Files
git.stella-ops.org/docs/modules/sbomservice/architecture.md

122 lines
11 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# SBOM Service architecture (2025Q4)
> Scope: canonical SBOM projections, lookup and timeline APIs, asset metadata overlays, and events feeding Advisory AI, Console, Graph, Policy, and Vuln Explorer.
## 1) Mission & boundaries
- Mission: serve deterministic, tenant-scoped SBOM projections (Link-Not-Merge v1) and related metadata for downstream reasoning and overlays.
- Boundaries:
- Does not perform scanning; consumes Scanner outputs or supplied SPDX/CycloneDX blobs.
- Does not author verdicts/policy; supplies evidence and projections to Policy/Concelier/Graph.
- Append-only SBOM versions; mutations happen via new versions, never in-place edits.
- Owns the SBOM lineage ledger for versioned uploads, diffs, and retention pruning.
## 2) Project layout
- `src/SbomService/StellaOps.SbomService` — REST API + event emitters + orchestrator integration.
- Storage: PostgreSQL tables (proposed)
- `sbom_snapshots` (immutable versions; tenant + artifact + digest + createdAt)
- `sbom_projections` (materialised views keyed by snapshotId, entrypoint/service node flags)
- `sbom_assets` (asset metadata, criticality/owner/env/exposure; append-only history)
- `sbom_catalog` (console catalog surface; indexed by artifact, scope, license, assetTags.*, createdAt for deterministic pagination)
- `sbom_component_neighbors` (component lookup graph edges; indexed by purl+artifact for cursor pagination)
- `sbom_paths` (resolved dependency paths with runtime flags, blast-radius hints)
- `sbom_events` (outbox for event delivery + watermark/backfill tracking)
### 2.1) SBOM + provenance spine (Nov 2026)
The service now owns an idempotent spine that converts OCI images into SBOMs and provenance bundles with DSSE and in-toto. The flow is intentionally air-gap ready:
- **Extract** OCI manifest/layers (hash becomes `contentAddress`).
- **Build SBOM** in CycloneDX 1.7 and/or SPDX 3.0.1; canonicalize JSON before hashing (`sbomHash`).
- **Sign** outputs as DSSE envelopes; predicate uses in-toto Statement with SLSA Provenance v1.
- **Publish** attestations optionally to a transparency backend: `rekor`, `local-merkle`, or `null` (no-op). Local Merkle log keeps proofs for later sync when online.
Minimal APIs exposed by SbomService (idempotent by hash):
- `POST /sbom/ingest` `{ imageDigest, sbom, format, dsseSignature? }``{ sbomId, status: stored|already_present, sbomHash }` keyed by `contentAddress + sbomHash`.
- `POST /attest/verify` `{ dsseEnvelope, expectedSubjects[] }``{ verified, predicateType, logIndex?, inclusionProof? }` and records attestation when verified.
Operational rules:
- Default media types: `application/vnd.cyclonedx+json`, `application/spdx+json`, `application/dsse+json`, `application/vnd.in-toto+json`.
- If the same SBOM/attestation arrives again, return HTTP 200 with `"status":"already_present"` and do not create a new version.
- Offline posture: no external calls required; Rekor publish remains optional and retryable when connectivity is restored.
## 3) APIs (first wave)
- `GET /sbom/paths?purl=...&artifact=...&scope=...&env=...` — returns ordered paths with runtime_flag/blast_radius and nearest-safe-version hint; supports `cursor` pagination.
- `GET /sbom/versions?artifact=...` time-ordered SBOM version timeline for Advisory AI; include provenance and source bundle hash.
- `POST /sbom/upload` BYOS upload endpoint; validates/normalizes SPDX 2.3/3.0 or CycloneDX 1.41.6 and registers a ledger version.
- `GET /sbom/ledger/history` list version history for an artifact (cursor pagination).
- `GET /sbom/ledger/point` resolve the SBOM version at a specific timestamp.
- `GET /sbom/ledger/range` query versions within a time range.
- `GET /sbom/ledger/diff` component/version/license diff between two versions.
- `GET /sbom/ledger/lineage` parent/child lineage edges for an artifact chain.
- `GET /console/sboms` Console catalog with filters (artifact, license, scope, asset tags), cursor pagination, evaluation metadata, immutable JSON projection for drawer views.
- `GET /components/lookup?purl=...` component neighborhood for global search/Graph overlays; returns caches hints + tenant enforcement.
- `POST /entrypoints` / `GET /entrypoints` manage entrypoint/service node overrides feeding Cartographer relevance; deterministic defaults when unset.
- `GET /sboms/{snapshotId}/projection` Link-Not-Merge v1 projection returning hashes plus asset metadata (criticality, owner, environment, exposure flags, tags) alongside package/component graph.
- `GET /internal/sbom/events` — internal diagnostics endpoint returning the in-memory event outbox for validation.
- `POST /internal/sbom/events/backfill` — replays existing projections into the event stream; deterministic ordering, clock abstraction for tests.
- `GET /internal/sbom/asset-events` — diagnostics endpoint returning emitted `sbom.asset.updated` envelopes for validation and air-gap parity checks.
- `GET/POST /internal/orchestrator/sources` — list/register orchestrator ingest/index sources (deterministic seeds; idempotent on artifactDigest+sourceType).
- `GET/POST /internal/orchestrator/control` — manage pause/throttle/backpressure signals per tenant; metrics emitted for control updates.
- `GET/POST /internal/orchestrator/watermarks` — fetch/set backfill watermarks for reconciliation and deterministic replays.
