hub: F-14 part 1 — host-deletion provenance (host_deletions, in-tx)

Every DeleteHost writes a provenance row INSIDE the cascade tx: host_id, customer_id, deleted_at, escrow_acked. escrow_acked = ack given over a PRESENT escrow row (acknowledged destruction, not a vacuous checkbox). LatestHostDeletion(customer) serves the F-14 gate — newest record only, so an old acked deletion never whitelists a newer un-acked one. No backfill by design: pre-record deletions stay on the manual path. Red-proof: dropping the in-tx INSERT fails TestDeleteHost_ProvenanceRecord ("no deletion record written") + the part-2 scenario-A test.
2026-07-13 14:44:42 +02:00
parent 74fa61c7bc
commit 2321077800
2 changed files with 168 additions and 8 deletions
@@ -174,6 +174,89 @@ func TestDeleteHost_EscrowFlagSemantics(t *testing.T) {
 	}
 }

+// v0.53.0 F-14 provenance — the deletion record is written IN the delete tx, with
+// escrow_acked reflecting an ACTUAL acknowledged destruction (ack over a present escrow).
+// RED-PROOF (Part 1): dropping the provenance INSERT from DeleteHost fails the acked case
+// (LatestHostDeletion returns nil — the gate finds nothing).
+func TestDeleteHost_ProvenanceRecord(t *testing.T) {
+	s := newTestStore(t)
+
+	// Escrow-ack delete → record with escrow_acked = true.
+	seedHostWithArtifacts(t, s, "prov-acked", "cust-f14")
+	if err := s.DeleteHost("prov-acked", true); err != nil {
+		t.Fatalf("DeleteHost: %v", err)
+	}
+	rec, err := s.LatestHostDeletion("cust-f14")
+	if err != nil {
+		t.Fatalf("LatestHostDeletion: %v", err)
+	}
+	if rec == nil {
+		t.Fatal("no deletion record written by the escrow-ack delete")
+	}
+	if rec.HostID != "prov-acked" || rec.CustomerID != "cust-f14" || !rec.EscrowAcked {
+		t.Errorf("record = %+v, want host=prov-acked customer=cust-f14 escrow_acked=true", rec)
+	}
+	if rec.DeletedAt.IsZero() {
+		t.Error("deleted_at not populated")
+	}
+
+	// Delete WITHOUT escrow (none present) → record exists but escrow_acked = false, even
+	// though deleteEscrow=true was passed: ticking the box over NOTHING is not an
+	// acknowledged destruction.
+	if err := s.UpsertHost(&Host{HostID: "prov-noescrow", CustomerID: "cust-noesc", APIKey: "k"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.DeleteHost("prov-noescrow", true); err != nil {
+		t.Fatalf("DeleteHost: %v", err)
+	}
+	rec, err = s.LatestHostDeletion("cust-noesc")
+	if err != nil || rec == nil {
+		t.Fatalf("LatestHostDeletion = %+v, %v; want a record", rec, err)
+	}
+	if rec.EscrowAcked {
+		t.Error("escrow_acked = true for a host with NO escrow row — vacuous ack must record false")
+	}
+
+	// The refused delete (escrow present, no ack) writes NOTHING — the tx never ran.
+	seedHostWithArtifacts(t, s, "prov-refused", "cust-refused")
+	if err := s.DeleteHost("prov-refused", false); !errors.Is(err, ErrHostEscrowPresent) {
+		t.Fatalf("expected escrow refusal, got %v", err)
+	}
+	if rec, _ := s.LatestHostDeletion("cust-refused"); rec != nil {
+		t.Errorf("refused delete wrote a provenance record: %+v", rec)
+	}
+
+	// Customer with no deletions ever → nil, nil (the pre-v0.53.0 shape — manual path).
+	if rec, err := s.LatestHostDeletion("cust-never"); err != nil || rec != nil {
+		t.Errorf("LatestHostDeletion(no deletions) = %+v, %v; want nil, nil", rec, err)
+	}
+}
+
+// The gate reads the MOST RECENT record: an old acked deletion must not whitelist a newer
+// un-acked one (the F-14 law is about the deletion that orphaned the CURRENT tenancy).
+func TestLatestHostDeletion_NewestWins(t *testing.T) {
+	s := newTestStore(t)
+
+	seedHostWithArtifacts(t, s, "gen1-host", "cust-seq")
+	if err := s.DeleteHost("gen1-host", true); err != nil { // acked
+		t.Fatal(err)
+	}
+	if err := s.UpsertHost(&Host{HostID: "gen2-host", CustomerID: "cust-seq", APIKey: "k2"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.DeleteHost("gen2-host", false); err != nil { // no escrow → un-acked record
+		t.Fatal(err)
+	}
+
+	rec, err := s.LatestHostDeletion("cust-seq")
+	if err != nil || rec == nil {
+		t.Fatalf("LatestHostDeletion = %+v, %v", rec, err)
+	}
+	if rec.HostID != "gen2-host" || rec.EscrowAcked {
+		t.Errorf("latest record = %+v, want the NEWER un-acked gen2-host row", rec)
+	}
+}
+
 func TestCountHostArtifacts(t *testing.T) {
 	s := newTestStore(t)
 	seedHostWithArtifacts(t, s, "impact-host", "cust-e")
@@ -555,6 +555,27 @@ func (s *Store) migrate() error {
 		return err
 	}

