Base Command Manager / Bright Cluster Manager Release Notes
Release notes for Bright 6.0-21 (released 2012-10-26)
buildmaster: Default disk layout for nodes now uses 16G instead of 2G for /var.
buildmaster: package needed for hvm machines in the cloud
buildmaster: Added ext2 module to slave node kernel module list, for the case a ext2 fs is configured for a slave node, for sles11 and rhel6 based distros.
buildmaster: remote-install: read ofedstack,hwvendor from buildconfig
buildmaster: remote-install: fix grub.conf writing in SLES
cluster-tools: wait for service stop and manually stop in wlm-setup workload manager disable
cluster-tools: cm-create-image: document --mininal option in usage
cluster-tools: cm-register-node:provide option (-u) to only update node license
cmburn: Added a failsafe to make sure post install burn detects if burn was already complete.(Should normally not happen.)
cmburn: Changed bblocks test so that it uses far less disk space in the burn spool directory.
cmburn: Not needed to spawn a process per core anymore as this is now done by mprime itself (threads)
cmburn: Making use of newest mprime, 64bits
cmburn: making compile less chatty
cmdaemon: Failed to update node: billing service is reporting an error or the number of nodes for which a license is required has exceeded the pre-paid node count in your license
cmdaemon: cmsh prints a warning message if the user homedirectory cannot be created when running commit while adding a new user.
cmdaemon: add extra tokens to readonly profile
cmdaemon: add nfs-rdma service to check if it's needed
cmdaemon: conman only on active head
cmdaemon: log shutdown/reboot from cmdaemon
cmdaemon: Burn status now also includes total burn time, in addition to phase time.
cmdaemon: Don't flush monitoring data to other head node on stop of cmdaemon, as this can cause issues
cmdaemon: When services are stopped in cmsh/cmgui: don't restart them when a config file is updated
cmdaemon: Improve cmdaemon-backup: don't make a backup if msqyl is not running
cmdaemon: force reload of conman when node pass through the node installer, so conman tries to reconnect
cmdaemon: don't mark-stop conman when it's not need
cmdaemon: Measurement unit -> Metric unit
cmdaemon: Fixed an issue in the provisioning subsystem. When a request came in at a specific time, the main scheduling loop would not get triggered. This could cause the subsystem to remain idle until some other request happened to come in.
cmgui: add start time to monitoring job groups
cmgui: make it easy to create a job filter #2898
cmgui: Option for delayed monitoring tree fill
cmgui: fix finalize script sles cloud
cmgui: fix model check
cmgui: fix monitoring model check
cmgui: total/phase time burn
cmgui: fix duplicate monconf for generic devices
cmgui: extra thread for monitoring load
cmgui: improve display of unit in graphs
cmgui: improve fill for parallel shell
cmgui: add option to only display single page pexec output
finalise-base: Add 0.9 metadata type to auto-assemble line of mdadm.conf, if file/line exists, and if not already configured.
finalise-base: Set cmhealth UID explicitly the same on a head node and in a software image
manuals: added: --basedistrepo --cmrepo --minimal options to cm-create-image command
manuals: added: burn log relocation notes
manuals: added: note about Externally visible IP addresses
manuals: added: additionalhostnames note
manuals: added: note on InfiniBand MTU recommendation
manuals: CUDA 5 upgrade notes
mpich: do not load gcc module on RHEL6 for fortran tests
mvapich: unload gcc recent module for fortran tests
node-installer: Post installer burn is invoked from inittab/upstart config files. We relied on provisioning reverting these files to the default contents when burn completed. This resulted in a problem when the installmode is set to NOSYNC. Recent changes to other burn related scripts caused this to result in an endless reboot loop when burn completes and NOSYNC is used. This fix has the node installer backup/restore the original inittab/upstart config files, regardless of the install mode.
node-installer: Further improvements to motd handling during burn.
node-installer: fixed the issue of /etc/resolv.conf not being properlu written to on SLES cloud nodes
node-installer: Fix to the scenario in which cloud nodes instances attempted to identify themselves using port detection. This happened only during the very first boot of the spot instance. The instance tried to identify itself with the CMDaemon using instance ID, but the CMDamon was not YET aware of the instanceID for this node. Therefore, the node tried to use port detection. With this fix the node continuesly retries to identify itself using the instance ID, until it succeeds, or until the retry limit is reached (at which point the installer fails).
node-installer: In the interactive prompts in the node installer now only a subset of keys cancels the timeout. Previously pressing any key would cancel timeout, which means that timeout cancels were more likely to have been caused by faulty hardware/console input. Now the set keys pressing which cancels timeout includes arrow keys and pageUP/DOWN keys.
node-installer-slave: fix clone install RHEL6 check to support RHEL6u2
node-installer-slave: Updated kernel-update.log, kernel dist package got updated
node-installer-slave: Updated dependencies-rev.log, revision of package changed, on which node-installer-slave depends.
node-installer-slave: Updated dependencies-rev.log, revision of package changed, on which node-installer-slave depends.
node-installer-slave: Updated kernel-update.log, kernel dist package got updated
web-burntool: Fixed broken history search. And fixed compatiability issue with new burn status format.