| 1 | // Tests for `carlos relay`'s client half: the wire shape the five apiSession |
| 2 | // methods send (matching internal/console/apirelay_test.go's server side), |
| 3 | // and the honesty rules printRelayStatus and printRelayRotate must follow. |
| 4 | // |
| 5 | // The printers get the most attention here, because `carlos relay status` is |
| 6 | // the ONLY sight a member with no box access has of whether their calls can |
| 7 | // connect. Every rule pinned below is one where a wrong line would leave them |
| 8 | // believing something the console did not say: that a route has the key when |
| 9 | // it is waiting on them, that a relay is healthy when it stopped reading five |
| 10 | // minutes ago, that a disable is finished when allocations are still live. |
| 11 | package main |
| 12 | |
| 13 | import ( |
| 14 | "bytes" |
| 15 | "context" |
| 16 | "encoding/json" |
| 17 | "net/http" |
| 18 | "net/http/httptest" |
| 19 | "strings" |
| 20 | "testing" |
| 21 | ) |
| 22 | |
| 23 | // ---- apiSession.relay* wire shape ------------------------------------------- |
| 24 | |
| 25 | // TestAPISessionRelayWireShapes pins the path and body of each of the four |
| 26 | // app-scoped calls plus rotate. rotate is the one with no account segment at |
| 27 | // all — a literal path, because there is nothing per-app about moving a |
| 28 | // region's issuer pointer. |
| 29 | func TestAPISessionRelayWireShapes(t *testing.T) { |
| 30 | var gotPath, gotMethod, gotAuth, gotBody string |
| 31 | ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 32 | gotPath, gotMethod, gotAuth = r.URL.RequestURI(), r.Method, r.Header.Get("Authorization") |
| 33 | body := new(bytes.Buffer) |
| 34 | body.ReadFrom(r.Body) |
| 35 | gotBody = strings.TrimSpace(body.String()) |
| 36 | json.NewEncoder(w).Encode(map[string]any{"app": "meet", "enabled": true}) |
| 37 | })) |
| 38 | defer ts.Close() |
| 39 | sess := newTestSession(ts.URL) |
| 40 | a := apiAcct{Seg: "meet-team"} |
| 41 | |
| 42 | if _, err := sess.relayEnable(context.Background(), a, "meet", nil); err != nil { |
| 43 | t.Fatalf("relayEnable: %v", err) |
| 44 | } |
| 45 | if gotMethod != http.MethodPost || gotPath != "/api/cli/apps/meet-team/meet/relay" { |
| 46 | t.Fatalf("enable = %s %s", gotMethod, gotPath) |
| 47 | } |
| 48 | // No regions named sends the literal {} the door documents as its |
| 49 | // default, never {"regions":null}. |
| 50 | if gotBody != "{}" { |
| 51 | t.Fatalf("enable body = %q, want {}", gotBody) |
| 52 | } |
| 53 | if gotAuth != "Bearer cli-token" { |
| 54 | t.Fatalf("auth = %q", gotAuth) |
| 55 | } |
| 56 | |
| 57 | if _, err := sess.relayEnable(context.Background(), a, "meet", []string{"eu-west-1"}); err != nil { |
| 58 | t.Fatalf("relayEnable(regions): %v", err) |
| 59 | } |
| 60 | if gotBody != `{"regions":["eu-west-1"]}` { |
| 61 | t.Fatalf("enable body with regions = %q", gotBody) |
| 62 | } |
| 63 | |
| 64 | if _, err := sess.relayGet(context.Background(), a, "meet"); err != nil { |
| 65 | t.Fatalf("relayGet: %v", err) |
| 66 | } |
| 67 | if gotMethod != http.MethodGet || gotPath != "/api/cli/apps/meet-team/meet/relay" { |
| 68 | t.Fatalf("status = %s %s", gotMethod, gotPath) |
| 69 | } |
| 70 | |
| 71 | if _, err := sess.relayDisable(context.Background(), a, "meet"); err != nil { |
