# # # Sample stanza file used as input to the rollupdate command # This sample assumes that you have a simple set of updates that can # be applied to all nodes in any order. Only the list of all nodes needs # to be specified. You do not need to specify individual update groups. # # For more complex updates, see the sample file rollupdate.input.sample # # Unless otherwise noted in the descriptions below, if multiple stanza # lines are specified for the same keyword, only the FIRST entry will be # used and all others will be ignored. # # # scheduler: The job scheduler used to submit the rolling update jobs # Currently only LoadLeveler is supported scheduler=loadleveler # scheduser: The userid with authority to submit scheduler jobs # Note: LL does not allow jobs to be submitted by root scheduser=loadl # updateall: Allows you to apply updates to all nodes without specifying # specific update groups. This should be used for simple compute node # updates that have no special dependencies on other nodes and update order # is not important. Only those nodes that are currently active in # in the scheduler will be updated. # For LoadLeveler reservation jobs, you can control which nodes will be # updated by specifying different machine requirements, classes, or other # controls that LL allows you in the LL job template that you provide. # If updateall is used, the following entries MUST be specified in this file: # updateall_nodes # updateall_nodecount # job_template # job_dir # If updateall is used, the following entries can also be used: # maxupdates # oldfeature # newfeature # reservationduration # update_if_down # prescript (but NO prescriptnodes) # shutdowntimeout # outofbandcmd (but NO outofbandnodes) # bringupstatus,bringupappstatus # bringuptimeout # All other entries in this file will be ignored updateall=yes # updateall_nodes: The xCAT noderange of the nodes to be updated. updateall_nodes=compute # updateall_nodecount: The number of nodes that will be reserved at one time # in the scheduler and updated together. The smaller the number, the more # scheduler reservation jobs that will be submitted. # NOTE: LoadLeveler performance decreases with large numbers of # reservations. Do not set this value so low that you will exceed # the maximum number of reservations allowed for your cluster or # that you will degrade LL performance for your production jobs. updateall_nodecount=3 # translatenames: # If your scheduler will be using names for nodes that are different from # xCAT node names (e.g. the scheduler is using a different administrative # network), you will need to tell xCAT how to translate from xCAT node names # to the node names registered with your scheduler. # # Syntax: # translatenames=<xCAT_noderange>:/<pattern>/<replacement>/ # where <pattern> and <replacement> are perl regular expressions to # be performed on the node names in <xCAT_noderange>. # See the xcatdb man page for more details on using regular expressions. # Multiple translatenames statements are allowed. If an xCAT nodename # exists in more than one xCAT_noderange, the last translated value # will be used. #translatenames=service:|bb(\d+)s(\d+)|bb($1)sn($2)| #translatenames=compute:/\z/-hf2/ # Scheduler Feature values # Node feature values that will be changed in the scheduler during the # update process. These features can be used if users need to run jobs # on only old nodes or only new nodes, but cannot have the job span both # old and new nodes due to software incompatibilities. # oldfeature: This feature value will be removed from the node definition # in the scheduler after the node has been updated oldfeature=oldvalue # newfeature: A new feature value that will be set in the scheduler for each # node after it has been updated. newfeature=newvalue # maxupdates: Maximum number of updategroups that can be updated at one time # This allows you to ensure you will always have enough computing # resources in your cluster and that not all nodes will attempt to # be updated at once. # A value of "all" specifies that there is no restriction #maxupdates=all maxupdates=16 # reconfiglist: For LoadLeveler, the list of nodes (as known by xCAT) # that xCAT will xdsh an 'llctl reconfig' command to. xCAT will always send # the reconfig command to the local xCAT management node, and to all nodes # listed as the LL central managers and LL resource managers in the LL # database. This is a list of additional machines required to immediately see # any database changes xCAT may make. For example, all LL submit-only nodes # should be added to this list so that any machine FEATURE changes are # visible for job submission. #reconfiglist=login1,login2 #jobtemplate: Scheduler job template file. See this sample LoadLeveler file # for details on how the template will be processed: jobtemplate=/opt/xcat/share/xcat/rollupdate/llall.tmpl #jobdir: Directory to write the job command files to # For LL, this directory needs to be on a filesystem available to all nodes. jobdir=/u/loadl/rollupdate_jobs #reservationcallback: INTERNAL KEYWORD used for development only. # This is the reservation notify or callback command. # For Loadleveler, this script must reside on the LoadLeveler central # manager and will be called when the reservation for an updategroup # becomes active. # The default is: #reservationcallback=/opt/xcat/bin/runrollupdate # #reservationduration: Maximum time to hold a LoadLeveler reservation for # the update process. This value in minutes should be longer than the # expected time to shutdown, update, and reboot all the nodes in an update # group. xCAT will release the nodes from the reservation as they come # back up, and will cancel the reservation when the last node has completed. reservationduration=15 #update_if_down: If set to "yes", also attempt the update for any node in an # updategroup that is down or not active/available in the scheduler # (useful if you have nodes that are not part of your scheduler's cluster). # If set to "no", any node in an updategroup that is not active in the # scheduler will be skipped. # If set to "cancel", if any node in an updategroup is not active, # the entire updategroup will be skipped. # NOTE: For updateall operation, update_if_down must be "no" or "cancel". # If set to "yes", "cancel" will be assumed. #update_if_down=yes update_if_down=no #update_if_down=cancel # prescript # (optional) Command to be run on the xCAT management node before issuing # the shutdown command for the nodes in the updategroup. # NOTE: For updateall operation, prescriptnodes are not supported. # Prescripts will be run for all nodes. # You may specify the string $NODELIST if you would like the # comma-delimited list of xCAT nodenames passed into your command. # Multiple prescript entries can be specified. Each command will be run in # order. #prescript=/u/admin/bin/shutdownGPFS $NODELIST # skipshutdown: Should a shutdown command be sent to the nodes. # Shutdown is required for diskless nodes. For diskfull nodes, simple # updates may be applied to the nodes through prescripts, and a node # reboot may not be required. # Default is "no" - a node shutdown will be performed. skipshutdown=no # shutdowntimeout: (optional) The number of minutes xCAT should wait for # an OS shutdown to complete before giving up and issuing a hard power off # command and continuing with the rolling update process. shutdowntimeout=5 # outofbandcmd # (optional) Out-of-band command to run on the xCAT management # node once the node has been shutdown but before it is rebooted. # NOTE: For updateall operation, outofbandnodes are not supported. # Out-of-band commands will be run for all nodes. # You may specify the string $NODELIST if you would like the # comma-delimited list of xCAT nodenames passed into your command. # Multiple outofbandcmd (with optional outofbandnodes) entries can be # specified. Each command will be run in order. #outofbandcmd=/u/admin/bin/myfirmwareupdates $NODELIST # bringupstatus: # OR # bringupappstatus: # The xCAT database node status or appstatus value that xCAT will check and # will wait for to determine that the node is up. Once this status is # reached, xCAT will continue bringing up more nodes (if bringuporder is set) # and will release this node from the scheduler reservation. # If both attributes are set, only bringupappstatus will be used. #bringupappstatus="gpfs=ready" bringupstatus=booted # bringuptimeout: (optional) The maximum number of minutes xCAT should wait # after issuing the rpower on command for the nodes to reach bringupstatus # before giving up. If using bringuporder and this timeout is reached for # one set of nodes, no additional nodes will be attempted to be brought up. # The scheduler reservation will be cancelled. bringuptimeout=10