From Kubernetes resource allocation to Linux task scheduling.
Gthulhu is a cloud-native runtime scheduling project built around Kubernetes, eBPF, and Linux sched_ext.
Our current direction is Claim2Core:
DRA chooses what and where; Gthulhu controls how it actually runs.
Kubernetes can allocate a workload to a Node, GPU, NIC, CPU set, NUMA domain, or other device. Gthulhu focuses on the execution gap that follows: mapping workload intent and actual allocation to the Linux tasks that need CPU service, then applying bounded runtime policy through sched_ext and verifying the result with eBPF telemetry.
Kueue / Workload API
│ admission / quota
▼
kube-scheduler / DRA
│ Node + device + topology allocation
▼
Gthulhu Runtime Plane
│ Claim → Pod/cgroup → TGID/TID
▼
sched_ext + eBPF
│ runtime policy + verification
▼
Delivered workload SLO
- pod-level scheduling observability with eBPF;
- Prometheus / Grafana / KEDA integration;
- distributed scheduling intents through a Manager and per-node Decision Makers;
- custom CPU scheduling on Linux 6.12+ with
sched_ext; - TID-aware node-policy matching for non-leader worker threads;
- user-space and kernel-mode priority handling with explicit non-boosting semantics.
The next roadmap steps focus on DRA semantic correctness, read-only ResourceClaim observation, Claim-to-Task provenance, and a static DRA-aware execution policy. See Gthulhu 2026 Roadmap — Claim2Core.
| Repository | Status | Role |
|---|---|---|
| Gthulhu/Gthulhu | Active | Main runtime/control-plane implementation |
| Gthulhu/qumun | Active | Go framework for custom sched_ext schedulers |
| Gthulhu/plugin | Active | User-space scheduling strategy implementation |
| Gthulhu/docs | Active | Official documentation at gthulhu.org |
| Gthulhu/gtp5g-operator | Active | Telecom / GTP5G integration work |
| Gthulhu/kina | Active | Related runtime experimentation |
| Gthulhu/libbpfgo | Active | Project-maintained libbpfgo fork |
| Gthulhu/api | Archived | Historical standalone API repository; code has moved into the main repository |
| Gthulhu/chart | Archived | Historical standalone Helm chart repository; deployment assets now live with the main project |
| Gthulhu/mcp | Archived | Experimental MCP work |
- Claim2Core —
ResourceClaim → Pod/cgroup → TGID/TID → sched_extexecution lineage. - Device-local execution domains — respect allocated CPU/cgroup boundaries while preferring NUMA / PCIe locality.
- Scheduling provenance — preview, explain, intended-vs-actual runtime state, and stale-state verification.
- Workload adapters — free5GC/UPF first for fast end-to-end validation; LLM prefill/decode/NCCL roles for accelerator-focused research.
- Closed-loop runtime control — only after static policy and provenance are trustworthy.
- Main repository: Gthulhu/Gthulhu
- Roadmap: Issue #141
- Documentation: gthulhu.org
- Contributing: Contribution guide
- Discussions: GitHub Discussions
Gthulhu is Apache-2.0 licensed and welcomes contributions across Kubernetes, DRA, eBPF, sched_ext, NUMA/topology, observability, telecom, and accelerator workloads.