- `GET /internal/sbom/resolver-feed` list resolver candidates (artifact, purl, version, paths, scope, runtime_flag, nearest_safe_version).
- `POST /internal/sbom/resolver-feed/backfill` clear and repopulate resolver feed from current projections.
- `GET /internal/sbom/resolver-feed/export` NDJSON export of resolver candidates for air-gap delivery.
- `GET /internal/sbom/ledger/audit` audit trail for ledger changes (created/pruned).
- `GET /internal/sbom/analysis/jobs` list analysis jobs triggered by BYOS uploads.
- `POST /internal/sbom/retention/prune` apply retention policy and emit audit entries.
## 3.1) Ledger + BYOS workflow (Sprint 4600)
- Uploads are validated, normalized, and stored as ledger versions chained per artifact identity.
- Diffs compare normalized component keys and surface version/license deltas with deterministic ordering.
- Lineage is derived from parent version references and emitted for Graph lineage edges.
- Lineage relationships include parent links plus build links (shared CI build IDs when provided).
- Retention policy prunes old versions while preserving audit entries and minimum keep counts.
- See `docs/modules/sbomservice/ledger-lineage.md` for request/response examples.
- See `docs/modules/sbomservice/byos-ingestion.md` for supported formats and troubleshooting.
## 4) Ingestion & orchestrator integration
- Ingest sources: Scanner pipeline (preferred) or uploaded SPDX 2.3/3.0 and CycloneDX 1.41.6 bundles.
- Orchestrator: register SBOM ingest/index jobs; worker SDK emits artifact hash + job metadata; honor pause/throttle; report backpressure metrics; support watermark-based backfill for idempotent replays.
- Idempotency: combine `(tenant, artifactDigest, sbomVersion)` as primary key; duplicate ingests short-circuit.
## 5) Events & streaming
- `sbom.version.created` — emitted per new SBOM snapshot; payload: tenant, artifact digest, sbomVersion, projection hash, source bundle hash, import provenance; replay/backfill via outbox with watermark.
- `sbom.asset.updated` — emitted when asset metadata changes; idempotent payload keyed by `(tenant, assetId, version)`.
- Inventory/resolver feeds — queue/topic delivering `(artifact, purl, version, paths, runtime_flag, scope, nearest_safe_version)` for Vuln Explorer/Findings Ledger.
- Current implementation uses an in-memory event store/publisher (with clock abstraction) plus `/internal/sbom/events` + `/internal/sbom/events/backfill` to validate envelopes until the PostgreSQL-backed outbox is wired.
- Entrypoint/service node overrides are exposed via `/entrypoints` (tenant-scoped) and should be mirrored into Cartographer relevance jobs when the outbox lands.
## 6) Determinism & offline posture
- Stable ordering for projections and paths; timestamps in UTC ISO-8601; hash inputs canonicalised.
- Add-only evolution for schemas; LNM v1 fixtures published alongside API docs and replayable tests.
- Offline-friendly: uses mirrored packages, avoids external calls during projection; exports NDJSON bundles for air-gapped replay.
## 7) Tenancy & security
- All APIs require tenant context (token claims or mTLS binding); collection filters must include tenant keys.
- Enforce least-privilege queries; avoid cross-tenant caches; log tenant IDs in structured logs.
- Input validation: schema-validate incoming SBOMs; reject oversized/unsupported media types early.
## 8) Observability
- Metrics: `sbom_projection_seconds`, `sbom_projection_size_bytes`, `sbom_projection_queries_total`, `sbom_paths_latency_seconds`, `sbom_paths_cache_hit_ratio`, `sbom_events_backlog`, `sbom_ledger_uploads_total`, `sbom_ledger_diffs_total`, `sbom_ledger_retention_pruned_total`.
- Tracing: ActivitySource `StellaOps.SbomService` (entrypoints, component lookup, console catalog, projections, events).
- Traces: wrap ingest, projection build, and API handlers; propagate orchestrator job IDs.
- Logs: structured, include tenant + artifact digest + sbomVersion; classify ingest failures (schema, storage, orchestrator, validation).
- Alerts: backlog thresholds for outbox/event delivery; high latency on path/timeline endpoints.
## 9) Configuration (PostgreSQL-backed catalog & lookup)
- Enable PostgreSQL storage for `/console/sboms` and `/components/lookup` by setting `SbomService:PostgreSQL:ConnectionString` (env: `SBOM_SbomService__PostgreSQL__ConnectionString`).
- Optional overrides: `SbomService:PostgreSQL:Schema`, `SbomService:PostgreSQL:CatalogTable`, `SbomService:PostgreSQL:ComponentLookupTable`; defaults are `sbom_service`, `sbom_catalog`, `sbom_component_neighbors`.
- When the connection string is absent the service falls back to fixture JSON or deterministic in-memory seeds to keep air-gapped workflows alive.
- Ledger retention settings (env prefix `SBOM_SbomService__Ledger__`): `MaxVersionsPerArtifact`, `MaxAgeDays`, `MinVersionsToKeep`.
## 10) Open questions / dependencies
- Confirm orchestrator pause/backfill contract (shared with Runtime & Signals 140-series).
- Finalise storage table names and indexes (compound on tenant+artifactDigest+version, TTL for transient staging).
- Publish canonical LNM v1 fixtures and JSON schemas for projections and asset metadata.
- See `docs/modules/sbomservice/api/projection-read.md` for `/sboms/{snapshotId}/projection` (LNM v1, tenant-scoped, hash-returning).
- See `docs/modules/sbomservice/lineage-ledger.md` for ledger endpoints and lineage relationships.
- See `docs/modules/sbomservice/retention-policy.md` for retention configuration and audit expectations.