| 72 | t.Fatalf("relayDisable: %v", err) |
| 73 | } |
| 74 | if gotMethod != http.MethodPost || gotPath != "/api/cli/apps/meet-team/meet/relay/disable" { |
| 75 | t.Fatalf("disable = %s %s", gotMethod, gotPath) |
| 76 | } |
| 77 | |
| 78 | if _, err := sess.relayAdopted(context.Background(), a, "meet", "x.meet.example"); err != nil { |
| 79 | t.Fatalf("relayAdopted: %v", err) |
| 80 | } |
| 81 | if gotPath != "/api/cli/apps/meet-team/meet/relay/adopted" { |
| 82 | t.Fatalf("adopted path = %s", gotPath) |
| 83 | } |
| 84 | if gotBody != `{"host":"x.meet.example"}` { |
| 85 | t.Fatalf("adopted body = %q", gotBody) |
| 86 | } |
| 87 | |
| 88 | if _, err := sess.relayRotate(context.Background(), "eu-west-1"); err != nil { |
| 89 | t.Fatalf("relayRotate: %v", err) |
| 90 | } |
| 91 | if gotPath != "/api/cli/relay/rotate" { |
| 92 | t.Fatalf("rotate path = %s, want the literal deployment-wide door", gotPath) |
| 93 | } |
| 94 | if gotBody != `{"region":"eu-west-1"}` { |
| 95 | t.Fatalf("rotate body = %q", gotBody) |
| 96 | } |
| 97 | } |
| 98 | |
| 99 | // TestAPISessionRelayRotateNotFoundNamesTheOperatorList: the rotate door |
| 100 | // answers a non-operator with the mux's own byte-identical 404, which is |
| 101 | // deliberate — a 403 would confirm to any token holder that this deployment |
| 102 | // runs a relay. So the CLI cannot tell "not allowed" from "no such door", and |
| 103 | // the message must say BOTH rather than fall back to the generic membership |
| 104 | // advice, which names --account and --app: two flags this verb does not have, |
| 105 | // sending an operator to check something they cannot even type. |
| 106 | func TestAPISessionRelayRotateNotFoundNamesTheOperatorList(t *testing.T) { |
| 107 | ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 108 | http.NotFound(w, r) |
| 109 | })) |
| 110 | defer ts.Close() |
| 111 | |
| 112 | _, err := newTestSession(ts.URL).relayRotate(context.Background(), "eu-west-1") |
| 113 | if err == nil { |
| 114 | t.Fatal("rotate against a 404 door returned no error") |
| 115 | } |
| 116 | if !strings.Contains(err.Error(), "CARLOS_RELAY_OPERATORS") { |
| 117 | t.Fatalf("error = %q, want the operator list named", err) |
| 118 | } |
| 119 | if strings.Contains(err.Error(), "--account") || strings.Contains(err.Error(), "--app") { |
| 120 | t.Fatalf("error = %q, want no advice about flags this verb does not have", err) |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | // TestAPISessionRelayServerDiagnosisSurvives: a 404 that DOES carry a JSON |
| 125 | // error is a real server sentence ("the relay has never been enabled...") and |
| 126 | // must reach the member verbatim, not be rewritten into the generic one. |
| 127 | func TestAPISessionRelayServerDiagnosisSurvives(t *testing.T) { |
| 128 | ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 129 | w.WriteHeader(http.StatusNotFound) |
| 130 | json.NewEncoder(w).Encode(map[string]any{"error": "the relay has never been enabled for this app"}) |
| 131 | })) |
| 132 | defer ts.Close() |
| 133 | |
| 134 | _, err := newTestSession(ts.URL).relayGet(context.Background(), apiAcct{Seg: "t"}, "meet") |
| 135 | if err == nil || !strings.Contains(err.Error(), "never been enabled") { |
