- expect Rv1 from resource.acquire, use flux resource reload (#766)
- update configs for travis-ci.com migration (#769)
- Add support for nodelist for RV1 change and R_lite compression (#764)
- Add job shell plugin for data staging on tiered storage systems (#743)
- qmanager: remove
t_estimate
when job alloc succeeds (#746) - rc: avoid unloading sched-simple, which is no longer loaded by default (#749)
- rc: only load/unload scheduler on rank 0 (#741)
- makefile: workaround boost errors under latest compilers (#757)
Note: Fluxion can now annotate its jobs with their
start-time estimate and queue info (to be displayed by flux-jobs
).
- Handle invalid sibling requests in jobspecs more gracefully (#737)
- Fix various errors detected when tests are run under valgrind (#734)
- fluxion-resource: check return value of
find_first_not_of
(#715)
- Add support for job annotation including its start-time estimate (#735)
- Add support for node-local storage to hwloc reader (#714)
- flux module load returns early for staggered system-instance bring-up (#719)
- Update for non-integer jobid encodings (#718)
- Use version 9999 for test jobspecs (#739)
Note: Fluxion now uses flux-core's resource service to acquire resources to schedule.
- qmanager: support for ordering pending jobs by priority/submit time (#694)
- testsuite: increase wait-event timeouts (#692)
- testsuite: enable valgrind testing (#703)
- resource: errno support for std::ifstream open (#698)
- Undo compiler workarounds (#697)
- Add support for hwloc 2.0+ (#677)
- Augment multi-tiered storage scheduling test (#673)
- Integrate with flux-core's resource module (#665, #667, #674, #693, #694)
- Update for flux-core annotation support changes (#681)
- Update pkg dependency matrix in README.md (#696)
- Globally rename whitelist to allowlist (#690)
Note: flux-sched has been re-branded to "Fluxion"!
- testsuite: break a race in a qmgr reload test (#617)
- testsuite: avoid rc3 hang in qmanager-reload test (#593)
- Fix build/test issues and enable travis on CentOS 8 (#565)
- t4000-match-params.t: suppress flux module unload failure (#561)
- Fix t2002-qmanager-reload.t with new core resource module (#661)
- docker: add ENTRYPOINT to start munged by default (#648)
- testsuite: prevent python version mismatch in test scripts (#568)
- Add state recovery support (#663, #666)
- qmanager: add multi-queue support (#649)
- Add flux-tree (#622, #621, #616, #582)
- Add configuration file support (#599, #645, #630)
- Multi-tiered storage scheduling support demo and tests (#620)
- Update the project name to Fluxion (#655)
- qmanager: add exception safety for schedutil callbacks (#646)
- Add path, vertex pair to by_path map (#641)
- testsuite: update flux-sharness.sh from flux-core (#657)
- Use Jansson for JSON-based writers + time keys in RV1 (#614)
- rc: use reload -f to load qmanager/resource (#619)
- simplify build process and instructions (#603, #578, #595, #578, #579)
- testsuite: update flux-job list usage (#571)
- mergify: add config from flux-core (#562)
- github: add workflow action to validate PR commits (#553)
- drop flux module -r option (#557)
- resource/sched/simulator: update KVS API usage (#435)
- build: fix build issues for priority_mod_fair_tree.so (#434)
- api: sync flux-sched with removed interfaces in flux-core (#440)
- resource: update for kvs watch API removal (#442)
- API: remove the use of deprecated Python API (#450)
- resource: update to flux_respond_error() (#457)
- resource: bug fix for incorrectly handling implicit exclusivity (#502)
- libschedutil: remove vendored copy and use flux-core's exported lib (#516)
- testsuite: update flux-sharness.sh to version from flux-core (#523)
- rc3: Fix a bug in RC3 dir definition within configure.ac (#525)
- resource: fix buffer overflow when handling slot type in a jobspec (#548)
- qmanager: integrate with the new exec system (#481)
- qmanager: add hello/exception callback support (#493)
- qmanager: add EASY/HYBRID/CONSERVATIVE policies (#504)
- resource: RFC20 resource set specification version 1 support (#455)
- resource: add hwloc whitelist support (#467)
- resource: add set- and get-property support (#490, #513)
- resource: add support for checking a job's satisfiability (#503)
- resource: support for variation-aware scheduler (#517)
- resource: add JGF reader support (#521)
- resource: resource graph metadata (by_path) optimization (#536)
- libjobspec: update command to be list instead of list or string (#549)
- resource: add resource update support (#543)
- add smart pointer support and misc. cleanup (#537)
- test: add test cases to support systems with disaggregated resources (#460)
