Melvin Strobl
1dc891aa1b
- NodeDef `device`/`device_policy` become `worker`/`worker_policy`, the issue `ignored_device` becomes `ignored_worker`, and the SDK's `@node` takes `worker=` and `worker_policy=` - FlowStore._migrate renames the keys in published and draft documents once, bumping their version; a queued run's `needs` reads the old key - placement passes the label as `label`, client regenerated - docs: worker placement wording, MCP issuer without DOMAIN, device rows in the CLI data directory and the API table, device links in compose comments
fluksio-worker
Runs Fluksio nodes on a machine the engine cannot reach — a GPU box, a cluster node, anything behind a NAT. It dials out to the engine over one authenticated websocket, so nothing has to be exposed here.
pip install fluksio-worker
fluksio-worker \
--url wss://api.example.com/api/v1/workers/attach \
--labels gpu,cuda12 \
--python /opt/torch-venv/bin/python
--python is the interpreter node code runs on, which is how this machine keeps
its own wheels without the engine ever installing them. Mint a token from the
engine with POST /api/v1/workers/tokens; it carries this worker's name and
labels, and --labels may repeat those but not change them.
It sends the token as an Authorization header, which an engine from before
that does not read: upgrade the engine first, then its workers.
A plain ws:// address is refused unless the engine is on this machine, since
the token and node secrets would cross the network in clear; --insecure
allows it on a network you trust.
Linux and macOS.
License
Copyright (C) 2026 Melvin Strobl — GNU Affero General Public License v3.0 or later. Running a modified version over a network obliges you to offer its users the corresponding source (AGPL §13).