event.', #for RMC, it's the condition name monnode => 'The node that monitors this event.', node => 'The node where the event occurred.', application => 'The application that reports the event.', #RMC, Ganglia component => 'The component where the event occurred.', #in RMC, it's the resource class name id => 'The location or the resource name where the event occurred.', #In RMC it's the resource name and attribute name severity => 'The severity of the event. Valid values are: informational, warning, critical.', message => 'The full description of the event.', rawdata => ' The data that associated with the event. ', # in RMC, it's the attribute value, it takes the format of attname=attvalue[,atrrname=attvalue....] comments => 'Any user-provided notes.', disable => "Do not use. tabprune will not work if set to yes or 1", }, }, auditlog => { cols => [qw(recid audittime userid clientname clienttype command noderange args status comments disable)], keys => [qw(recid)], types => { recid => 'INTEGER AUTO_INCREMENT', }, compress =>'YES', tablespace =>'XCATTBS32K', table_desc => ' Audit Data log.', descriptions => { recid => 'The record id.', audittime => 'The timestamp for the audit entry.', userid => 'The user running the command.', clientname => 'The client machine, where the command originated.', clienttype => 'Type of command: cli,java,webui,other.', command => 'Command executed.', noderange => 'The noderange on which the command was run.', args => 'The command argument list.', status => 'Allowed or Denied.', comments => 'Any user-provided notes.', disable => "Do not use. tabprune will not work if set to yes or 1", }, }, prescripts => { cols => [qw(node begin end comments disable)], keys => [qw(node)], tablespace =>'XCATTBS16K', table_desc => 'The scripts that will be run at the beginning and the end of the nodeset(Linux), nimnodeset(AIX) or mkdsklsnode(AIX) command.', descriptions => { node => 'The node name or group name.', # Do not put description text past column 88, so it displays well in a 100 char wide window. # ----------------------------------------------------------------------------------| begin => "The scripts to be run at the beginning of the nodeset(Linux),\n" . " nimnodeset(AIX) or mkdsklsnode(AIX) command.\n". " The format is:\n". " [action1:]s1,s2...[|action2:s3,s4,s5...]\n". " where:\n". " - action1 and action2 for Linux are the nodeset actions specified in the command. \n" . " For AIX, action1 and action1 can be 'diskless' for mkdsklsnode command'\n" . " and 'standalone for nimnodeset command. \n" . " - s1 and s2 are the scripts to run for action1 in order.\n". " - s3, s4, and s5 are the scripts to run for actions2.\n". " If actions are omitted, the scripts apply to all actions.\n". " Examples:\n". " myscript1,myscript2 (all actions)\n". " diskless:myscript1,myscript2 (AIX)\n". " install:myscript1,myscript2|netboot:myscript3 (Linux)\n\n". " All the scripts should be copied to /install/prescripts directory.\n". " The following two environment variables will be passed to each script: \n". " NODES a coma separated list of node names that need to run the script for\n". " ACTION current nodeset action.\n\n". " If '#xCAT setting:MAX_INSTANCE=number' is specified in the script, the script\n". " will get invoked for each node in parallel, but no more than number of instances\n". " will be invoked at at a time. If it is not specified, the script will be invoked\n". " once for all the nodes.\n", end => "The scripts to be run at the end of the nodeset(Linux),\n". " nimnodeset(AIX),or mkdsklsnode(AIX) command. \n". " The format is the same as the 'begin' column.", comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, routes => { cols => [qw(routename net mask gateway ifname comments disable)], keys => [qw(routename)], table_desc => 'Describes the additional routes needed to be setup in the os routing table. These routes usually are used to connect the management node to the compute node using the servie node as gateway.', descriptions => { routename => 'Name used to identify this route.', net => 'The network address.', mask => 'The network mask.', ifname => '(optional) The interface name of the management node facing the gateway.', gateway => 'The gateway that routes the ip traffic from the mn to the nodes. It is usually a service node.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, zvm => { cols => [qw(node hcp userid nodetype parent comments disable)], keys => [qw(node)], table_desc => 'List of z/VM virtual servers.', descriptions => { node => 'The node name.', hcp => 'The hardware control point for this node.', userid => 'The z/VM userID of this node.', nodetype => 'The node type. Valid values: cec (Central Electronic Complex), lpar (logical partition), zvm (z/VM host operating system), and vm (virtual machine).', parent => 'The parent node. For LPAR, this specifies the CEC. For z/VM, this specifies the LPAR. For VM, this specifies the z/VM host operating system.', comments => 'Any user provided notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, firmware => { cols => [qw(node cfgfile comments disable)], keys => [qw(node)], required => [qw(node)], table_desc => 'Maps node to firmware values to be used for setup at node discovery or later', descriptions => { node => 'The node id.', cfgfile => 'The file to use.', comments => 'Any user-written notes.', disable => "Set to 'yes' or '1' to comment out this