feat: agent-capability gate for coupled features — typed StatusError + Supports probe/cache + netstorage add gate (option-1)

- agentapi: non-2xx GETs now surface as typed *StatusError (same text); features.go adds Feature/SupportState/SupportCache (route probe, TTL 5m, Yes/No cached, Unknown never cached or refused) + Client.Supports - web: handleNetStorageAdd refuses up front (412, code agent_outdated, Hungarian message) when the agent predates /netstorage/verify-status (= pre-0.81 add semantics); gate runs BEFORE the single-flight claim; SupportUnknown passes through to the existing agent-error paths - netAddSupport page-render helper lands here; its template consumer follows - tests: T1 gate refusal (job never starts, slot free), T2 unchanged happy path + warm-cache negative assertion, T3 indeterminate never 'too old', T4 classification incl. the string-match trap, T6 TTL, wire-level 404-typing; red-proofs RP1-RP4 run and reverted Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PSK5g6qYLknKj8u3QAFEr6
2026-07-11 13:45:49 +02:00
parent 5d06ecf374
commit d347dc48d2
6 changed files with 511 additions and 1 deletions
@@ -0,0 +1,168 @@
+package web
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/agentapi"
+)
+
+// Capability-gate tests (agent-outdated backstop): the probe runs through the SAME netAgent seam
+// as the orchestration — fakeNetAgent's NetVerifyStatus doubles as the probed route.
+
+// counts reads the fake's recorded call counters under its lock.
+func (f *fakeNetAgent) counts() (adds, verifyCalls int) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.addCalls, f.verifyPoll
+}
+
+// setAddRes swaps the scripted add result between requests (under the fake's lock).
+func (f *fakeNetAgent) setAddRes(res agentapi.NetStorageAddResult) {
+	f.mu.Lock()
+	f.addRes = res
+	f.mu.Unlock()
+}
+
+// postNetAdd drives the REAL handler (the pipeline a user triggers) with a valid body.
+func postNetAdd(t *testing.T, s *Server, name string) *httptest.ResponseRecorder {
+	t.Helper()
+	body := `{"name":"` + name + `","protocol":"nfs","server":"10.0.0.5","export":"/srv/` + name + `"}`
+	r := httptest.NewRequest(http.MethodPost, "/api/storage/netstorage/add", strings.NewReader(body))
+	w := httptest.NewRecorder()
+	s.handleNetStorageAdd(w, r)
+	return w
+}
+
+// --- T1 (Scenario A): old agent ⇒ sync refusal; the job NEVER starts ---------------------------
+// Companion red-proof: remove the gate call in handleNetStorageAdd → the job starts against the
+// old add semantics → the "never started" + zero-add-calls assertions fail.
+func TestNetAddGate_OldAgent_RefusedUpFront(t *testing.T) {
+	s := testServer(t)
+	// The ≤0.80 shape: the probed route 404s (typed — the wire path is pinned in agentapi tests).
+	agent := &fakeNetAgent{
+		addRes:    okAddRes("media"),
+		verifyErr: &agentapi.StatusError{Path: "/netstorage/verify-status", Code: http.StatusNotFound},
+	}
+	s.netAgentFn = func() (netAgent, error) { return agent, nil }
+	s.netProbeFn = func(context.Context, string) probeOutcome { t.Error("probe must never run on a gated add"); return probeOutcome{} }
+
+	w := postNetAdd(t, s, "media")
+	if w.Code != http.StatusPreconditionFailed {
+		t.Fatalf("gate: got %d want 412 (%s)", w.Code, w.Body.String())
+	}
+	if !strings.Contains(w.Body.String(), `"code":"agent_outdated"`) {
+		t.Errorf("refusal must carry the machine code agent_outdated: %s", w.Body.String())
+	}
+	if !strings.Contains(w.Body.String(), "Az ügynök frissítése szükséges ehhez a funkcióhoz") {
+		t.Errorf("refusal must carry the §Part-3 Hungarian message: %s", w.Body.String())
+	}
+	// The orchestration was NEVER started and no agent add/remove ran.
+	if j := s.netAdd.snapshot(); j != nil {
+		t.Errorf("job slot must stay empty on a gated add, got %+v", j)
+	}
+	adds, _ := agent.counts()
+	if adds != 0 {
+		t.Errorf("agent add calls = %d, want 0", adds)
+	}
+	if got := agent.removed(); len(got) != 0 {
+		t.Errorf("no rollback should ever run (removes=%v)", got)
+	}
+	if got := networkPathCount(s); got != 0 {
+		t.Errorf("nothing may register on a gated add (got %d)", got)
+	}
+	// The single-flight slot was NEVER claimed — it must still be acquirable.
