Back to oar_2

OAR 2.6.1

Release date: 2025-03-24

Changelog

  • [pam_oar_adopt] Make it easier to install and configure the oar-node PAM profile
  • [pam_oar_adopt] Fix and improve the script
  • [pam_oar_adopt] Move pam_oar_adopt back to the oar-node package
  • [pam_oar_adopt] Rework man page
  • [job_resource_manager_systemd/pam_oar_adopt] Make OAR env compatible with pam_env
  • [job_resource_manager_systemd] 1 liner log message for tmp clean-up
  • [job_resource_manager_systemd/oarsh] Fix cgroup v2 path extraction
  • [setup] cleanup script
  • [oarsh_shell] Add support for cgroupv2/systemd and use hwloc
  • [job_resource_manager] Add support for cgroupv2/systemd and use hwloc
  • [oarcgde] Add the oarcgdev tool to manage devices blacklist in cgroupv2
  • [suspend_resume_manager] Add support for cgroupv2/systemd
  • [Monika] Fix display of wanted resources
  • [all] Fix for Perl exporter (Perl 5.39.1)
  • [oarsub] Ignore ^Z from terminal (SIGTSTP)
  • [API] Remove cycle limits
  • [IO.pm] Fix issue with resource_log
  • [IO.pm/MetaSched] Add functions to allow scheduling perf statistics
  • [oaraccounting] Now use a systemd timer
  • [pam_adopt_oar] Add PAM script to enable SSH connection to jobs
  • [oar_resources_init] Now use hwloc to find resources on nodes
  • [oar_resources_add] Add support for resources in the hwloc format
  • [man] Move administrator's commands to section 8 and some rewritings
  • [cron] Provide both cron and systemd timers (oaraccounting, oarnodecheck)
  • [oarnodecheck] Add support for cgroupv2/systemd, refactor

Starting from OAR 2.6, a new mechanism using cgroupv2/systemd and hwloc is in place to map OAR resources to physical compute resources (cores, GPUs) on nodes and manage the processes of jobs.

As a consequence, on an existing installation, the cpuset property of the OAR resources may need some changes to adopt the hwloc description of the machines. The oar_resource_init script can be used to look at the expected values for the cpuset resource property.

Regarding the gpudevice resource property, it must now contain the list of special devices (e.g., /dev/nvidia0) associated with a resource.

Finally, cgroupv1 activation is not needed anymore, only the unified cgroupv2 file hierarchy is required now. Thus, cgroupv1-related directives in the kernel command line can be removed on OAR nodes.

oar_2.6.1.txt · Last modified: by neyron
Recent changes RSS feed GNU Free Documentation License 1.3 Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki