ROCm/spur

View on GitHub

spur your compute

Stars: 120
Forks: 48
Open beginner issues: 2
Indexed issues: 86
Avg merge: 3d 14h
Merged PRs (30d): 89

Dominant language: Rust
License: Apache-2.0
Last GitHub push: Sep 19, 2026
Latest indexed: Sep 20, 2026
Contributing guide: No contributing guide
Code of conduct: No code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

81 open issues indexed

spurd inventory convergence: controller does not drain/fail-requeue a job whose allocated GPU vanished Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

ROCm/spur#892 · Opened Sep 18, 2026
spurd re-registration under non-default auth: label revert, token expiry, unbounded retry Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

ROCm/spur#891 · Opened Sep 17, 2026
Preemption: an unschedulable job preempts one running job per cycle, with no memory and no feasibility gate Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

ROCm/spur#888 · 1 comment · Opened Sep 17, 2026
`srun --jobid --overlap` attaches via PTY and ignores `--mpi=pmix` Open

area: cli area: networking priority: high slurm-compat type: bug

ROCm/spur#876 · 1 assignee · Opened Sep 12, 2026
Nested overlapping job steps share `~/.spur_step_$JOBID` and clobber PMIx wrappers Open

area: job-lifecycle area: networking priority: high slurm-compat type: bug

ROCm/spur#874 · 1 assignee · Opened Sep 12, 2026
spur-k8s: launch_job reports success when the Pod already exists Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ROCm/spur#873 · Opened Sep 11, 2026
Concurrent srun --overlap container steps fail with "unsquashfs ... File exists" Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#865 · Opened Sep 10, 2026
OpenMPI mpirun fails on Spur due to unsupported srun --kill-on-bad-exit argument Open

Difficulty 4/5 3-5 days Newbie friendliness 42/100

ROCm/spur#862 · Opened Sep 9, 2026
spur k8s up on OCI: pod traffic between nodes is dropped with the default CNI Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#852 · Opened Sep 8, 2026
[feat]: sbcast is missing: no way to stage a file onto a running job's allocated nodes Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#828 · Opened Sep 5, 2026
[Issue]: --mem unit suffixes are inconsistent across sbatch/srun/salloc (K only in sbatch, T nowhere) Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ROCm/spur#827 · Opened Sep 5, 2026
[Issue]: A zero token TTL (--ttl 0d) creates a never-expiring token instead of being rejected Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

ROCm/spur#826 · Opened Sep 5, 2026
[Issue]: build_node_allocation can leave an empty "gpu" device list, so is_empty() and has_devices() disagree Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ROCm/spur#825 · Opened Sep 5, 2026
[Issue]: scontrol show hostnames/hostlist/hostnamestr is missing Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ROCm/spur#824 · Opened Sep 5, 2026
[Issue]: Job array with overlapping ranges that dedup within the cap is wrongly rejected Open

Difficulty 3/5 1-2 days Newbie friendliness 70/100

ROCm/spur#823 · Opened Sep 5, 2026
[Issue]: Over-cap or malformed --nodelist/--exclude is accepted at submission, then the job never runs Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

ROCm/spur#822 · Opened Sep 5, 2026
GPU time is never charged to fair-share: gpu_seconds is never written, fair-share weights CPU only, and QOS usage_factor is inert Open

Difficulty 5/5 Over a week Newbie friendliness 38/100

ROCm/spur#805 · Opened Sep 2, 2026
A spurd restart orphans running batch jobs: the agent never re-adopts them and the controller never reaps them Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#803 · 1 comment · Opened Sep 2, 2026
No node health verification between job release and next assignment Open

Difficulty 5/5 Over a week Newbie friendliness 42/100

ROCm/spur#801 · Opened Sep 2, 2026
Node resource inventory is captured once at spurd startup and never refreshed; heartbeats carry no resources Open

Difficulty 4/5 3-5 days Newbie friendliness 65/100

ROCm/spur#800 · Opened Sep 2, 2026
SSH sessions survive job cancellation and retain GPU access on the released node Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

ROCm/spur#799 · Opened Sep 2, 2026
salloc cancels valid pending allocations after five minutes Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

ROCm/spur#789 · Opened Sep 1, 2026
Compute-node SSH sessions are neither adopted into jobs nor tied to allocations (no `pam_slurm_adopt` equivalent) Open

Difficulty 5/5 Over a week Newbie friendliness 38/100

ROCm/spur#782 · 1 comment · Opened Aug 31, 2026
AMD CDI injection silently overlays host ROCm libraries onto every GPU container Open