row.", }, }, ); # end of tabspec definition ################################################### # adding user defined external tables ################################################## foreach my $tabname (keys(%xCAT::ExtTab::ext_tabspec)) { if (exists($tabspec{$tabname})) { xCAT::MsgUtils->message('ES', "\n Warning: Conflict when adding user defined tablespec. Duplicate table name: $tabname. \n"); } else { $tabspec{$tabname}=$xCAT::ExtTab::ext_tabspec{$tabname}; } } #################################################### # # Data abstraction definitions # For each table entry added to the database schema, # a corresponding attribute should be added to one of # the data objects below, or new data objects should # be created as needed. # # Definition format: # List of data object hashes: # => # {attrs => # [ {attr_name => '', # only_if => '=', # # optional, used to define conditional attributes. # # is a previously resolved attribute from # # this data object. # tabentry => '', # # where the data is stored in the database # access_tabentry => '=::=', # # how to look up tabentry. Now support multiple lookup entries, useful for 'multiple keys" in the table # For , # # if "attr:", use a previously resolved # # attribute value from the data object # # for now, only supports the objectname in attr: # # if "str:" use the value directly # description => '', # }, # {attr_name => , # ... # } ], # attrhash => {}, # internally generated hash of attrs array # # to allow code direct access to an attr def # objkey => 'attribute_name' # key attribute for this data object # } # # #################################################### %defspec = ( node => { attrs => [], attrhash => {}, objkey => 'node' }, osimage => { attrs => [], attrhash => {}, objkey => 'imagename' }, network => { attrs => [], attrhash => {}, objkey => 'netname' }, route => { attrs => [], attrhash => {}, objkey => 'routename' }, group => { attrs => [], attrhash => {}, objkey => 'groupname' }, site => { attrs => [], attrhash => {}, objkey => 'master' }, policy => { attrs => [], attrhash => {}, objkey => 'priority' }, monitoring => { attrs => [], attrhash => {}, objkey => 'name' }, notification => { attrs => [], attrhash => {}, objkey => 'filename' }, eventlog => { attrs => [], attrhash => {}, objkey => 'recid' }, auditlog => { attrs => [], attrhash => {}, objkey => 'recid' }, boottarget => { attrs => [], attrhash => {}, objkey => 'bprofile' }, ); ############### # @nodeattrs ia a list of node attrs that can be used for # BOTH node and group definitions ############## my @nodeattrs = ( {attr_name => 'nodetype', tabentry => 'nodetype.nodetype', access_tabentry => 'nodetype.node=attr:node', }, #################### # postscripts table# #################### {attr_name => 'postscripts', tabentry => 'postscripts.postscripts', access_tabentry => 'postscripts.node=attr:node', }, {attr_name => 'postbootscripts', tabentry => 'postscripts.postbootscripts', access_tabentry => 'postscripts.node=attr:node', }, #################### # noderes table # #################### {attr_name => 'xcatmaster', tabentry => 'noderes.xcatmaster', access_tabentry => 'noderes.node=attr:node', }, ### # TODO: Need to check/update code to make sure it really uses servicenode as # default if other server value not set ### {attr_name => 'servicenode', tabentry => 'noderes.servicenode', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'tftpserver', tabentry => 'noderes.tftpserver', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'tftpdir', tabentry => 'noderes.tftpdir', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'nfsserver', tabentry => 'noderes.nfsserver', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'nimserver', tabentry => 'noderes.nimserver', access_tabentry => 'noderes.node=attr:node', }, ### # TODO: Is noderes.nfsdir used anywhere? Could not find any code references # to this attribute. ### {attr_name => 'nfsdir', tabentry => 'noderes.nfsdir', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'monserver', tabentry => 'noderes.monserver', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'kernel', tabentry => 'bootparams.kernel', access_tabentry => 'bootparams.node=attr:node', }, {attr_name => 'initrd', tabentry => 'bootparams.initrd', access_tabentry => 'bootparams.node=attr:node', }, {attr_name => 'kcmdline', tabentry => 'bootparams.kcmdline', access_tabentry => 'bootparams.node=attr:node', }, {attr_name => 'addkcmdline', tabentry => 'bootparams.addkcmdline', access_tabentry => 'bootparams.node=attr:node', }, # Note that the serialport attr is actually defined down below # with the other serial* attrs from the nodehm table #{attr_name => 'serialport', # tabentry => 'noderes.serialport', # access_tabentry => 'noderes.node=attr:node', # }, {attr_name => 'primarynic', tabentry => 'noderes.primarynic', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'installnic', tabentry => 'noderes.installnic', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'netboot', tabentry => 'noderes.netboot', access_tabentry => 'noderes.node=attr:node', }, {attr_name => 'routenames', tabentry => 'noderes.routenames', access_tabentry => 'noderes.node=attr:node', }, ###################### # servicenode table # ###################### {attr_name => 'setupnameserver', tabentry => 'servicenode.nameserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setupdhcp', tabentry => 'servicenode.dhcpserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setuptftp', tabentry => 'servicenode.tftpserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setupnfs', tabentry => 'servicenode.nfsserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setupconserver', tabentry => 'servicenode.conserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setupipforward', tabentry => 'servicenode.ipforward', access_tabentry => 'servicenode.node=attr:node', }, # - moserver not used yet # {attr_name => 'setupmonserver', # tabentry => 'servicenode.monserver', # access_tabentry => 'servicenode.node=attr:node', # }, {attr_name => 'setupldap', tabentry => 'servicenode.ldapserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setupntp', tabentry => 'servicenode.ntpserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setupftp', tabentry => 'servicenode.ftpserver', access_tabentry => 'servicenode.node=attr:node', }, {attr_name => 'setupnim', tabentry => 'servicenode.nimserver', access_tabentry => 'servicenode.node=attr:node', }, ###################### # nodetype table # ###################### {attr_name => 'arch', tabentry => 'nodetype.arch', access_tabentry => 'nodetype.node=attr:node', }, {attr_name => 'supportedarchs', tabentry => 'nodetype.supportedarchs', access_tabentry => 'nodetype.node=attr:node', }, {attr_name => 'os', tabentry => 'nodetype.os', access_tabentry => 'nodetype.node=attr:node', }, # TODO: need to decide what to do with the profile attr once the osimage # stuff is implemented. May want to move this to the osimage table. {attr_name => 'profile', tabentry => 'nodetype.profile', access_tabentry => 'nodetype.node=attr:node', }, {attr_name => 'provmethod', tabentry => 'nodetype.provmethod', access_tabentry => 'nodetype.node=attr:node', }, #################### # iscsi table # #################### {attr_name => 'iscsiserver', tabentry => 'iscsi.server', access_tabentry => 'iscsi.node=attr:node', }, {attr_name => 'iscsitarget', tabentry => 'iscsi.target', access_tabentry => 'iscsi.node=attr:node', }, {attr_name => 'iscsiuserid', tabentry => 'iscsi.userid', access_tabentry => 'iscsi.node=attr:node', }, {attr_name => 'iscsipassword', tabentry => 'iscsi.passwd', access_tabentry => 'iscsi.node=attr:node', }, #################### # nodehm table # #################### {attr_name => 'mgt', tabentry => 'nodehm.mgt', access_tabentry => 'nodehm.node=attr:node', }, {attr_name => 'power', tabentry => 'nodehm.power', access_tabentry => 'nodehm.node=attr:node', }, {attr_name => 'cons', tabentry => 'nodehm.cons', access_tabentry => 'nodehm.node=attr:node', }, {attr_name => 'termserver', tabentry => 'nodehm.termserver', access_tabentry => 'nodehm.node=attr:node', }, {attr_name => 'termport', tabentry => 'nodehm.termport', access_tabentry => 'nodehm.node=attr:node', }, ### # TODO: is nodehm.conserver used anywhere? I couldn't find any code references ### {attr_name => 'conserver', tabentry => 'nodehm.conserver', access_tabentry => 'nodehm.node=attr:node', }, ### # TODO: is nodehm.getmac used anywhere? I couldn't find any code references ### {attr_name => 'getmac', tabentry => 'nodehm.getmac', access_tabentry => 'nodehm.node=attr:node', }, {attr_name => 'serialport', tabentry => 'nodehm.serialport', access_tabentry => 'nodehm.node=attr:node', }, {attr_name => 'serialspeed', tabentry => 'nodehm.serialspeed', access_tabentry => 'nodehm.node=attr:node', }, {attr_name => 'serialflow', tabentry => 'nodehm.serialflow', access_tabentry => 'nodehm.node=attr:node', }, ################## # vpd table # ################## {attr_name => 'serial', tabentry => 'vpd.serial', access_tabentry => 'vpd.node=attr:node', }, {attr_name => 'mtm', tabentry => 'vpd.mtm', access_tabentry => 'vpd.node=attr:node', }, {attr_name => 'side', tabentry => 'vpd.side', access_tabentry => 'vpd.node=attr:node', }, ################## # mac table # ################## {attr_name => 'interface', tabentry => 'mac.interface', access_tabentry => 'mac.node=attr:node', }, {attr_name => 'mac', tabentry => 'mac.mac', access_tabentry => 'mac.node=attr:node', }, ################## # chain table # ################## ### # TODO: Need user documentation from Jarrod on how to use chain, what each # action does, valid ordering, etc. ### {attr_name => 'chain', tabentry => 'chain.chain', access_tabentry => 'chain.node=attr:node', }, ### # TODO: What is chain.ondiscover used for? Could not find any code references # to this table entry ### {attr_name => 'ondiscover', tabentry => 'chain.ondiscover', access_tabentry => 'chain.node=attr:node', }, {attr_name => 'currstate', tabentry => 'chain.currstate', access_tabentry => 'chain.node=attr:node', }, {attr_name => 'currchain', tabentry => 'chain.currchain', access_tabentry => 'chain.node=attr:node', }, #################### # ppchcp table # #################### {attr_name => 'username', only_if => 'nodetype=ivm', tabentry => 'ppchcp.username', access_tabentry => 'ppchcp.hcp=attr:node', }, {attr_name => 'password', only_if => 'nodetype=ivm', tabentry => 'ppchcp.password', access_tabentry => 'ppchcp.hcp=attr:node', }, {attr_name => 'username', only_if => 'nodetype=hmc', tabentry => 'ppchcp.username', access_tabentry => 'ppchcp.hcp=attr:node', }, {attr_name => 'password', only_if => 'nodetype=hmc', tabentry => 'ppchcp.password', access_tabentry => 'ppchcp.hcp=attr:node', }, {attr_name => 'username', only_if => 'nodetype=ppc', tabentry => 'ppchcp.username', access_tabentry => 'ppchcp.hcp=attr:node', }, {attr_name => 'password', only_if => 'nodetype=ppc', tabentry => 'ppchcp.password', access_tabentry => 'ppchcp.hcp=attr:node', }, #################### # ppc table # #################### {attr_name => 'hcp', tabentry => 'ppc.hcp', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'id', tabentry => 'ppc.id', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'pprofile', only_if => 'mgt=hmc', tabentry => 'ppc.pprofile', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'pprofile', only_if => 'mgt=ivm', tabentry => 'ppc.pprofile', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'pprofile', only_if => 'mgt=fsp', tabentry => 'ppc.pprofile', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'parent', only_if => 'mgt=hmc', tabentry => 'ppc.parent', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'parent', only_if => 'mgt=ivm', tabentry => 'ppc.parent', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'parent', only_if => 'mgt=bpa', tabentry => 'ppc.parent', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'parent', only_if => 'mgt=fsp', tabentry => 'ppc.parent', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'hwtype', only_if => 'mgt=fsp', tabentry => 'ppc.nodetype', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'hwtype', only_if => 'mgt=bpa', tabentry => 'ppc.nodetype', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'hwtype', only_if => 'mgt=ivm', tabentry => 'ppc.nodetype', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'hwtype', only_if => 'mgt=hmc', tabentry => 'ppc.nodetype', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'hwtype', only_if => 'mgt=zvm', tabentry => 'zvm.nodetype', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'hwtype', only_if => 'mgt=blade', tabentry => 'mp.nodetype', access_tabentry => 'mp.node=attr:node', }, {attr_name => 'supernode', tabentry => 'ppc.supernode', access_tabentry => 'ppc.node=attr:node', }, {attr_name => 'sfp', tabentry => 'ppc.sfp', access_tabentry => 'ppc.node=attr:node', }, ####################### # ppcdirect table # ####################### {attr_name => 'passwd.HMC', only_if => 'mgt=fsp', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:HMC', }, {attr_name => 'passwd.hscroot', only_if => 'mgt=fsp', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:hscroot', }, {attr_name => 'passwd.admin', only_if => 'mgt=fsp', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:admin', }, {attr_name => 'passwd.general', only_if => 'mgt=fsp', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:general', }, {attr_name => 'passwd.celogin', only_if => 'mgt=fsp', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:celogin', }, {attr_name => 'passwd.celogin', only_if => 'mgt=bpa', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:celogin', }, {attr_name => 'passwd.HMC', only_if => 'mgt=bpa', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:HMC', }, {attr_name => 'passwd.hscroot', only_if => 'mgt=bpa', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:hscroot', }, {attr_name => 'passwd.admin', only_if => 'mgt=bpa', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:admin', }, {attr_name => 'passwd.general', only_if => 'mgt=bpa', tabentry => 'ppcdirect.password', access_tabentry => 'ppcdirect.hcp=attr:node::ppcdirect.username=str:general', }, #################### # zvm table # #################### {attr_name => 'hcp', only_if => 'mgt=zvm', tabentry => 'zvm.hcp', access_tabentry => 'zvm.node=attr:node', }, {attr_name => 'userid', only_if => 'mgt=zvm', tabentry => 'zvm.userid', access_tabentry => 'zvm.node=attr:node', }, ################## # ipmi table # ################## {attr_name => 'bmc', only_if => 'mgt=ipmi', tabentry => 'ipmi.bmc', access_tabentry => 'ipmi.node=attr:node', }, {attr_name => 'bmcport', only_if => 'mgt=ipmi', tabentry => 'ipmi.bmcport', access_tabentry => 'ipmi.node=attr:node', }, {attr_name => 'bmcusername', only_if => 'mgt=ipmi', tabentry => 'ipmi.username', access_tabentry => 'ipmi.node=attr:node', }, {attr_name => 'bmcpassword', only_if => 'mgt=ipmi', tabentry => 'ipmi.password', access_tabentry => 'ipmi.node=attr:node', }, ################ # mp table # ################ {attr_name => 'mpa', # Remove the restriction so that fsp also can # write to mpa attribute #only_if => 'mgt=blade', tabentry => 'mp.mpa', access_tabentry => 'mp.node=attr:node', }, {attr_name => 'slotid', only_if => 'mgt=fsp', tabentry => 'mp.id', access_tabentry => 'mp.node=attr:node', }, {attr_name => 'id', only_if => 'mgt=blade', tabentry => 'mp.id', access_tabentry => 'mp.node=attr:node', }, ################# # mpa table # ################# {attr_name => 'username', only_if => 'nodetype=mm', tabentry => 'mpa.username', access_tabentry => 'mpa.mpa=attr:node', }, {attr_name => 'password', only_if => 'nodetype=mm', tabentry => 'mpa.password', access_tabentry => 'mpa.mpa=attr:node', }, ###################### # nodepos table # ###################### {attr_name => 'rack', tabentry => 'nodepos.rack', access_tabentry => 'nodepos.node=attr:node', }, {attr_name => 'unit', tabentry => 'nodepos.u', access_tabentry => 'nodepos.node=attr:node', }, {attr_name => 'chassis', tabentry => 'nodepos.chassis', access_tabentry => 'nodepos.node=attr:node', }, {attr_name => 'slot', tabentry => 'nodepos.slot', access_tabentry => 'nodepos.node=attr:node', }, {attr_name => 'room', tabentry => 'nodepos.room', access_tabentry => 'nodepos.node=attr:node', }, ###################### # vm