feat(auth): add GET /v1/me/can capability probe and OIDC token-refresh logging (#129)

Author: hyp3rd

CHANGELOG.md

@@ -8,6 +8,27 @@ All notable changes to HyperCache are recorded here. The format follows
 
 ### Added
 
+- **Token-refresh visibility for the OIDC source.** Closes RFC 0003 open question 6: the
+  `WithOIDCClientCredentials` source now wraps its `oauth2.TokenSource` with a logger that emits one
+  `"oidc token rotated"` Info line per real rotation (expiry change), staying silent on cached returns.
+  Operators debugging "why are my requests suddenly 401?" now see token age in the structured log alongside
+  the other lifecycle events. The wrapper holds the `*Client` by reference rather than capturing
+  `c.logger` at construction time, so `WithLogger` applied AFTER `WithOIDCClientCredentials` still reaches
+  the rotation log surface. Three unit tests in [`pkg/client/oidc_logging_test.go`](pkg/client/oidc_logging_test.go)
+  cover the rotation-logs case, the cached-returns-stay-silent case, and the nil-Client defensive path.
+- **`GET /v1/me/can` capability probe + `Client.Can(ctx, capability)` SDK method.** Closes RFC 0003 open
+  question 5: callers can now check "do I have write?" without the speculative-write pattern (try the
+  action, catch the 403). The server endpoint validates against a closed set of capability strings
+  (`cache.read` / `cache.write` / `cache.admin`); unknown values return 400 BAD_REQUEST so typos surface as
+  client errors rather than silently degrading to allowed=false. The SDK method mirrors this:
+  `(true, nil)` / `(false, nil)` for the allow/deny answers; `errors.Is(err, ErrBadRequest)` for the
+  spelling-mistake path. `Identity.HasCapability` added to [`pkg/httpauth/policy.go`](pkg/httpauth/policy.go)
+  as the single authoritative check used by both the server handler and the SDK. Three handler tests in
+  [`cmd/hypercache-server/me_test.go`](cmd/hypercache-server/me_test.go) cover allowed/denied/unknown;
+  three SDK tests in [`pkg/client/client_test.go`](pkg/client/client_test.go) cover the parallel surface.
+  OpenAPI spec ([`cmd/hypercache-server/openapi.yaml`](cmd/hypercache-server/openapi.yaml)) gains the
+  `/v1/me/can` operation + `CanResponse` schema. New "Probing a single capability with `Can`" and
+  "Token-refresh visibility" sections in [`docs/client-sdk.md`](docs/client-sdk.md).
 - **Chaos hooks for resilience testing (Phase 7).** New
   [`backend.WithDistChaos(*Chaos)`](pkg/backend/dist_chaos.go) option transparently wraps the dist transport
   with configurable fault injection — drop rate and latency injection, both with per-call probability rolls

Makefile

@@ -216,6 +216,8 @@ docs-serve: docs-build
 	PYENV_VERSION=mkdocs mkdocs serve
 
 pre-commit:
+	@eval "$$(pyenv init -)" && \
+	pyenv activate pre-commit && \
 	pre-commit run -a trailing-whitespace && \
 	pre-commit run -a end-of-file-fixer && \
 	pre-commit run -a markdownlint && \

cmd/hypercache-server/main.go

@@ -465,6 +465,7 @@ func registerClientRoutes(app *fiber.App, policy httpauth.Policy, nodeCtx *nodeC
 	app.Delete("/v1/cache/:key", write, func(c fiber.Ctx) error { return handleDelete(c, nodeCtx) })
 	app.Get("/v1/owners/:key", read, func(c fiber.Ctx) error { return handleOwners(c, nodeCtx) })
 	app.Get("/v1/me", read, handleMe)
+	app.Get("/v1/me/can", read, handleCan)
 
 	app.Post("/v1/cache/batch/get", read, func(c fiber.Ctx) error { return handleBatchGet(c, nodeCtx) })
 	app.Post("/v1/cache/batch/put", write, func(c fiber.Ctx) error { return handleBatchPut(c, nodeCtx) })
@@ -1335,6 +1336,74 @@ func handleMe(c fiber.Ctx) error {
 	})
 }
 
