config-atlas/workplans/ATLAS-WP-0004-explain-and-graph.md

166 lines
6.6 KiB
Markdown
Raw Normal View History

---
id: ATLAS-WP-0004
type: workplan
title: "Effective-config explain and graph"
domain: infotech
repo: config-atlas
status: finished
owner: codex
topic_slug: custodian
created: "2026-06-26"
updated: "2026-06-27"
state_hub_workstream_id: "fbfdbf2b-ca6b-450e-a654-a61c5939f068"
---
# Effective-config explain and graph
Turn the registry into the **read-first control-plane MVP**: given a key, show its
layer path, what overrides what, its owner, its consumers, and its blast radius —
**statically, from source links, without reading any live value**.
This is **Phase 3** of [`specs/ArchitectureBlueprint.md`](../specs/ArchitectureBlueprint.md)
§6 and realizes PRD FR-5 and FR-7
([`specs/ProductRequirementsDocument.md`](../specs/ProductRequirementsDocument.md)).
The configuration knowledge graph reuses the **State Hub** relationship store rather
than building a new graph database ([`docs/ecosystem-boundaries.md`](../docs/ecosystem-boundaries.md)
§2.5; [`research/configuration-control-plane.md`](../research/configuration-control-plane.md) §5).
**Depends on:** ATLAS-WP-0002 (entries carry `sources[].role` and `relations`).
Benefits from, but does not require, ATLAS-WP-0003 (connectors enrich coverage).
**Hard boundary:** this renders the effective-config *path* only. Resolving the
actual effective *value* is out of scope (delegated downstream); `feature-flag`
surfaces link to feature-control and are never re-resolved here.
**Exit condition:** `config explain <surface-id>` emits an ordered override path
with winning layer, what it overrode, validating schema, and owner; config-typed
edges are expressible to the State Hub graph; a blast-radius view lists affected
consumers.
**Sequencing:** T01 (resolver) → T02 (CLI) and T05 (tests); T03 (graph edges) →
T04 (blast-radius). T01 and T03 may start in parallel.
## Effective-config path resolver (static)
```task
id: ATLAS-WP-0004-T01
status: done
priority: high
state_hub_task_id: "cee293aa-b407-4b97-a462-b67d7aa0f170"
```
Result 2026-06-27: Added `tools/effective_config.py` — a pure, deterministic
resolver that orders a surface's `sources[]` by canonical L0-L9 layer rank and
emits the override path (contributing layers, winning layer, what each overrode,
validator, owner). Renders the PATH only, never a value; non-layer roles
(feature-control-key) are listed as linked authority. Verified order-independent
(8x shuffle -> identical path).
Implement a static resolver that, from a surface's `sources[]` (with layer `role`),
the L0L9 ordering, and the explicit merge rules, produces the **override path**:
ordered contributing layers, the winning layer, what each overrode, the validating
schema, and the owner — never a live value. Honors non-overridable guardrails. See
the `config explain` shape in [`wiki/ConfigLayering.md`](../wiki/ConfigLayering.md).
- **Acceptance:** given a surface with ordered sources, the resolver returns a
deterministic, order-independent override path with owner + validator refs and no
resolved value.
## `config explain` tool
```task
id: ATLAS-WP-0004-T02
status: done
priority: high
state_hub_task_id: "5d66a8c1-74bd-4c6f-9ea6-839e884ad103"
```
Result 2026-06-27: Added `tools/config_explain.py` and `make explain
SURFACE=...`. Renders the resolver output (layer path, overrides, winning layer,
validator, owner, consumers, secret refs) for any `surface.*` id. Verified on all
4 seeded surfaces; no values or secret values appear in output. Documented in
`.claude/rules/stack-and-commands.md`.
Add a `tools/` command (e.g. `config_explain.py`) that takes a `surface.*` id and
renders the resolver output as the human/agent-readable `config explain` view.
Reuse the existing `tools/` + Makefile pattern (`make explain SURFACE=...`).
- **Acceptance:** `config explain surface.infotech.state-hub.api-config` prints the
ordered layer path, overrides, validator, owner, and consumers.
## Config knowledge graph edges to State Hub
```task
id: ATLAS-WP-0004-T03
status: done
priority: medium
state_hub_task_id: "7b16eaa0-f5e1-4ff3-809d-729b312fd154"
```
Result 2026-06-27: Added `tools/config_graph.py` + `make graph`/`graph-query`.
Emits config-typed edges (consumed_by/overrides/depends_on_secret/related_to) to
`registry/indexes/graph.yaml` (10 nodes, 10 edges) — queryable by surface id in
both directions. config-atlas owns edge semantics; the State Hub owns storage
(artifact is hub-ingestible). A `--check` staleness gate runs in `make validate`.
Emit config-typed edges (`consumed_by`, `overrides`, `depends_on_secret`,
`related_to`) from surface entries to the **State Hub** relationship/graph model,
contributing the config semantics of each edge while the hub stores topology
(PRD FR-7; ecosystem-boundaries §2.5). Do not build a separate graph store.
- **Acceptance:** declared edges from the seed surfaces are expressed to the State
Hub without duplicating its store; edges are queryable by surface id.
## Blast-radius / dependency view
```task
id: ATLAS-WP-0004-T04
status: done
priority: medium
state_hub_task_id: "57e085c7-25e8-4e2d-bb3e-43b82e351aa9"
```
Result 2026-06-27: Added `tools/blast_radius.py` + `make blast-radius`. For a
surface it lists direct consumers, transitively-dependent surfaces (cycle-safe),
referenced secrets, owner, and a fan-out risk band. Verified: state-hub config is
high risk (fan-out 4); leaf surfaces lower.
Build a view that, for a given surface, traverses the graph to list affected
consumers, dependent surfaces, and referenced secrets — the
`config key → service → tenant → feature → secret → owner` chain from
research §5. Supports change-risk reasoning before a change.
- **Acceptance:** for a surface, the view lists downstream consumers and dependent
surfaces; a change to a high-fan-out surface is visibly higher risk.
## Resolver determinism tests
```task
id: ATLAS-WP-0004-T05
status: done
priority: medium
state_hub_task_id: "77b19d5f-f55c-48bb-8129-ba1478e47223"
```
Result 2026-06-27: Added `tests/test_effective_config.py` (6 tests) proving the
override path is order-independent (16x shuffle), the winner is the most-specific
layer, and no schema default / live / secret value leaks into explain output.
Wired into `make validate` and CI (`validate-tests`).
Add tests proving the override path is deterministic and order-independent
(consistent with the merge-rule and CUE-unification rationale,
research §3.3), and that no live or secret value ever appears in explain output.
Wire into the CI gate (`make validate`).
- **Acceptance:** tests pass in CI; an explain output containing a literal value or
secret fails the test.
---
After workplan or registry file updates, sync from `~/state-hub`:
```bash
make fix-consistency REPO=config-atlas
```
</content>