feat(audit): DegradingSink + durable buffer + loki/ntfy + reconcile (#54)

The classification-aware I5 audit path (§4.4): - DegradingSink.Reserve: central up → AuditCentral; central down + confidential → refuse (no buffer); central down + internal/public + buffer writable → AuditBuffered; central down + buffer unwritable → floor refuse. Record executes the reserved outcome and, when buffered, fires an ntfy alert. - FileBuffer: durable JSONL buffer that survives process restart; Confirm rewrites the file without a record, so a buffered record is cleared ONLY after its central write is confirmed. - LokiCentral: /ready probe + /loki/api/v1/push (full audit entry as the structured line). NtfyNotifier: degraded-state alerts; token only in the auth header, never logged (regression-tested). - Reconcile + StartReconcile: replay buffered records to central on recovery, confirm-then-clear per record; a failed push keeps the record buffered (no loss). SlogSink updated to the two-phase shape (always central, never fails) — the default when no loki endpoint is set. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-22 23:54:24 +02:00
parent 202212e8d5
commit 77f5e06d6b
9 changed files with 773 additions and 6 deletions
@@ -0,0 +1,191 @@
+package audit_test
+
+import (
+	"context"
+	"errors"
+	"path/filepath"
+	"testing"
+
+	"github.com/mathiasbq/hyperguild/ingestion/internal/audit"
+	"github.com/mathiasbq/hyperguild/ingestion/internal/capture"
+	"github.com/mathiasbq/hyperguild/ingestion/internal/classification"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// --- fakes ---
+
+type fakeCentral struct {
+	down    bool
+	pushed  []capture.AuditEntry
+	pushErr error
+}
+
+func (f *fakeCentral) Ready(context.Context) error {
+	if f.down {
+		return errors.New("loki down")
+	}
+	return nil
+}
+
+func (f *fakeCentral) Push(_ context.Context, e capture.AuditEntry) error {
+	if f.pushErr != nil {
+		return f.pushErr
+	}
+	f.pushed = append(f.pushed, e)
+	return nil
+}
+
+type fakeNotifier struct{ msgs []string }
+
+func (f *fakeNotifier) Notify(_ context.Context, msg string) error {
+	f.msgs = append(f.msgs, msg)
+	return nil
+}
+
+// unwritableBuffer always reports it cannot be written (floor condition).
+type unwritableBuffer struct{}
+
+func (unwritableBuffer) Writable() error                       { return errors.New("disk full") }
+func (unwritableBuffer) Append(capture.AuditEntry) error       { return errors.New("disk full") }
+func (unwritableBuffer) Pending() ([]audit.Buffered, error)    { return nil, nil }
+func (unwritableBuffer) Confirm(string) error                  { return nil }
+
+func newFileBuffer(t *testing.T) *audit.FileBuffer {
+	t.Helper()
+	b, err := audit.NewFileBuffer(filepath.Join(t.TempDir(), "audit-buffer.jsonl"))
+	require.NoError(t, err)
+	return b
+}
+
+func entry(principal string) capture.AuditEntry {
+	return capture.AuditEntry{Principal: principal, EffectiveClassification: "internal", Items: []string{"insight:x"}}
+}
+
+// --- Reserve: classification-aware decision ---
+
+func TestReserveCentralUpGrantsCentral(t *testing.T) {
+	d := audit.NewDegradingSink(&fakeCentral{}, newFileBuffer(t), &fakeNotifier{})
+	for _, lvl := range []classification.Level{classification.Public, classification.Internal, classification.Confidential} {
+		out, err := d.Reserve(context.Background(), lvl)
+		require.NoError(t, err)
+		assert.Equal(t, capture.AuditCentral, out)
+	}
+}
+
+func TestReserveConfidentialSinkDownRefuses(t *testing.T) {
+	d := audit.NewDegradingSink(&fakeCentral{down: true}, newFileBuffer(t), &fakeNotifier{})
+	_, err := d.Reserve(context.Background(), classification.Confidential)
+	require.Error(t, err, "confidential + sink down → refuse, no buffer")
+}
+