+// canResponse is the body of GET /v1/me/can?capability=<name>.
+// `Allowed` is the discrimination result; `Capability` echoes the
+// caller's input so log scraping ties allow/deny to the asked
+// capability without parsing the query string again.
+type canResponse struct {
+	Capability string `json:"capability"`
+	Allowed    bool   `json:"allowed"`
+}
+
+// capability strings — closed set in the `cache.` namespace.
+// Unknown values return 400 rather than silently false so callers
+// detect typos instead of shipping broken authz logic to prod.
+const (
+	capabilityCacheRead  = "cache.read"
+	capabilityCacheWrite = "cache.write"
+	capabilityCacheAdmin = "cache.admin"
+)
+
+// isKnownCapability reports whether s is one of the three
+// recognized capability strings. Switch-based so a future
+// capability is one named const + one case.
+func isKnownCapability(s string) bool {
+	switch s {
+	case capabilityCacheRead, capabilityCacheWrite, capabilityCacheAdmin:
+		return true
+	default:
+		return false
+	}
+}
+
+// handleCan implements GET /v1/me/can?capability=cache.write —
+// per-capability authorization probe. Caller passes a capability
+// string; the response says whether the resolved identity holds
+// it. Cheaper than the speculative-write pattern (try the write,
+// catch the 403), and stable across future scope-to-capability
+// refactors (clients key off the capability string, not the
+// internal scope shape).
+//
+// Requires the `read` scope — same threshold as /v1/me. Unknown
+// capability values fail BAD_REQUEST so typos don't silently
+// answer "not allowed" when the real issue is the caller's
+// spelling.
+func handleCan(c fiber.Ctx) error {
+	capability := c.Query("capability")
+	if capability == "" {
+		return jsonErr(c, fiber.StatusBadRequest, codeBadRequest, "missing 'capability' query parameter")
+	}
+
+	if !isKnownCapability(capability) {
+		return jsonErr(c, fiber.StatusBadRequest, codeBadRequest, "unknown capability '"+capability+"'")
+	}
+
+	identity, ok := c.Locals(httpauth.IdentityKey).(httpauth.Identity)
+	if !ok {
+		return jsonErr(
+			c,
+			fiber.StatusInternalServerError,
+			codeInternal,
+			"identity not resolved by middleware (wiring bug)",
+		)
+	}
+
+	return c.JSON(canResponse{
+		Capability: capability,
+		Allowed:    identity.HasCapability(capability),
+	})
+}
+
 func main() { os.Exit(run()) }
 