+	// v0.53.0 — host-deletion provenance (F-14, operator ruling 2026-07-13): one row per DeleteHost,
+	// written INSIDE the delete transaction. escrow_acked records whether the host was removed
+	// through the escrow-ack flow (the operator explicitly acknowledged destroying a PRESENT escrow
+	// row — acknowledged key destruction). The PBS-DR enable path may auto-re-issue a surviving ep0
+	// tenancy ONLY when the customer's most recent record here has escrow_acked=1; no record (all
+	// pre-v0.53.0 deletions — deliberately NO backfill) or an un-acked record keeps the manual
+	// re-issue path the only one (never-silently-re-key law).
+	_, err = s.db.Exec(`
+		CREATE TABLE IF NOT EXISTS host_deletions (
+			id           INTEGER PRIMARY KEY AUTOINCREMENT,
+			host_id      TEXT NOT NULL,
+			customer_id  TEXT NOT NULL,
+			deleted_at   DATETIME NOT NULL DEFAULT (datetime('now')),
+			escrow_acked INTEGER NOT NULL DEFAULT 0
+		);
+		CREATE INDEX IF NOT EXISTS idx_host_deletions_customer ON host_deletions(customer_id, id DESC);
+	`)
+	if err != nil {
+		return err
+	}
+
 	// v0.51.0 dr_tier one-time legacy backfill — see the ALTER above; runs last so every table
 	// it touches (hosts, customer_configs) exists on a fresh DB too (where it finds nothing).
 	if drTierAlterErr == nil {
@@ -1893,18 +1914,21 @@ func (s *Store) CountHostArtifacts(hostID string) (HostArtifacts, error) {
 // reconciler's 5-minute declarative full-list push converges the endpoint after the row
 // disappears — no bump, no reconciler change. log_bundle rows die by scope_id == host_id
 // (agent channel); customer-scoped bundles (scope_id == customer_id) are NOT touched.
+//
+// v0.53.0 (F-14 provenance): every delete also writes a host_deletions row IN THE SAME tx.
+// escrow_acked = deleteEscrow AND an escrow row was actually present — "removed through the
+// escrow-ack flow" means an acknowledged destruction happened, not merely that the checkbox
+// was ticked over nothing.
 func (s *Store) DeleteHost(hostID string, deleteEscrow bool) error {
 	if hostID == "" {
 		return fmt.Errorf("DeleteHost: empty host_id")
 	}
-	if !deleteEscrow {
-		var n int
-		if err := s.db.QueryRow(`SELECT EXISTS(SELECT 1 FROM host_escrow WHERE host_id = ?)`, hostID).Scan(&n); err != nil {
-			return fmt.Errorf("DeleteHost %s: escrow check: %w", hostID, err)
-		}
-		if n != 0 {
-			return ErrHostEscrowPresent
-		}
+	var escrowPresent int
+	if err := s.db.QueryRow(`SELECT EXISTS(SELECT 1 FROM host_escrow WHERE host_id = ?)`, hostID).Scan(&escrowPresent); err != nil {
+		return fmt.Errorf("DeleteHost %s: escrow check: %w", hostID, err)
+	}
+	if !deleteEscrow && escrowPresent != 0 {
+		return ErrHostEscrowPresent
 	}

