Files
git.stella-ops.org/docs/scanner-core-contracts.md

8.6 KiB
Raw Blame History

Scanner Core Contracts

The Scanner Core library provides shared contracts, observability helpers, and security utilities consumed by Scanner.WebService, Scanner.Worker, analyzers, and tooling. These primitives guarantee deterministic identifiers, timestamps, and log context for all scanning flows.

Canonical DTOs

  • ScanJob & ScanJobStatus canonical job metadata (image reference/digest, tenant, correlation ID, timestamps, failure details). Constructors normalise timestamps to UTC microsecond precision and canonicalise image digests. Round-trips with JsonSerializerDefaults.Web using ScannerJsonOptions.
  • ScanProgressEvent & ScanStage/ScanProgressEventKind stage-level progress surface for queue/stream consumers. Includes deterministic sequence numbers, optional progress percentage, attributes, and attached ScannerError.
  • ScannerError & ScannerErrorCode shared error taxonomy spanning queue, analyzers, storage, exporters, and signing. Carries severity, retryability, structured details, and microsecond-precision timestamps.
  • ScanJobId strongly-typed identifier rendered as Guid (lowercase N format) with deterministic parsing.

Canonical JSON samples

The golden fixtures consumed by ScannerCoreContractsTests document the wire shape shared with downstream services. They live under src/StellaOps.Scanner.Core.Tests/Fixtures/ and a representative extract is shown below.

{
  "id": "8f4cc9c582454b9d9b4f5ae049631b7d",
  "status": "running",
  "imageReference": "registry.example.com/stellaops/scanner:1.2.3",
  "imageDigest": "sha256:abcdef",
  "createdAt": "2025-10-18T14:30:15.123456+00:00",
  "updatedAt": "2025-10-18T14:30:20.123456+00:00",
  "correlationId": "scan-analyzeoperatingsystem-8f4cc9c582454b9d9b4f5ae049631b7d",
  "tenantId": "tenant-a",
  "metadata": {
    "requestId": "req-1234",
    "source": "ci"
  },
  "failure": {
    "code": "analyzerFailure",
    "severity": "error",
    "message": "Analyzer failed to parse layer",
    "timestamp": "2025-10-18T14:30:15.123456+00:00",
    "retryable": false,
    "stage": "AnalyzeOperatingSystem",
    "component": "os-analyzer",
    "details": {
      "layerDigest": "sha256:deadbeef",
      "attempt": "1"
    }
  }
}

Progress events follow the same conventions (jobId, stage, kind, timestamp, attributes, optional embedded ScannerError). The fixtures are verified via deterministic JSON comparison in every CI run.

Deterministic helpers

  • ScannerIdentifiers derives ScanJobId, correlation IDs, and SHA-256 hashes from normalised inputs (image reference/digest, tenant, salt). Ensures case-insensitive stability and reproducible metric keys.
  • ScannerTimestamps trims to microsecond precision, provides ISO-8601 (yyyy-MM-ddTHH:mm:ss.ffffffZ) rendering, and parsing helpers.
  • ScannerJsonOptions standard JSON options (web defaults, camel-case enums) shared by services/tests.
  • ScanAnalysisStore & ScanAnalysisKeys shared in-memory analysis cache flowing through Worker stages. OS analyzers populate analysis.os.packages (raw output), analysis.os.fragments (per-analyzer component fragments), and merge into analysis.layers.fragments so emit/diff stages can compose SBOMs and diffs without knowledge of individual analyzer implementations.

Observability primitives

  • ScannerDiagnostics global ActivitySource/Meter for scanner components. StartActivity seeds deterministic tags (job_id, stage, component, correlation_id).
  • ScannerMetricNames centralises metric prefixes (stellaops.scanner.*) and deterministic job/event tag builders.
  • ScannerCorrelationContext & ScannerCorrelationContextAccessor ambient correlation propagation via AsyncLocal for log scopes, metrics, and diagnostics.
  • ScannerLogExtensions ILogger scopes for jobs/progress events with automatic correlation context push, minimal allocations, and consistent structured fields.