 // run is the testable main body — separated so deferred cleanup

cmd/hypercache-server/me_test.go

@@ -35,7 +35,7 @@ func TestHandleMe_BodyShape(t *testing.T) {
 			want: meResponse{
 				ID:           "ops-readonly",
 				Scopes:       []string{"read"},
-				Capabilities: []string{"cache.read"},
+				Capabilities: []string{capabilityCacheRead},
 			},
 		},
 		{
@@ -47,7 +47,7 @@ func TestHandleMe_BodyShape(t *testing.T) {
 			want: meResponse{
 				ID:           "ops-rw",
 				Scopes:       []string{"read", "write"},
-				Capabilities: []string{"cache.read", "cache.write"},
+				Capabilities: []string{capabilityCacheRead, capabilityCacheWrite},
 			},
 		},
 		{
@@ -59,7 +59,7 @@ func TestHandleMe_BodyShape(t *testing.T) {
 			want: meResponse{
 				ID:           "anonymous",
 				Scopes:       []string{"read", "write", "admin"},
-				Capabilities: []string{"cache.read", "cache.write", "cache.admin"},
+				Capabilities: []string{capabilityCacheRead, capabilityCacheWrite, capabilityCacheAdmin},
 			},
 		},
 	}
@@ -172,3 +172,144 @@ func TestHandleMe_MissingLocals(t *testing.T) {
 		t.Fatalf("status: got %d, want 500", resp.StatusCode)
 	}
 }
+
+// TestHandleCan_AllowedAndDenied pins the canonical happy paths:
+// an identity holding a capability gets allowed=true; the same
+// identity probed against a capability it lacks gets allowed=false.
+// Both produce 200 — "allowed=false" is a successful probe, not
+// an authorization failure.
+func TestHandleCan_AllowedAndDenied(t *testing.T) {
+	t.Parallel()
+
+	readWriteIdentity := httpauth.Identity{
+		ID:     "ops-rw",
+		Scopes: []httpauth.Scope{httpauth.ScopeRead, httpauth.ScopeWrite},
+	}
+
+	tests := []struct {
+		name        string
+		capability  string
+		wantAllowed bool
+	}{
+		{"read holder asks for read", capabilityCacheRead, true},
+		{"rw holder asks for write", capabilityCacheWrite, true},
+		{"rw holder asks for admin", capabilityCacheAdmin, false},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			body := callCanWithIdentity(t, readWriteIdentity, tc.capability)
+			if body.Capability != tc.capability {
+				t.Errorf("capability echo: got %q, want %q", body.Capability, tc.capability)
+			}
+
+			if body.Allowed != tc.wantAllowed {
+				t.Errorf("allowed: got %v, want %v", body.Allowed, tc.wantAllowed)
+			}
+		})
+	}
+}
+
+// TestHandleCan_MissingCapabilityParam pins the input-validation
+// posture: a request without the `capability` query param fails
+// 400 with the canonical error envelope. We don't silently
+// default to allowed=false — that would let typos pass as
+// "you can't do it" when the real issue is the missing argument.
+func TestHandleCan_MissingCapabilityParam(t *testing.T) {
+	t.Parallel()
+
+	app := fiber.New()
+	app.Use(func(c fiber.Ctx) error {
+		c.Locals(httpauth.IdentityKey, httpauth.Identity{ID: "x", Scopes: []httpauth.Scope{httpauth.ScopeRead}})
+
+		return c.Next()
+	})
+	app.Get("/v1/me/can", handleCan)
+
+	req := httptest.NewRequestWithContext(t.Context(), http.MethodGet, "/v1/me/can", strings.NewReader(""))
+
+	resp, err := app.Test(req)
+	if err != nil {
+		t.Fatalf("app.Test: %v", err)
+	}
+
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode != http.StatusBadRequest {
+		t.Fatalf("status: got %d, want 400", resp.StatusCode)
+	}
+}
+
+// TestHandleCan_UnknownCapability pins that unrecognized
+// capability strings fail 400, not silently allowed=false. A
+// typo like `cache.reaad` should surface as a client error so
+// the caller fixes their code rather than shipping a broken
+// authz check.
+func TestHandleCan_UnknownCapability(t *testing.T) {
+	t.Parallel()
+
+	app := fiber.New()
+	app.Use(func(c fiber.Ctx) error {
+		c.Locals(httpauth.IdentityKey, httpauth.Identity{
+			ID:     "x",
+			Scopes: []httpauth.Scope{httpauth.ScopeRead, httpauth.ScopeWrite, httpauth.ScopeAdmin},
+		})
+
+		return c.Next()
+	})
+	app.Get("/v1/me/can", handleCan)
+
+	req := httptest.NewRequestWithContext(t.Context(), http.MethodGet,
+		"/v1/me/can?capability=cache.reaad", strings.NewReader(""))
+
+	resp, err := app.Test(req)
+	if err != nil {
+		t.Fatalf("app.Test: %v", err)
+	}
+
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode != http.StatusBadRequest {
+		t.Fatalf("status: got %d, want 400 (unknown capability)", resp.StatusCode)
+	}
+}
+
+// callCanWithIdentity drives /v1/me/can with a pre-populated
+// IdentityKey local. Returns the decoded canResponse; failed
+// status / decode trips the test fatally.
+func callCanWithIdentity(t *testing.T, identity httpauth.Identity, capability string) canResponse {
+	t.Helper()
+
+	app := fiber.New()
+	app.Use(func(c fiber.Ctx) error {
+		c.Locals(httpauth.IdentityKey, identity)
+
+		return c.Next()
+	})
+	app.Get("/v1/me/can", handleCan)
+
+	req := httptest.NewRequestWithContext(t.Context(), http.MethodGet,
+		"/v1/me/can?capability="+capability, strings.NewReader(""))
+
+	resp, err := app.Test(req)
+	if err != nil {
+		t.Fatalf("app.Test: %v", err)
+	}
+
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode != http.StatusOK {
+		t.Fatalf("status: got %d, want 200", resp.StatusCode)
+	}
+
+	var got canResponse
+
+	err = json.NewDecoder(resp.Body).Decode(&got)
+	if err != nil {
+		t.Fatalf("decode body: %v", err)
+	}
+
+	return got
+}

