hero_cloud_explorer — multi-node aggregator and RPC proxy #3

New issue

Closed

opened 2026-03-05 08:16:18 +00:00 by mahmoud · 1 comment

mahmoud commented

2026-03-05 08:16:18 +00:00

Owner

Description

A new standalone service that sits in front of multiple hero_cloud nodes. Nodes register themselves with the Explorer every ~5 minutes. Users connect to the Explorer and can deploy VMs across any registered node without knowing the node's address directly.

How It Works

Each hero_cloud node periodically pings the Explorer with:

Node specs (RAM, disk, slice count, available slices)
Node status (online/offline/degraded)
Mycelium address (used for routing RPC calls)

The Explorer:

Aggregates all registered nodes into a directory
Exposes the same RPC interface as hero_cloud_server
Proxies VM operations to the correct backend node
Adds extra methods: list_nodes, node_heartbeat

Deployment Model

Deploy 2-3 Explorers under DNS names:

explorer.cloud1.hero.tf
explorer.cloud2.hero.tf
explorer.cloud3.hero.tf

Nodes connect to all Explorers via Mycelium.
Users connect to any Explorer — they all have the same view.

New Crate

crates/hero_cloud_explorer

Same Hero RPC / OSIS framework as hero_cloud_server
New schema: explorer.oschema
Methods: node_heartbeat, list_nodes, get_node,
and all 14 hero_cloud methods proxied to backends

Notes

Nodes report via Mycelium — no public IP needed on nodes
Explorer needs a public IP or DNS
Secret auth (Issue #2) must be done first — the Explorer passes secrets through to backend nodes unchanged
Start with one Explorer instance, scale to 3 later

### Description A new standalone service that sits in front of multiple hero_cloud nodes. Nodes register themselves with the Explorer every ~5 minutes. Users connect to the Explorer and can deploy VMs across any registered node without knowing the node's address directly. ### How It Works Each hero_cloud node periodically pings the Explorer with: - Node specs (RAM, disk, slice count, available slices) - Node status (online/offline/degraded) - Mycelium address (used for routing RPC calls) The Explorer: - Aggregates all registered nodes into a directory - Exposes the same RPC interface as hero_cloud_server - Proxies VM operations to the correct backend node - Adds extra methods: `list_nodes`, `node_heartbeat` ### Deployment Model Deploy 2-3 Explorers under DNS names: - `explorer.cloud1.hero.tf` - `explorer.cloud2.hero.tf` - `explorer.cloud3.hero.tf` Nodes connect to all Explorers via Mycelium. Users connect to any Explorer — they all have the same view. ### New Crate `crates/hero_cloud_explorer` - Same Hero RPC / OSIS framework as hero_cloud_server - New schema: `explorer.oschema` - Methods: `node_heartbeat`, `list_nodes`, `get_node`, and all 14 hero_cloud methods proxied to backends ### Notes - Nodes report via Mycelium — no public IP needed on nodes - Explorer needs a public IP or DNS - Secret auth (Issue #2) must be done first — the Explorer passes secrets through to backend nodes unchanged - Start with one Explorer instance, scale to 3 later