table # ###################### {attr_name => 'vmhost', tabentry => 'vm.host', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'migrationdest', tabentry => 'vm.migrationdest', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmstorage', tabentry => 'vm.storage', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmstoragemodel', tabentry => 'vm.storagemodel', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmcfgstore', tabentry => 'vm.cfgstore', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmmemory', tabentry => 'vm.memory', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmcpus', tabentry => 'vm.cpus', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmnics', tabentry => 'vm.nics', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmnicnicmodel', tabentry => 'vm.nicmodel', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmbootorder', tabentry => 'vm.bootorder', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmvirtflags', tabentry => 'vm.virtflags', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmvncport', tabentry => 'vm.vncport', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmtextconsole', tabentry => 'vm.textconsole', access_tabentry => 'vm.node=attr:node', }, {attr_name => 'vmbeacon', tabentry => 'vm.beacon', access_tabentry => 'vm.node=attr:node', }, ###################### # websrv table # ###################### {attr_name => 'webport', only_if => 'nodetype=websrv', tabentry => 'websrv.port', access_tabentry => 'websrv.node=attr:node', }, {attr_name => 'username', only_if => 'nodetype=websrv', tabentry => 'websrv.username', access_tabentry => 'websrv.node=attr:node', }, {attr_name => 'password', only_if => 'nodetype=websrv', tabentry => 'websrv.password', access_tabentry => 'websrv.node=attr:node', }, ###################### # switch table # ###################### {attr_name => 'switch', tabentry => 'switch.switch', access_tabentry => 'switch.node=attr:node', }, {attr_name => 'switchport', tabentry => 'switch.port', access_tabentry => 'switch.node=attr:node', }, {attr_name => 'switchvlan', tabentry => 'switch.vlan', access_tabentry => 'switch.node=attr:node', }, {attr_name => 'switchinterface', tabentry => 'switch.interface', access_tabentry => 'switch.node=attr:node', }, #################### # hosts table # #################### {attr_name => 'ip', tabentry => 'hosts.ip', access_tabentry => 'hosts.node=attr:node', }, {attr_name => 'hostnames', tabentry => 'hosts.hostnames', access_tabentry => 'hosts.node=attr:node', }, {attr_name => 'otherinterfaces', tabentry => 'hosts.otherinterfaces', access_tabentry => 'hosts.node=attr:node', }, #################### # prescripts table# #################### {attr_name => 'prescripts-begin', tabentry => 'prescripts.begin', access_tabentry => 'prescripts.node=attr:node', }, {attr_name => 'prescripts-end', tabentry => 'prescripts.end', access_tabentry => 'prescripts.node=attr:node', }, ); # end of @nodeattrs that applies to both nodes and groups #################### # node definition - nodelist & hosts table parts # #################### @{$defspec{node}->{'attrs'}} = ( #################### # nodelist table # #################### {attr_name => 'node', tabentry => 'nodelist.node', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'groups', tabentry => 'nodelist.groups', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'status', tabentry => 'nodelist.status', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'statustime', tabentry => 'nodelist.statustime', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'appstatus', tabentry => 'nodelist.appstatus', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'appstatustime', tabentry => 'nodelist.appstatustime', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'primarysn', tabentry => 'nodelist.primarysn', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'usercomment', tabentry => 'nodelist.comments', access_tabentry => 'nodelist.node=attr:node', }, {attr_name => 'hidden', tabentry => 'nodelist.hidden', access_tabentry => 'nodelist.node=attr:node', }, ); # add on the node attrs from other tables push(@{$defspec{node}->{'attrs'}}, @nodeattrs); ######################### # osimage data object # ######################### @{$defspec{osimage}->{'attrs'}} = ( {attr_name => 'imagename', tabentry => 'osimage.imagename', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'imagetype', tabentry => 'osimage.imagetype', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'provmethod', tabentry => 'osimage.provmethod', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'rootfstype', only_if => 'imagetype=linux', tabentry => 'osimage.rootfstype', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'profile', tabentry => 'osimage.profile', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'osname', tabentry => 'osimage.osname', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'osvers', tabentry => 'osimage.osvers', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'osdistro', tabentry => 'osimage.osdistro', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'osarch', tabentry => 'osimage.osarch', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'synclists', tabentry => 'osimage.synclists', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'postscripts', tabentry => 'osimage.postscripts', access_tabentry => 'osimage.imagename=attr:imagename', }, {attr_name => 'postbootscripts', tabentry => 