Menu

flux-accounting v0.63.0

Published: Nov 3, 2026 by flux-framework

Download from GitHub here

Release Notes

flux-accounting version 0.63.0 - 2026-11-03

Features

  • job usage: add a “continuous” usage decay mode (#996)

  • cmd: add new estimate-priority command (#1006)

  • view-usage-report: add byjobname grouping (#1010)

Fixes

  • plugin: remove max_nodes_per_assoc as dependency (#993)

  • plugin: fix queue priority bug, separate queue validation from queue priority access (#1002)

  • usage: don’t double-count past usage in period 0 (#1004)

  • plugin: check for bank attribute during rediscovery of active jobs after reload (#1008)

  • edit-bank: convert --ignore-older-than to timestamps instead of storing raw value (#1012)

  • plugin: return lowest possible runnable priority for non-held jobs (#1014)

Testsuite

  • t: check for flux-security before running sharness (#990)

  • ci: drop Python 3.9 from CI, upgrade pylint (#1008)

  • testsuite: fix race in t1048-issue575.t (#1015)

  • github: bump crate-ci/typos from 1.50.3 to 1.51.1 in the github-actions group across 1 directory (#1017)

important note for administrators: queue node limits

The max_nodes_per_assoc queue property is deprecated in this release and is no longer enforced by the multi-factor priority plugin. Previously, this property added a max-resources-queue dependency when an association’s aggregate node usage in a queue reached the configured limit. The property is still accepted by the queue commands, retained in the flux-accounting database and JSON payload, and included in flux jobtap query mf_priority.so output for compatibility, but changing it no longer affects scheduling policy.

Aggregate node usage by an association across jobs in SCHED and RUN is now enforced exclusively by max_sched_nodes_per_assoc. This property also provides the per-job validation check: a job requesting more nodes than the configured value is rejected during validation.

Existing queue configuration is not migrated automatically. Before deploying this release, administrators who use max_nodes_per_assoc should:

  1. Review the configured value for each queue.
  2. Set max_sched_nodes_per_assoc to the desired aggregate SCHED and RUN node limit with flux account edit-queue QUEUE --max-sched-nodes-per-assoc=VALUE.
  3. Run flux account-priority-update to send the updated queue configuration to the running priority plugin.
  4. Optionally reset max_nodes_per_assoc to unlimited once local tooling no longer depends on the legacy field.

The plugin no longer creates new max-resources-queue dependencies. During a plugin reload or job-manager restart, it makes a best-effort attempt to remove this dependency from jobs that were already in DEPEND under an older plugin version. No database schema, queue command option, accounting RPC payload, or plugin query field is removed by this change.