| 136 | t.Fatalf("relayGet 404 = %v, want the console's own text", err) |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | // ---- printRelayStatus -------------------------------------------------------- |
| 141 | // |
| 142 | // Driven against literal replies, no stub server: the wording rules are the |
| 143 | // thing under test, and they must not depend on the wire plumbing above. |
| 144 | |
| 145 | func TestPrintRelayStatus(t *testing.T) { |
| 146 | healthy := apiRelayReply{ |
| 147 | App: "meet", Enabled: true, Incarnation: "0123456789abcdef", |
| 148 | Regions: []string{"eu-west-1"}, Op: 3, Phase: "done", |
| 149 | Issued: map[string]string{"eu-west-1": "abc:meet:0123456789abcdef:k1"}, |
| 150 | RegionStatus: []apiRelayRegion{{ |
| 151 | Region: "eu-west-1", Box: "relay-1", HeartbeatAge: "12s", Current: true, |
| 152 | LeaseLive: true, Listeners: []string{"udp:3478"}, Kids: []string{"k1"}, |
| 153 | }}, |
| 154 | } |
| 155 | |
| 156 | cases := []struct { |
| 157 | name string |
| 158 | res apiRelayReply |
| 159 | want []string |
| 160 | wantNot []string |
| 161 | modifier func(*apiRelayReply) |
| 162 | }{ |
| 163 | { |
| 164 | name: "healthy", |
| 165 | res: healthy, |
| 166 | want: []string{"relay for meet", "enabled in eu-west-1", "0123456789abcdef", |
| 167 | "op 3", "done", "issued eu-west-1: abc:meet:0123456789abcdef:k1", |
| 168 | "relay relay-1 in eu-west-1: heartbeat 12s", "udp:3478", "masters k1", "lease live"}, |
| 169 | // A healthy relay must never be described as stale or as holding |
| 170 | // an old record — the two words a member acts on. |
| 171 | wantNot: []string{"STALE", "older record", "disable pending"}, |
| 172 | }, |
| 173 | { |
| 174 | // The operator's mid-rotation question, and the only place the |
| 175 | // CLI can answer it: when may the old key leave this box's |
| 176 | // /etc/carlos/relay-masters (discussion #120). |
| 177 | name: "a retiring master names when it may be dropped", |
| 178 | res: healthy, |
| 179 | modifier: func(r *apiRelayReply) { |
| 180 | r.RegionStatus = []apiRelayRegion{{ |
| 181 | Region: "eu-west-1", Box: "relay-1", HeartbeatAge: "12s", Current: true, |
| 182 | LeaseLive: true, Listeners: []string{"udp:3478"}, Kids: []string{"k1", "k2"}, |
| 183 | Retiring: map[string]string{"k2": "2026-09-10T12:00:00Z", "k1": "2026-09-09T12:00:00Z"}, |
| 184 | }} |
| 185 | }, |
| 186 | // Two of them, deliberately seeded out of order: the kids are |
| 187 | // sorted, and the retiring clause uses the same tight separator |
| 188 | // the masters list does, so its items cannot be misread as |
| 189 | // further clauses of the line. |
| 190 | want: []string{"masters k1,k2", "retiring k1 after 2026-09-09T12:00:00Z,k2 after 2026-09-10T12:00:00Z"}, |
| 191 | }, |
| 192 | { |
| 193 | name: "stale names the relay's own reason", |
| 194 | res: healthy, |
| 195 | modifier: func(r *apiRelayReply) { |
| 196 | r.RegionStatus = []apiRelayRegion{{ |
| 197 | Region: "eu-west-1", Box: "relay-1", HeartbeatAge: "6m0s", Current: false, |
| 198 | LeaseLive: true, Stale: true, Kids: []string{"k1"}, |
| 199 | AuthorityError: "get control/relay.json: 503", |
| 200 | }} |
| 201 | }, |
| 202 | want: []string{"STALE", "holding an older record", "6m0s", "authority", "503"}, |
| 203 | // Still inside its fifteen-minute lease, so it IS still serving: |