+func TestReserveInternalSinkDownBuffers(t *testing.T) {
+	d := audit.NewDegradingSink(&fakeCentral{down: true}, newFileBuffer(t), &fakeNotifier{})
+	out, err := d.Reserve(context.Background(), classification.Internal)
+	require.NoError(t, err)
+	assert.Equal(t, capture.AuditBuffered, out)
+}
+
+func TestReserveFloorRefusesWhenNothingCanRecord(t *testing.T) {
+	d := audit.NewDegradingSink(&fakeCentral{down: true}, unwritableBuffer{}, &fakeNotifier{})
+	_, err := d.Reserve(context.Background(), classification.Internal)
+	require.Error(t, err, "central down AND buffer unwritable → floor refuse")
+}
+
+// --- Record: executes the reserved outcome ---
+
+func TestRecordCentralPushes(t *testing.T) {
+	c := &fakeCentral{}
+	d := audit.NewDegradingSink(c, newFileBuffer(t), &fakeNotifier{})
+	require.NoError(t, d.Record(context.Background(), entry("p"), capture.AuditCentral))
+	assert.Len(t, c.pushed, 1)
+}
+
+func TestRecordBufferedAppendsAndNotifies(t *testing.T) {
+	buf := newFileBuffer(t)
+	nt := &fakeNotifier{}
+	d := audit.NewDegradingSink(&fakeCentral{down: true}, buf, nt)
+	require.NoError(t, d.Record(context.Background(), entry("p"), capture.AuditBuffered))
+
+	pending, err := buf.Pending()
+	require.NoError(t, err)
+	assert.Len(t, pending, 1)
+	assert.NotEmpty(t, nt.msgs, "degraded state alerts via ntfy")
+}
+
+// --- FileBuffer durability + Confirm ---
+
+func TestFileBufferSurvivesRestart(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "buf.jsonl")
+	b1, err := audit.NewFileBuffer(path)
+	require.NoError(t, err)
+	require.NoError(t, b1.Append(entry("p1")))
+	require.NoError(t, b1.Append(entry("p2")))
+
+	// "restart": a fresh FileBuffer over the same file sees the records.
+	b2, err := audit.NewFileBuffer(path)
+	require.NoError(t, err)
+	pending, err := b2.Pending()
+	require.NoError(t, err)
+	assert.Len(t, pending, 2)
+}
+
+func TestFileBufferConfirmRemovesOnlyThatRecord(t *testing.T) {
+	buf := newFileBuffer(t)
+	require.NoError(t, buf.Append(entry("keep")))
+	require.NoError(t, buf.Append(entry("drop")))
+
+	pending, _ := buf.Pending()
+	require.Len(t, pending, 2)
+	var dropID string
+	for _, p := range pending {
+		if p.Entry.Principal == "drop" {
+			dropID = p.ID
+		}
+	}
+	require.NoError(t, buf.Confirm(dropID))
+
+	after, _ := buf.Pending()
+	require.Len(t, after, 1)
+	assert.Equal(t, "keep", after[0].Entry.Principal)
+}
+
+// --- Reconcile ---
+
+func TestReconcileReplaysAndClearsOnlyAfterConfirmedWrite(t *testing.T) {
+	buf := newFileBuffer(t)
+	require.NoError(t, buf.Append(entry("a")))
+	require.NoError(t, buf.Append(entry("b")))
+	c := &fakeCentral{} // up
+	nt := &fakeNotifier{}
+
+	n, err := audit.Reconcile(context.Background(), c, buf, nt)
+	require.NoError(t, err)
+	assert.Equal(t, 2, n)
+	assert.Len(t, c.pushed, 2, "buffered records replayed to central")
+
+	pending, _ := buf.Pending()
+	assert.Empty(t, pending, "buffer cleared after confirmed central writes")
+}
+
+func TestReconcileNoopWhenCentralDown(t *testing.T) {
+	buf := newFileBuffer(t)
+	require.NoError(t, buf.Append(entry("a")))
+	n, err := audit.Reconcile(context.Background(), &fakeCentral{down: true}, buf, &fakeNotifier{})
+	require.NoError(t, err)
+	assert.Equal(t, 0, n)
+	pending, _ := buf.Pending()
+	assert.Len(t, pending, 1, "records stay buffered while central is down")
+}
+
+func TestReconcileKeepsRecordWhenPushFails(t *testing.T) {
+	buf := newFileBuffer(t)
+	require.NoError(t, buf.Append(entry("a")))
+	// Ready ok but Push fails → record must remain buffered (not lost).
+	c := &fakeCentral{pushErr: errors.New("push rejected")}
+	n, err := audit.Reconcile(context.Background(), c, buf, &fakeNotifier{})
+	require.NoError(t, err)
+	assert.Equal(t, 0, n)
+	pending, _ := buf.Pending()
+	assert.Len(t, pending, 1)
+}