 	tx, err := s.db.Begin()
@@ -1913,6 +1937,25 @@ func (s *Store) DeleteHost(hostID string, deleteEscrow bool) error {
 	}
 	defer tx.Rollback()

+	// Provenance first (reads the host row this tx is about to delete). A host_id that has no
+	// row deletes nothing anyway — skip the record rather than inventing an empty customer_id.
+	var customerID string
+	switch err := tx.QueryRow(`SELECT customer_id FROM hosts WHERE host_id = ?`, hostID).Scan(&customerID); err {
+	case nil:
+		acked := 0
+		if deleteEscrow && escrowPresent != 0 {
+			acked = 1
+		}
+		if _, err := tx.Exec(`INSERT INTO host_deletions (host_id, customer_id, escrow_acked) VALUES (?, ?, ?)`,
+			hostID, customerID, acked); err != nil {
+			return fmt.Errorf("DeleteHost %s: provenance record: %w", hostID, err)
+		}
+	case sql.ErrNoRows:
+		// no host row — fall through, the deletes below are no-ops
+	default:
+		return fmt.Errorf("DeleteHost %s: customer lookup: %w", hostID, err)
+	}
+
 	stmts := []string{
 		`DELETE FROM guests WHERE host_id = ?`,
 		`DELETE FROM host_reports WHERE host_id = ?`,
@@ -1935,6 +1978,40 @@ func (s *Store) DeleteHost(hostID string, deleteEscrow bool) error {
 	return tx.Commit()
 }

+// HostDeletion is one host-removal provenance record (v0.53.0, F-14). EscrowAcked means the
+// operator removed the host through the escrow-ack flow — an acknowledged destruction of the
+// host's key custody, the ONLY state that permits the PBS-DR auto-re-issue.
+type HostDeletion struct {
+	HostID      string
+	CustomerID  string
+	DeletedAt   time.Time
+	EscrowAcked bool
+}
+
+// LatestHostDeletion returns the customer's MOST RECENT host-deletion record (nil when the
+// customer has none — every pre-v0.53.0 deletion, by design: no backfill invents provenance).
+// The latest record is the one that orphaned a surviving ep0 tenancy, so the F-14 gate reads
+// exactly this row — an older acked record must not whitelist a newer un-acked deletion.
+func (s *Store) LatestHostDeletion(customerID string) (*HostDeletion, error) {
+	var d HostDeletion
+	var deletedAt string
+	var acked int
+	err := s.db.QueryRow(`
+		SELECT host_id, customer_id, deleted_at, escrow_acked
+		FROM host_deletions WHERE customer_id = ?
+		ORDER BY id DESC LIMIT 1`, customerID,
+	).Scan(&d.HostID, &d.CustomerID, &deletedAt, &acked)
+	if err == sql.ErrNoRows {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, err
+	}
+	d.DeletedAt = parseSQLiteTime(deletedAt)
+	d.EscrowAcked = acked != 0
+	return &d, nil
+}
+
 // UpsertHost creates or updates a host identity (used by the admin mint). On
 // conflict it updates only operator-settable identity fields + updated_at; it does
 // NOT touch the reality columns (agent_version/last_report_at) or the inert intent