ADR-047: Terminal operations console shares daemon events and controls¶

Status¶

Accepted.

Context¶

The web dashboard already reads identifier-only change notices from GET /api/events and fetches authoritative state from REST. Operators also need a full-screen terminal view of mining jobs, readiness and maintenance without moving business logic into a CLI process. Unlike cooperative cancellation, a force-cancel may stop a local worker before it checkpoints; an interrupted request must remain durable across a daemon restart, and a different daemon's lease must never be overwritten.

Decision¶

memcastle tui uses ratatui and crossterm for rendering and input, and client::DaemonClient for every daemon call. Its Rust SSE reader uses the web dashboard's existing authenticated endpoint and wire format, never a parallel bus. Events prompt rereads; reconnect, resync, a manual refresh and periodic reconciliation cover gaps in that process-local stream.

The REST-only POST /api/jobs/{id}/force-cancel applies to a locally owned running mining job. The scheduler first persists cancel_requested, then aborts and joins its task. It applies JobEvent::Cancel through Job::apply and checks running status and lease owner atomically before answering. A crash between those steps is recovered as a cancellation, and subsequent fenced writes from the old run are refused. Blocking work outside the task may continue briefly and previously filed data remains filed; force-cancel does not claim process isolation or instant termination. The ordinary pause and cancel routes remain cooperative, preserving checkpoint-friendly behavior.

Consequences¶

The terminal and browser share one domain and event model, while each keeps its own presentation and stream reader. No new storage, credential handling, MCP control or event replay is needed. An owner on a different daemon cannot be force-cancelled through this daemon; the caller receives a rejection and can ask for graceful cancellation instead.