+	if !s.netAdd.acquire(&netAddJob{Name: "slot-check"}) {
+		t.Error("single-flight slot was consumed by the refused add")
+	}
+	s.netAdd.release()
+}
+
+// --- T2 (Scenario B): current agent ⇒ pre-gate behavior + ONE probe per cache window ------------
+// Companion red-proof: drop the cache in SupportCache.Supports (always probe) → the warm-cache
+// negative assertion (verify calls stays 1) fails with 2.
+func TestNetAddGate_CurrentAgent_UnchangedAndProbeCached(t *testing.T) {
+	s := testServer(t)
+	// Pre-verify-shaped OK result (Verify != "started"): the orchestrator skips the verify poll, so
+	// the fake's NetVerifyStatus count measures the GATE PROBE alone.
+	res1 := okAddRes("m1")
+	res1.Verify, res1.JobID = "", ""
+	agent := &fakeNetAgent{addRes: res1, verify: agentapi.NetVerifyStatus{Phase: "none"}}
+	s.netAgentFn = func() (netAgent, error) { return agent, nil }
+	s.netProbeFn = func(context.Context, string) probeOutcome { return probeOutcome{OK: true} }
+
+	// Add #1 — identical end-to-end shape to the pre-gate happy path (C1's contract).
+	w := postNetAdd(t, s, "m1")
+	if w.Code != http.StatusOK || !strings.Contains(w.Body.String(), `"started":true`) {
+		t.Fatalf("add #1: got %d %s, want 200 started", w.Code, w.Body.String())
+	}
+	if job := waitNetAdd(t, s); job.Phase != netAddPhaseDone {
+		t.Fatalf("add #1 phase = %s (category=%s detail=%s), want done", job.Phase, job.Category, job.Detail)
+	}
+	if got := networkPathCount(s); got != 1 {
+		t.Fatalf("registered paths after add #1 = %d, want 1", got)
+	}
+	if got := agent.removed(); len(got) != 0 {
+		t.Fatalf("happy path must not roll back: %v", got)
+	}
+
+	// Add #2 inside the TTL window — the probe must answer from the cache.
+	res2 := okAddRes("m2")
+	res2.Verify, res2.JobID = "", ""
+	agent.setAddRes(res2)
+	w = postNetAdd(t, s, "m2")
+	if w.Code != http.StatusOK {
+		t.Fatalf("add #2: got %d (%s)", w.Code, w.Body.String())
+	}
+	if job := waitNetAdd(t, s); job.Phase != netAddPhaseDone || job.Name != "m2" {
+		t.Fatalf("add #2 job = %+v, want done/m2", job)
+	}
+	adds, verifyCalls := agent.counts()
+	if adds != 2 {
+		t.Errorf("agent add calls = %d, want 2", adds)
+	}
+	if verifyCalls != 1 { // the NEGATIVE assertion: no probe increment on a warm cache
+		t.Errorf("probe (verify-status) calls across two adds = %d, want exactly 1 (cached)", verifyCalls)
+	}
+}
+
+// --- T3 (Scenario C): probe indeterminate ⇒ NEVER "too old"; the existing error paths speak -----
+// Companion red-proof (the classification trap): mutate classifySupportErr to return SupportNo on
+// ANY error → both subtests fail with the false 412 agent_outdated refusal.
+func TestNetAddGate_ProbeIndeterminate_PassesThrough(t *testing.T) {
+	cases := map[string]error{
+		"transport error": errors.New("dial tcp 10.0.0.9:8443: connect: connection refused"),
+		"http 5xx":        &agentapi.StatusError{Path: "/netstorage/verify-status", Code: http.StatusBadGateway},
+	}
+	for name, perr := range cases {
+		t.Run(name, func(t *testing.T) {
+			s := testServer(t)
+			agent := &fakeNetAgent{
+				verifyErr: perr,
+				addErr:    errors.New("agentapi: POST /netstorage/add: connection refused"),
+			}
+			s.netAgentFn = func() (netAgent, error) { return agent, nil }
+
+			w := postNetAdd(t, s, "media")
+			// The gate must NOT refuse: the add is accepted and fails through the EXISTING
+			// agent-error path with its honest message.
+			if w.Code != http.StatusOK {
+				t.Fatalf("indeterminate probe must pass the gate: got %d (%s)", w.Code, w.Body.String())
+			}
+			if strings.Contains(w.Body.String(), "agent_outdated") || strings.Contains(w.Body.String(), "frissítés") {
+				t.Fatalf("a down agent must never be called outdated: %s", w.Body.String())
+			}
+			job := waitNetAdd(t, s)
+			if job.Phase != netAddPhaseFailed || job.Category != "agent_error" {
+				t.Errorf("job = %s/%s, want failed/agent_error (the existing path)", job.Phase, job.Category)
+			}
+			if got := networkPathCount(s); got != 0 {
+				t.Errorf("nothing may register (got %d)", got)
+			}
+		})
+	}
+}