| 204 | // saying otherwise would send someone hunting a dead box. |
| 205 | wantNot: []string{"no longer serving"}, |
| 206 | }, |
| 207 | { |
| 208 | name: "lease expired", |
| 209 | res: healthy, |
| 210 | modifier: func(r *apiRelayReply) { |
| 211 | r.RegionStatus = []apiRelayRegion{{ |
| 212 | Region: "eu-west-1", Box: "relay-1", HeartbeatAge: "40m0s", |
| 213 | Current: true, LeaseLive: false, Stale: false, |
| 214 | }} |
| 215 | }, |
| 216 | want: []string{"lease expired", "no longer serving"}, |
| 217 | }, |
| 218 | { |
| 219 | name: "no heartbeat at all", |
| 220 | res: healthy, |
| 221 | modifier: func(r *apiRelayReply) { |
| 222 | r.RegionStatus = nil |
| 223 | }, |
| 224 | want: []string{"no relay in eu-west-1 has heartbeated at all", "will not connect"}, |
| 225 | wantNot: []string{"lease live"}, |
| 226 | }, |
| 227 | { |
| 228 | name: "disable pending", |
| 229 | res: healthy, |
| 230 | modifier: func(r *apiRelayReply) { |
| 231 | r.Enabled, r.Op, r.Phase = false, 4, "done" |
| 232 | r.Issued = nil |
| 233 | r.DisablePending = []string{"eu-west-1"} |
| 234 | }, |
| 235 | // A disabled record still shows its operation and phase: a |
| 236 | // disable stuck at phase "config" is not the same state as one |
| 237 | // that finished, and "not enabled" alone would hide the |
| 238 | // difference. |
| 239 | want: []string{"disabled (op 4, done)", "disable pending on eu-west-1", |
| 240 | "next authority read, within a minute"}, |
| 241 | wantNot: []string{"revoked"}, |
| 242 | }, |
| 243 | { |
| 244 | name: "revoked", |
| 245 | res: healthy, |
| 246 | modifier: func(r *apiRelayReply) { |
| 247 | r.Enabled, r.Op = false, 4 |
| 248 | r.Issued = nil |
| 249 | }, |
| 250 | want: []string{"revoked: every relay has read a record without this app"}, |
| 251 | wantNot: []string{"disable pending"}, |
| 252 | }, |
| 253 | { |
| 254 | name: "never enabled", |
| 255 | res: apiRelayReply{App: "meet"}, |
| 256 | want: []string{"not enabled", "carlos relay enable --app meet"}, |
| 257 | // Nothing was ever issued, so there is no operation to report. |
| 258 | wantNot: []string{"op 0", "revoked"}, |
| 259 | }, |
| 260 | { |
| 261 | name: "route deliveries", |
| 262 | res: healthy, |
| 263 | modifier: func(r *apiRelayReply) { |
| 264 | r.Routes = []apiRelayRoute{ |
| 265 | {Host: "a.meet", Shape: "unit", Delivery: "written, awaiting delivery"}, |
| 266 | {Host: "b.meet", Shape: "socket-exec", Delivery: "written, awaiting delivery"}, |
| 267 | {Host: "c.meet", Shape: "tcp-exec-file", |
| 268 | Delivery: "file delivered, awaiting `carlos relay adopted --host c.meet` after you restart it"}, |
| 269 | {Host: "d.meet", Shape: "no-delivery", |
| 270 | Delivery: "no config delivery: relay will not reach this route; opt it into exec_delivery file or move it to a unit"}, |
| 271 | {Host: "e.meet", Shape: "tcp-exec-file", Cutover: true, |
| 272 | AckedAt: "2026-09-07T10:00:00Z", Delivery: "adopted by member at 2026-09-07T10:00:00Z"}, |
| 273 | // f, g, h: the three states Task 12's config-ack adds for a |
| 274 | // unit or socket-exec route (internal/console/apirelay.go's |
| 275 | // relayMatchingAck) — printRelayStatus computes none of this, |
| 276 | // it only prints whatever Delivery sentence it was handed, so |
