Menu

flux-core v0.38.0

Published: Apr 4, 2022 by flux-framework

Download from GitHub here

Release Notes

This release makes a few improvements that are visible in the flux-mini(1) command:

  • The -N,--nnodes option may be used without the -n,--nprocs option.
  • The --exclusive option requests the allocation of whole nodes.
  • The --requires option requests resources with generic properties.

Additionally, Flux system administrators should be aware of these changes:

  • Named properties may be assigned to resources in the configured R.
  • flux-shutdown(1) is now the preferred way to stop a Flux system instance.
  • The default archive.dbpath is now /var/lib/flux/job-archive.sqlite.
  • systemd-coredump(8) can now capture a Flux broker core dump. Consider enabling this on the management nodes of early access systems to help gather information on critical rank 0 broker failures, should they occur.
  • flux resource drain now shows nodes as “draining” if they are still running work.
  • Flux may be configured to reject jobs that do not run in a Flux sub-instance.

For more information, see the Flux Administrator’s Guide:

https://flux-framework.readthedocs.io/en/latest/adminguide.html

New Features

  • add flux-shutdown command (#4250)
  • add flux-dump and flux-restore (#4208, #4225)
  • support for node exclusive allocations (#4245)
  • add support for resource properties (#4236)
  • flux-resource: support properties when listing resources (#4249)
  • job-ingest: add TOML config (#4238)
  • flux-dmesg: add –new option, plus logging cleanup (#4237)
  • add ‘require-instance’ job validator plugin (#4239)
  • job-manager: add builtin job duration validator plugin (#4224)
  • sched-simple: set expiration of jobs with no duration to instance lifetime (#4223)
  • flux-resource: differentiate drained vs draining ranks (#4205)
  • librlist: support hwloc discovery of AMD RSMI GPUs (#4203)
  • broker: reject remote exec requests on rank 0 (#4258)
  • python: allow resource count of 0 in jobspec v1 (#4259)
  • job-archive: use statedir path if dbpath is not set (#4260)

Fixes

  • content-sqlite: ensure that module load fails if initialization fails (#4265)
  • job-archive: use statedir path if dbpath not set (#4260)
  • broker: emit error when running interactive shell without tty (#4253)
  • broker: add statedir attribute, drop content.backing-path (#4248)
  • broker: prevent systemd restart if rc1 fails (#4246)
  • flux.service: use StateDirectory for content.sqlite (#4244)
  • rc3: ensure exit code reflects any errors (#4243)
  • broker: don’t leave shutdown state prematurely (#4241)
  • libjob: improve flux_job_statetostr(), flux_job_resulttostr() interface (#4235)
  • job-list: fix bugs in some error paths (#4233)
  • broker: fine tune logging and enable core dump on SIGSEGV (#4231)
  • kvs: always store empty directory object to content store (#4229)
  • restrict access to content service used as KVS blob layer (#4216)
  • content-sqlite: check that file has rw permission (#4215)
  • broker: block SIGPIPE (#4211)
  • shell: add hostname to a couple key log messages (#4200)
  • python: add missing methods and improve efficiency of IDset class (#4209)
  • systemd: set SyslogIdentifier to flux (#4206)
  • misc minor fixes and cleanup (#4197)
  • job-exec: fix more potential hangs after early shell failure (#4199)
  • sched-simple: fix counting bug that can cause scheduler to fail after a restart (#4196)
  • flux-top: add man page, minor bug fixes (#4194)

Cleanup

  • broker: clean up shutdown logs (#4257)
  • libsdprocess: minor fixups (#4252)
  • job-manager: misc cleanup (#4232)

Testsuite

  • testsuite: fix a couple intermittent test failures (#4247)
  • ci: run 32bit build under linux32 personality (#4240)
  • testsuite: ensure tests can run concurrently with --root=$FLUX_JOB_TMPDIR (#4212)