cmd/hypercache-server/openapi.yaml

@@ -274,6 +274,43 @@ paths:
                 $ref: "#/components/schemas/IdentityResponse"
         "401": { $ref: "#/components/responses/Unauthorized" }
 
+  /v1/me/can:
+    get:
+      operationId: canPerform
+      tags: [ meta ]
+      summary: Per-capability authorization probe.
+      description: |
+        Returns whether the resolved identity holds the requested
+        capability. Cheaper than the speculative-write pattern
+        (try the action, catch the 403) and stable across future
+        scope-to-capability refactors — clients should key off the
+        capability string, not the internal scope shape.
+
+        Unknown capability values return 400 BAD_REQUEST so typos
+        don't silently answer "not allowed" when the real issue
+        is the caller's spelling.
+
+        Requires the `read` scope — same threshold as `/v1/me`.
+      parameters:
+        - in: query
+          name: capability
+          required: true
+          schema:
+            type: string
+            enum: [ cache.read, cache.write, cache.admin ]
+          description: |
+            The capability string to probe. Must be one of the
+            three values in the closed `cache.*` namespace.
+      responses:
+        "200":
+          description: Probe result.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/CanResponse"
+        "400": { $ref: "#/components/responses/BadRequest" }
+        "401": { $ref: "#/components/responses/Unauthorized" }
+
   /v1/cache/batch/get:
     post:
       operationId: batchGet
@@ -522,6 +559,19 @@ components:
           items:
             type: string
 
+    CanResponse:
+      type: object
+      required: [ capability, allowed ]
+      properties:
+        capability:
+          type: string
+          description: Echoes the queried capability string.
+        allowed:
+          type: boolean
+          description: |
+            True when the resolved identity holds the requested
+            capability; false otherwise.
+
     ItemEnvelope:
       type: object
       required: [ key, value, value_encoding, version, node, owners ]

cmd/hypercache-server/openapi_test.go

@@ -97,6 +97,7 @@ func declaredMethodsForPath() map[string]map[string]struct{} {
 		"/v1/cache/:key":         {fiber.MethodPut: {}, fiber.MethodGet: {}, fiber.MethodHead: {}, fiber.MethodDelete: {}},
 		"/v1/owners/:key":        {fiber.MethodGet: {}},
 		"/v1/me":                 {fiber.MethodGet: {}},
+		"/v1/me/can":             {fiber.MethodGet: {}},
 		"/v1/cache/batch/get":    {fiber.MethodPost: {}},
 		"/v1/cache/batch/put":    {fiber.MethodPost: {}},
 		"/v1/cache/batch/delete": {fiber.MethodPost: {}},

cspell.config.yaml

@@ -215,6 +215,7 @@ words:
   - pyenv
   - pygments
   - pymdownx
+  - reaad
   - recvcheck
   - rediscluster
   - Redocly