- test: add test cases for AMD GPUs (#464)
- resource: tidy up JGF match writer support (#520)
- sched/plugin: track flux-core module API changes (#414)
- Ensure scheduling correctness with different prune filter configurations (#419)
- travis: fix docker and github release deployment (#424)
- travis-ci: fix docker deploy better (#426)
- docker: update repo tag names, ensure flux-sched installed with prefix=/usr (#427)
- docker: add yaml-cpp dependency (#428)
- config: Fix a non-portable use of schell conditionals for automake (#405)
- config: Add libtoolize into autogen.sh (#406)
- Integrate jobspec into resource, etc (#398)
- Compilation fixes (#417)
- travis-ci: use docker for test builds (#392)
- resource: support for hwloc ingestion (#385)
- Add support for resource matching service (#386)
- planner: Replace zhash_t to zhashx_t for higher performance (#391)
- resource: wire in --prune-filters option for resource-query and matching module (#401)
- resource: Add run-level support for resource matching module (#418)
- Add better autoconf support for libboost (#394)
- build: do not install libflux-rdl.so, fix rebuild of aclocal.m4 on first make (#421)
- sched: remove trailing whitespaces (#376)
- sched: fix request handler type mismatch on i386 (#346)
- sched: Fix a job resrc_tree memory leak (#365)
- sched: Print error messages when schedule_job cannot run (#348)
- simulator: track a flux-core API change (#351)
- resource: Introduce planner_multi_t (#349)
- Add hwloc-based GPU scheduling support (#354)
- add sched parameter get & set commands to wreck (#362)
- rdl: Lua 5.2 compatibility (#373)
- travis: change email notifications to go to the flux-sched-dev listserv (#353)
- c++ configuration changes to mirror flux-core (#270)
- track upstream changes in flux-core APIs (#271, #272, #278)
- portability fixes for Ubuntu 16.04 and 17.10 (#298, #301)
- fixes for hwloc input xml (#310)
- add pending job cancellation support (#291)
- add support for excluding resources for specific nodes (#305)
- support for GPU scheduling (#313)
- add new, more scalable resource assignment schema in kvs (R_lite) (#321)
- include host names in R_lite resource assignment data (#324)
- optimize queue-depth=1 FCFS (#294)
- support for enhanced wreck submitted state event (#295)
- support for new completing state for wreck jobs (#341)
- add planner and resource-query utility support (#283, #323, #325, #274, #281)
- initial priority plugin development (#290)
- testsuite enhancements (#316, #328, #331)
- drop old rank.N method for resource assignment (#339)
- Fix allocation of cores unevenly across nodes (#226)
- Fixes and updates to track latest libflux-core API changes (#255, #253, #246, #238, #234)
- simulator: fix and re-enable sim integration tests (#250)
- Fix autoreconf when using side-installed autotools (#243)
- flux-waitjob: fixes to avoid flux-core issue #1027 (#244)
- Other fixes (#227, #231)
- simulator: add ISO time parsing to job traces (#260)
- resrc: add resource API context support (#236)
- simulator: refactor, cleanup, and ensure proper initialization. (#240, #250)
-
Support resource membership in multiple hierarchies and include "flow graphs" for modeling flow resources such as power and bandwidh. Enhance find/select of resources to support consideration of power, bandwidth, etc. (#189)
-
Add configurable scheduling optimizations for throughput-oriented workloads (#190, #191):
sched-params=queue-depth=<k>
: limits the number of jobs to consider per scheduling loopsched-params=delay-scheduling=true
: delays scheduling to batch up individual scheduling events
- Updates to track changes in flux-core (#188, #207, #212, #216)
-
Update JSON handling to libjansson (#215, #220)
-
Simulator updates and cleanup (#203, 218, #219)
-
RDL enhancements for multi-hierarchy support (#196)
-
Do not assume kvs path to jobid is in lwj.jobid (#216)
-
Versatile scheduling: replace resource tree list searches with tree searches.
-
Runtime selection of backfill algorithm (#159). Examples of selecting various algorithms are (first remove any loaded sched plugin, e.g., flux module remove sched.fcfs):
- Conservative :
flux module load sched.backfill reserve-depth=-1
- EASY :
flux module load sched.backfill reserve-depth=1
- Hybrid :
flux module load sched.backfill reserve-depth=4
- Pure :
flux module load sched.backfill reserve-depth=0
-
Update to latest flux-core logging apis
-
Fix segfaults in simulator during shutdown
-
Small leak fixes in simulator
- Remove
sim_sched
code that ran under simulator. The simulator now interfaces directly with sched module.
- Initial release for build testing only