Observability overhead validation

A micro-benchmark executed on 2025-10-19 (4vCPU runner, .NET 10.0.100-rc.1) measured the average scope cost across 1000000 iterations:

Scope Mean (µs/call)
BeginScanScope (logger attached) 0.80
BeginScanScope (noop logger) 0.31
BeginProgressScope 0.57

To reproduce, run dotnet test src/StellaOps.Scanner.Core.Tests -c Release (see ScannerLogExtensionsPerformanceTests) or copy the snippet below into a throwaway dotnet run console project and execute it with dotnet run -c Release:

using System.Collections.Generic;
using System.Diagnostics;
using Microsoft.Extensions.Logging;
using StellaOps.Scanner.Core.Contracts;
using StellaOps.Scanner.Core.Observability;
using StellaOps.Scanner.Core.Utility;

var factory = LoggerFactory.Create(builder => builder.AddFilter(static _ => true));
var logger = factory.CreateLogger("bench");

var jobId = ScannerIdentifiers.CreateJobId("registry.example.com/stellaops/scanner:1.2.3", "sha256:abcdef", "tenant-a", "benchmark");
var correlationId = ScannerIdentifiers.CreateCorrelationId(jobId, nameof(ScanStage.AnalyzeOperatingSystem));
var now = ScannerTimestamps.Normalize(new DateTimeOffset(2025, 10, 19, 12, 0, 0, TimeSpan.Zero));

var job = new ScanJob(jobId, ScanJobStatus.Running, "registry.example.com/stellaops/scanner:1.2.3", "sha256:abcdef", now, now, correlationId, "tenant-a", new Dictionary<string, string>(StringComparer.Ordinal) { ["requestId"] = "req-bench" });
var progress = new ScanProgressEvent(jobId, ScanStage.AnalyzeOperatingSystem, ScanProgressEventKind.Progress, 42, now, 10.5, "benchmark", new Dictionary<string, string>(StringComparer.Ordinal) { ["sample"] = "true" });

Console.WriteLine("Scanner Core Observability micro-bench (1,000,000 iterations)");
Report("BeginScanScope (logger)", Measure(static ctx => ctx.Logger.BeginScanScope(ctx.Job, ctx.Stage, ctx.Component), new ScopeContext(logger, job, nameof(ScanStage.AnalyzeOperatingSystem), "os-analyzer")));
Report("BeginScanScope (no logger)", Measure(static ctx => ScannerLogExtensions.BeginScanScope(null, ctx.Job, ctx.Stage, ctx.Component), new ScopeContext(logger, job, nameof(ScanStage.AnalyzeOperatingSystem), "os-analyzer")));
Report("BeginProgressScope", Measure(static ctx => ctx.Logger.BeginProgressScope(ctx.Progress!, ctx.Component), new ScopeContext(logger, job, nameof(ScanStage.AnalyzeOperatingSystem), "os-analyzer", progress)));

static double Measure(Func<ScopeContext, IDisposable> factory, ScopeContext context)
{
    const int iterations = 1_000_000;
    for (var i = 0; i < 10_000; i++)
    {
        using var scope = factory(context);
    }

    GC.Collect();
    GC.WaitForPendingFinalizers();
    GC.Collect();

    var sw = Stopwatch.StartNew();
    for (var i = 0; i < iterations; i++)
    {
        using var scope = factory(context);
    }

    sw.Stop();
    return sw.Elapsed.TotalSeconds * 1_000_000 / iterations;
}

static void Report(string label, double microseconds)
    => Console.WriteLine($"{label,-28}: {microseconds:F3} µs");

readonly record struct ScopeContext(ILogger Logger, ScanJob Job, string? Stage, string? Component, ScanProgressEvent? Progress = null);

Both guardrails enforce the ≤5µs acceptance target for SP9-G1.

Security utilities

  • AuthorityTokenSource caches short-lived OpToks per audience+scope using deterministic keys and refresh skew (default 30s). Integrates with StellaOps.Auth.Client.
  • DpopProofValidator validates DPoP proofs (alg allowlist, htm/htu, nonce, replay window, signature) backed by pluggable IDpopReplayCache. Ships with InMemoryDpopReplayCache for restart-only deployments.
  • RestartOnlyPluginGuard enforces restart-time plug-in registration (deterministic path normalisation; throws if new plug-ins added post-seal).
  • ServiceCollectionExtensions.AddScannerAuthorityCore DI helper wiring Authority client, OpTok source, DPoP validation, replay cache, and plug-in guard.

Testing guarantees

Unit tests (StellaOps.Scanner.Core.Tests) assert:

  • DTO JSON round-trips are stable and deterministic (ScannerCoreContractsTests + golden fixtures).
  • Identifier/hash helpers ignore case and emit lowercase hex.
  • Timestamp normalisation retains UTC semantics.
  • Log scopes push/pop correlation context predictably while staying under the 5µs envelope.
  • Authority token caching honours refresh skew and invalidation.
  • DPoP validator accepts valid proofs, rejects nonce mismatch/replay, and enforces signature validation.
  • Restart-only plug-in guard blocks runtime additions post-seal.