'osimage.postbootscripts', access_tabentry => 'osimage.imagename=attr:imagename', }, #################### # linuximage table# #################### {attr_name => 'template', only_if => 'imagetype=linux', tabentry => 'linuximage.template', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'pkglist', only_if => 'imagetype=linux', tabentry => 'linuximage.pkglist', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'pkgdir', only_if => 'imagetype=linux', tabentry => 'linuximage.pkgdir', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'otherpkglist', only_if => 'imagetype=linux', tabentry => 'linuximage.otherpkglist', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'otherpkgdir', only_if => 'imagetype=linux', tabentry => 'linuximage.otherpkgdir', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'exlist', only_if => 'imagetype=linux', tabentry => 'linuximage.exlist', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'postinstall', only_if => 'imagetype=linux', tabentry => 'linuximage.postinstall', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'rootimgdir', only_if => 'imagetype=linux', tabentry => 'linuximage.rootimgdir', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'kerneldir', only_if => 'imagetype=linux', tabentry => 'linuximage.kerneldir', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'nodebootif', only_if => 'imagetype=linux', tabentry => 'linuximage.nodebootif', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'otherifce', only_if => 'imagetype=linux', tabentry => 'linuximage.otherifce', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'netdrivers', only_if => 'imagetype=linux', tabentry => 'linuximage.netdrivers', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'kernelver', only_if => 'imagetype=linux', tabentry => 'linuximage.kernelver', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'krpmver', only_if => 'imagetype=linux', tabentry => 'linuximage.krpmver', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'permission', only_if => 'imagetype=linux', tabentry => 'linuximage.permission', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'dump', only_if => 'imagetype=linux', tabentry => 'linuximage.dump', access_tabentry => 'linuximage.imagename=attr:imagename', }, {attr_name => 'crashkernelsize', only_if => 'imagetype=linux', tabentry => 'linuximage.crashkernelsize', access_tabentry => 'linuximage.imagename=attr:imagename', }, #################### # nimimage table# #################### {attr_name => 'nimtype', only_if => 'imagetype=NIM', tabentry => 'nimimage.nimtype', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'nimmethod', only_if => 'imagetype=NIM', tabentry => 'nimimage.nimmethod', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'lpp_source', only_if => 'imagetype=NIM', tabentry => 'nimimage.lpp_source', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'spot', only_if => 'imagetype=NIM', tabentry => 'nimimage.spot', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'root', only_if => 'imagetype=NIM', tabentry => 'nimimage.root', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'dump', only_if => 'imagetype=NIM', tabentry => 'nimimage.dump', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'paging', only_if => 'imagetype=NIM', tabentry => 'nimimage.paging', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'resolv_conf', only_if => 'imagetype=NIM', tabentry => 'nimimage.resolv_conf', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'image_data', only_if => 'imagetype=NIM', tabentry => 'nimimage.image_data', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'tmp', only_if => 'imagetype=NIM', tabentry => 'nimimage.tmp', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'home', only_if => 'imagetype=NIM', tabentry => 'nimimage.home', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'shared_home', only_if => 'imagetype=NIM', tabentry => 'nimimage.shared_home', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'shared_root', only_if => 'imagetype=NIM', tabentry => 'nimimage.shared_root', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'script', only_if => 'imagetype=NIM', tabentry => 'nimimage.script', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'fb_script', only_if => 'imagetype=NIM', tabentry => 'nimimage.fb_script', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'bosinst_data', only_if => 'imagetype=NIM', tabentry => 'nimimage.bosinst_data', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'installp_bundle', only_if => 'imagetype=NIM', tabentry => 'nimimage.installp_bundle', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'otherpkgs', only_if => 'imagetype=NIM', tabentry => 'nimimage.otherpkgs', access_tabentry => 'nimimage.imagename=attr:imagename', }, {attr_name => 'mksysb', only_if => 'imagetype=NIM', tabentry => 'nimimage.mksysb', access_tabentry => 'nimimage.imagename=attr:imagename', }, # {attr_name => 'res_group', # tabentry => 'nimimage.res_group', # access_tabentry => 'nimimage.imagename=attr:imagename', # }, {attr_name => 'usercomment', only_if => 'imagetype=NIM', tabentry => 'nimimage.comments', access_tabentry => 'nimimage.imagename=attr:imagename', }, ); ######################### # network data object # ######################### # networks table # ######################### @{$defspec{network}->{'attrs'}} = ( ### # TODO: when creating networks table entries, create a default netname # See makenetworks command and networks.pm