Remove the Hub Self-Node Concept (YAGNI Cleanup)

Date: 2026-08-11 Status: Design approved, pending implementation plan Scope: Delete the hub “self-node” concept introduced in the admin-hardening work (PR #202). It is inert in production, read by no business logic, and not required for alert management.

Why remove it

The self-node was meant to materialize the hub as a first-class Node so the node axis would be “uniform.” In practice:

• It never works in production. Node.ensure_self() no-ops when settings.INSTANCE_ID is empty, and INSTANCE_ID is an agent-side setting the hub does not set. So the hub self-node is never created.
• Nothing consumes it. is_self is read by no business logic — only shown in the Node admin list/filter. ensure_self() has one functional caller: orchestrator._resolve_node stamps PipelineRun.node for the hub’s checker_generated runs.
• Alert management does not need it. Incident/alert grouping already works via incident_instance_key (instance_id → instance → hostname). Agent alerts group by instance_id (a registered Node); hub self-check alerts group by hostname. Both are visible on the alerts page and in the dashboard incident totals today — confirmed as the desired behavior.

Making it work would require inventing a stable hub identity (a new setting or a hostname fallback) to serve a convenience nobody consumes. Not worth it.

Considered and rejected: a per-source dashboard view

We considered building a dashboard panel that groups incidents/alerts by source node (which would give the hub a real reason to have an identity). Decision: not now. Hub alerts already appear in the dashboard incident totals and on the alerts page; a per-source breakdown is a separate, larger feature that is not currently wanted. This design is a pure deletion.

Removed

• Node.is_self field (apps/alerts/models.py) + a RemoveField migration.
• Node.ensure_self() classmethod (and the now-unused socket import if it becomes unused).
• bootstrap_self_node management command (apps/alerts/management/commands/bootstrap_self_node.py) and its test.
• is_self from NodeAdmin.list_display / list_filter (apps/alerts/admin.py).
• The is_self / ensure_self tests in apps/alerts/_tests/test_models.py and test_admin.py.
• The if origin == CHECKER_GENERATED: return Node.ensure_self() branch in orchestrator._resolve_node — all origins now flow through the single label-based resolution path.
• Self-node wording in apps/alerts/AGENTS.md and the two help_text strings that mention “hub self-node”.

Kept (deliberately — broader than the self-node)

• PipelineRun.node FK and the PipelineOrigin enum. origin (incoming_webhook / checker_generated / manual) still usefully classifies runs; checker_generated stays a valid label, just no longer node-mapped.
• The dashboard nodes-readiness card (uses Node.count() / last_seen, never is_self).
• docs/plans/* design docs (historical record) — left unchanged, including the admin-hardening design that introduced the self-node.

Behavior change

The hub’s own --checks-only pipelines, which would have received node=ensure_self(), now get node=null — via the single label-resolution path (no instance_id in a self-check payload). In production this is a no-op: ensure_self() already returns None there, so those runs are already node=null. Their alerts continue to group by hostname exactly as today.

Testing

• Delete the obsolete tests (is_self, ensure_self, bootstrap_self_node).
• Flip test_checker_generated_run_gets_self_node → assert run.node is None (rename accordingly).
• Full suite green; 100% branch coverage on changed lines.
• makemigrations --check --dry-run clean after the RemoveField migration.
• bandit / black / ruff clean.

Acceptance criteria

• No references to is_self, ensure_self, or bootstrap_self_node remain in apps/, config/, or bin/ (design docs excepted).
• Node has no is_self column (migration applied); _resolve_node has a single resolution path.
• Full CI green; coverage 100% on changed lines; no behavior change to deployed alert grouping.