Eru

A stateless, engine-agnostic resource scheduler. Eru core allocates CPU, memory, GPU and storage across pods of nodes and deploys workloads on Docker containers or virtual machines through a single gRPC API; agents on each node report health and workload status back.

eru cli / SDK ──► core (gRPC) ──► resource plugins: cpumem (built in) | gpu | storage
                      │
                      ├─ store: etcd | redis ── nodes, pods, workloads, deploy status
                      ├─ engine: docker | yavirt (VM, archived runtime) | systemd
                      └─ node ◄── agent ── health checks, logs, metrics, status heartbeat

Projects

Data plane

• mithril — fast thread-per-core Redis Cluster proxy in Rust: RESP2/RESP3, transparent topology absorption, per-slot multi-key fan-out, single-virtual-node cluster emulation.

Core

• core — the scheduler: gRPC API, resource allocation strategies, Docker / VM engines, etcd or redis metadata, pluggable resource plugins.
• agent — the per-node agent: workload health checks, log forwarding, node status heartbeat and Prometheus metrics for Docker and yavirt runtimes.
• cli — the command-line client: pods, nodes, workloads, images, networks and lambdas over the core gRPC API.
• resource-extend — out-of-tree resource plugins: GPU and storage/volume scheduling as external plugin binaries.

Tooling

• quickstart — start a standalone Eru and run something on it.
• footstone — container images for building and testing the Eru system.