plugin ### {attr_name => 'netname', tabentry => 'networks.netname', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'net', tabentry => 'networks.net', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'mask', tabentry => 'networks.mask', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'mgtifname', tabentry => 'networks.mgtifname', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'gateway', tabentry => 'networks.gateway', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'dhcpserver', tabentry => 'networks.dhcpserver', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'tftpserver', tabentry => 'networks.tftpserver', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'nameservers', tabentry => 'networks.nameservers', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'dynamicrange', tabentry => 'networks.dynamicrange', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'vlanid', tabentry => 'networks.vlanid', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'domain', tabentry => 'networks.domain', access_tabentry => 'networks.netname=attr:netname', }, {attr_name => 'usercomment', tabentry => 'networks.comments', access_tabentry => 'networks.netname=attr:netname', }, ); ######################### # route data object # ######################### # routes table # ######################### @{$defspec{route}->{'attrs'}} = ( {attr_name => 'routename', tabentry => 'routes.routename', access_tabentry => 'routes.routename=attr:routename', }, {attr_name => 'net', tabentry => 'routes.net', access_tabentry => 'routes.routename=attr:routename', }, {attr_name => 'mask', tabentry => 'routes.mask', access_tabentry => 'routes.routename=attr:routename', }, {attr_name => 'gateway', tabentry => 'routes.gateway', access_tabentry => 'routes.routename=attr:routename', }, {attr_name => 'ifname', tabentry => 'routes.ifname', access_tabentry => 'routes.routename=attr:routename', }, {attr_name => 'usercomment', tabentry => 'routes.comments', access_tabentry => 'routes.routename=attr:routename', }, ); ##################### # site data object # ##################### # site table # ##################### ############## # TODO: need to figure out how to handle a key for the site table. # since this is really implemented differently than all the other # data objects, it doesn't map as cleanly. # change format of site table so each column is an attr and there # is only a single row in the table keyed by xcatmaster name? ############# @{$defspec{site}->{'attrs'}} = ( {attr_name => 'master', tabentry => 'site.value', access_tabentry => 'site.key=str:master', description => 'The management node'}, {attr_name => 'installdir', tabentry => 'site.value', access_tabentry => 'site.key=str:installdir', description => 'The installation directory'}, {attr_name => 'xcatdport', tabentry => 'site.value', access_tabentry => 'site.key=str:xcatdport', description => 'Port used by xcatd daemon on master'}, ); ####################### # groups data object # ####################### # groups table # ####################### @{$defspec{group}->{'attrs'}} = ( {attr_name => 'groupname', tabentry => 'nodegroup.groupname', access_tabentry => 'nodegroup.groupname=attr:groupname', }, {attr_name => 'grouptype', tabentry => 'nodegroup.grouptype', access_tabentry => 'nodegroup.groupname=attr:groupname', }, {attr_name => 'members', tabentry => 'nodegroup.members', access_tabentry => 'nodegroup.groupname=attr:groupname', }, {attr_name => 'wherevals', tabentry => 'nodegroup.wherevals', access_tabentry => 'nodegroup.groupname=attr:groupname', }, {attr_name => 'usercomment', tabentry => 'nodegroup.comments', access_tabentry => 'nodegroup.groupname=attr:groupname', }, ### # TODO: Need to copy attrs that are common between nodes and static groups # Ideas: make a separate data structure that is linked/copied here. # need to figure out the perl dereferencing to make that work. ### ); # add on the generic node attrs push(@{$defspec{group}->{'attrs'}}, @nodeattrs); ####################### # policy data object # ####################### # policy table # ####################### @{$defspec{policy}->{'attrs'}} = ( ### # TODO: The policy validate subroutine in the xcatd daemon code does not # sort the rules in the policy table in priority order before # processing. Talk to Jarrod - I think it should. ### {attr_name => 'priority', tabentry => 'policy.priority', access_tabentry => 'policy.priority=attr:priority', }, {attr_name => 'name', tabentry => 'policy.name', access_tabentry => 'policy.priority=attr:priority', }, {attr_name => 'host', tabentry => 'policy.host', access_tabentry => 'policy.priority=attr:priority', }, {attr_name => 'commands', tabentry => 'policy.commands', access_tabentry => 'policy.priority=attr:priority', }, {attr_name => 'noderange', tabentry => 'policy.noderange', access_tabentry => 'policy.priority=attr:priority', }, {attr_name => 'parameters', tabentry => 'policy.parameters', access_tabentry => 'policy.priority=attr:priority', }, {attr_name => 'time', tabentry => 'policy.time', access_tabentry => 'policy.priority=attr:priority', }, {attr_name => 'rule', tabentry => 'policy.rule', access_tabentry => 'policy.priority=attr:priority' , }, {attr_name => 'usercomment', tabentry => 'policy.comments', access_tabentry => 'policy.priority=attr:priority', }, ); ############################# # notification data object # ############################# # notification table # ############################# @{$defspec{notification}->{'attrs'}} = ( {attr_name => 'filename', tabentry => 'notification.filename', access_tabentry => 'notification.filename=attr:filename', }, {attr_name => 'tables', tabentry => 'notification.tables', access_tabentry => 'notification.filename=attr:filename', }, {attr_name => 'tableops', tabentry => 'notification.tableops', access_tabentry => 'notification.filename=attr:filename', }, {attr_name => 'comments', tabentry => 'notification.comments', access_tabentry => 'notification.filename=attr:filename', }, ); ########################### # monitoring data object # ########################### # monitoring table # ########################### @{$defspec{monitoring}->{'attrs'}} = ( {attr_name => 'name', tabentry => 'monitoring.name', access_tabentry => 'monitoring.name=attr:name', }, {attr_name => 'nodestatmon', tabentry => 'monitoring.nodestatmon', access_tabentry => 'monitoring.name=attr:name', }, {attr_name => 'comments', tabentry => 'monitoring.comments', access_tabentry => 'monitoring.name=attr:name', }, {attr_name => 'disable', tabentry => 'monitoring.disable', access_tabentry => 'monitoring.name=attr:name', }, ); @{$defspec{eventlog}->{'attrs'}} = ( {attr_name => 'recid', tabentry => 'eventlog.recid', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'eventtime', tabentry => 'eventlog.eventtime', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'eventtype', tabentry => 'eventlog.eventtype', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'monitor', tabentry => 'eventlog.monitor', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'monnode', tabentry => 'eventlog.monnode', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'node', tabentry => 'eventlog.node', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'application', tabentry => 'eventlog.application', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'component', tabentry => 'eventlog.component', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'id', tabentry => 'eventlog.id', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'severity', tabentry => 'eventlog.severity', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'message', tabentry => 'eventlog.message', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'rawdata', tabentry => 'eventlog.rawdata', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'comments', tabentry => 'eventlog.comments', access_tabentry => 'eventlog.recid=attr:recid', }, {attr_name => 'disable', tabentry => 'eventlog.disable', access_tabentry => 'eventlog.recid=attr:recid', }, ); @{$defspec{auditlog}->{'attrs'}} = ( {attr_name => 'recid', tabentry => 'auditlog.recid', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'audittime', tabentry => 'auditlog.audittime', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'userid', tabentry => 'auditlog.userid', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'clientname', tabentry => 'auditlog.clientname', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'clienttype', tabentry => 'auditlog.clienttype', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'command', tabentry => 'auditlog.command', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'noderange', tabentry => 'auditlog.noderange', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'args', tabentry => 'auditlog.args', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'status', tabentry => 'auditlog.status', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'comments', tabentry => 'auditlog.comments', access_tabentry => 'auditlog.recid=attr:recid', }, {attr_name => 'disable', tabentry => 'auditlog.disable', access_tabentry => 'auditlog.recid=attr:recid', }, ); @{$defspec{firmware}->{'attrs'}} = ( { attr_name => 'cfgfile', tabentry => 'firmware.cfgfile', access_tabentry => 'firmware.file=attr:cfgfile', }, {attr_name => 'comments', tabentry => 'firmware.comments', access_tabentry => 'firmware.file=attr:cfgfile', }, {attr_name => 'disable', tabentry => 'auditlog.disable', access_tabentry => 'firmware.file=attr:cfgfile', }, ); ################################################### # adding user defined external defspec ################################################## foreach my $objname (keys(%xCAT::ExtTab::ext_defspec)) { if (exists($xCAT::ExtTab::ext_defspec{$objname}->{'attrs'})) { if (exists($defspec{$objname})) { my @extattr=@{$xCAT::ExtTab::ext_defspec{$objname}->{'attrs'}}; my @attr=@{$defspec{$objname}->{'attrs'}}; my %tmp_hash=(); foreach my $orig (@attr) { my $attrname=$orig->{attr_name}; $tmp_hash{$attrname}=1; } foreach(@extattr) { my $attrname=$_->{attr_name}; if (exists($tmp_hash{$attrname})) { xCAT::MsgUtils->message('ES', "\n Warning: Conflict when adding user defined defspec. Attribute name $attrname is already defined in object $objname. \n"); } else { push(@{$defspec{$objname}->{'attrs'}}, $_); } } } else { $defspec{$objname}=$xCAT::ExtTab::ext_defspec{$objname}; } } } #print "\ndefspec:\n"; #foreach(%xCAT::Schema::defspec) { # print " $_:\n"; # my @attr=@{$xCAT::Schema::defspec{$_}->{'attrs'}}; # foreach my $h (@attr) { # print " " . $h->{attr_name} . "\n"; # } #} # Build a corresponding hash for the attribute names to make # definition access easier foreach (keys %xCAT::Schema::defspec) { my $dataobj = $xCAT::Schema::defspec{$_}; my $this_attr; foreach $this_attr (@{$dataobj->{'attrs'}}){ $dataobj->{attrhash}->{$this_attr->{attr_name}} = $this_attr; } }; 1;