ROCm/spur#779 · 1 assignee · Opened Aug 31, 2026
`--container-remap-root` is accepted and silently ignored (dead flag) Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#778 · Opened Aug 31, 2026
spurctld: partition DefaultTime=0 gives an untimed job a zero-length time limit instead of refusing it Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#759 · Opened Aug 27, 2026
E2E: intermittent failures from shared runner state (fixed Postgres port, 60s wait deadline) Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

ROCm/spur#747 · Opened Aug 25, 2026
k0s adopters entry Open

Difficulty 2/5 1-3 hours Newbie friendliness 68/100

ROCm/spur#713 · 1 reaction · Opened Aug 21, 2026
fix(spurctld)!: controller/agent JWT default-key asymmetry breaks dispatch on a config with no [auth] jwt_key Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

ROCm/spur#710 · Opened Aug 21, 2026
scontrol update job bypasses submit-time limit and account-membership validation Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#694 · Opened Aug 19, 2026
GrpSubmitJobs limit is read per-user-row but counted account-wide Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#691 · Opened Aug 19, 2026
Submit-count limits can over-admit under concurrent submissions (TOCTOU) Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#690 · Opened Aug 19, 2026
spurd --hostname: jobs never complete and their allocations are never released Open

Difficulty 3/5 1-2 days Newbie friendliness 78/100

ROCm/spur#684 · Opened Aug 18, 2026
epic(tenancy): multi-tenancy isolation gaps beyond authorization Open

area: scheduling enhancement priority: high

Difficulty 5/5 Over a week Newbie friendliness 25/100

ROCm/spur#665 · 1 reaction · Opened Aug 17, 2026
tenancy: no GPU state scrub (VRAM/KFD) between jobs on a reassigned device Open

area: resources effort: large enhancement priority: high

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#661 · 1 comment · Opened Aug 17, 2026
tenancy: no per-tenant network isolation (flat WireGuard mesh + no netns on bare/srun paths) Open

area: networking effort: large enhancement priority: high

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#659 · 1 comment · Opened Aug 17, 2026
tenancy: replace the single symmetric jwt_key with per-principal asymmetric keys Open

area: controller effort: large enhancement priority: high

Difficulty 5/5 Over a week Newbie friendliness 25/100

ROCm/spur#658 · 1 comment · Opened Aug 17, 2026
tenancy: scheduler has no per-node tenant exclusivity (exclusive_user is dead; placement has no tenant field) Open

area: scheduling effort: large enhancement priority: high

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#657 · 1 comment · Opened Aug 17, 2026
spur-k8s: failure-backoff map can leak entries when a SpurJob bypasses the controller's cleanup path Open

ROCm/spur#632 · 1 assignee · Opened Aug 14, 2026
[RFC]: Measure Slurm compatibility with the Slurm test suite Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

ROCm/spur#621 · 1 comment · Opened Aug 12, 2026
Prolog/epilog hooks only accept a single script, not a directory of scripts Open

area: job-lifecycle priority: medium type: enhancement

Difficulty 3/5 1-2 days Newbie friendliness 76/100

ROCm/spur#615 · Opened Aug 11, 2026
Untimed multi-partition jobs get one min-derived wall-time instead of per-partition resolution Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#604 · Opened Aug 10, 2026
fix(spur-cli): reject unresolved k8s callers Open

Difficulty 2/5 1-3 hours Newbie friendliness 78/100

ROCm/spur#603 · Opened Aug 10, 2026
spur exec / srun --overlap lose supplementary groups (video, render) — GPU access broken on nsenter paths Open

priority: high

Difficulty 4/5 3-5 days Newbie friendliness 68/100

ROCm/spur#596 · 1 comment · Opened Aug 7, 2026
Heartbeat reclaim cancel for stale interactive allocations isn't epoch-gated against a concurrent requeue Open

area: fault-tolerance priority: low type: enhancement

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#592 · Opened Aug 7, 2026
Live-reloading admission.mode via scontrol reconfigure can strand registered agents and sign node tokens with the default key Open

Difficulty 5/5 Over a week Newbie friendliness 38/100

ROCm/spur#573 · Opened Aug 5, 2026
spur-cli: sbatch missing Slurm --priority and --nice submission flags Open

ROCm/spur#572 · 1 assignee · Opened Aug 5, 2026
Generalize internal k0s naming behind a distro abstraction when a second platform lands Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

ROCm/spur#559 · Opened Aug 2, 2026
sbatch batch script runs once per node, not once per job (Slurm parity gap) Open

area: job-lifecycle effort: large priority: high slurm-compat type: enhancement

ROCm/spur#553 · 1 assignee · Opened Jul 31, 2026
sacctmgr association/user changes are not reflected until the association cache refreshes Open

area: accounting effort: medium priority: medium type: bug

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ROCm/spur#543 · Opened Jul 30, 2026

ROCm/spur

Get new issues in your inbox