| 277 | // these pin that the sentences flow through unmangled. |
| 278 | {Host: "f.meet", Shape: "unit", ConfigAckAt: "2026-09-07T10:05:00Z", |
| 279 | Delivery: "adopted at 2026-09-07T10:05:00Z, restarted"}, |
| 280 | {Host: "g.meet", Shape: "unit", |
| 281 | Delivery: "restart failed: systemctl try-restart carlos-app@g.meet: exit status 1"}, |
| 282 | {Host: "h.meet", Shape: "unit", Delivery: "hibernating, delivered on wake"}, |
| 283 | } |
| 284 | }, |
| 285 | want: []string{ |
| 286 | "route a.meet (unit): written, awaiting delivery", |
| 287 | "route b.meet (socket-exec): written", |
| 288 | "route c.meet (tcp-exec-file): file delivered, awaiting `carlos relay adopted --host c.meet`", |
| 289 | "route d.meet (no-delivery): no config delivery", |
| 290 | "route e.meet (tcp-exec-file): adopted by member at 2026-09-07T10:00:00Z", |
| 291 | "route f.meet (unit): adopted at 2026-09-07T10:05:00Z, restarted", |
| 292 | "route g.meet (unit): restart failed: systemctl try-restart carlos-app@g.meet: exit status 1", |
| 293 | "route h.meet (unit): hibernating, delivered on wake", |
| 294 | }, |
| 295 | // The lesson restart.go paid for: never say a box has acted. |
| 296 | wantNot: []string{"no routes yet"}, |
| 297 | }, |
| 298 | { |
| 299 | name: "enabled with no routes", |
| 300 | res: healthy, |
| 301 | want: []string{"no routes yet"}, |
| 302 | }, |
| 303 | } |
| 304 | |
| 305 | for _, tc := range cases { |
| 306 | t.Run(tc.name, func(t *testing.T) { |
| 307 | res := tc.res |
| 308 | if tc.modifier != nil { |
| 309 | tc.modifier(&res) |
| 310 | } |
| 311 | var buf bytes.Buffer |
| 312 | printRelayStatus(&buf, res) |
| 313 | out := buf.String() |
| 314 | for _, want := range tc.want { |
| 315 | if !strings.Contains(out, want) { |
| 316 | t.Fatalf("missing %q:\n%s", want, out) |
| 317 | } |
| 318 | } |
| 319 | for _, no := range tc.wantNot { |
| 320 | if strings.Contains(out, no) { |
| 321 | t.Fatalf("must not contain %q:\n%s", no, out) |
| 322 | } |
| 323 | } |
| 324 | }) |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | // TestPrintRelayStatusNeverClaimsDelivery is the output-honesty rule of this |
| 329 | // whole file in one assertion: given only "written, awaiting delivery" — no |
| 330 | // config-ack, the state before Task 12's per-route ack (internal/appconfig/ |
| 331 | // configack.go) exists for a route — printRelayStatus must not claim a |
| 332 | // running process has picked anything up. It is a pure function of |
| 333 | // res.Routes[i].Delivery either way: what earns the fuller "adopted at |
| 334 | // <time>, restarted" sentence is relayReply's own config-ack matching |
| 335 | // (internal/console/apirelay.go), not anything this function decides. |
| 336 | func TestPrintRelayStatusNeverClaimsDelivery(t *testing.T) { |
| 337 | var buf bytes.Buffer |
| 338 | printRelayStatus(&buf, apiRelayReply{ |
| 339 | App: "meet", Enabled: true, Incarnation: "0123456789abcdef", |
| 340 | Regions: []string{"eu-west-1"}, Phase: "done", |
| 341 | Routes: []apiRelayRoute{{Host: "a.meet", Shape: "unit", Delivery: "written, awaiting delivery"}}, |
| 342 | }) |
| 343 | out := strings.ToLower(buf.String()) |
| 344 | for _, mustNot := range []string{"is using", "now running the", "has the key", "picked it up"} { |
| 345 | if strings.Contains(out, mustNot) { |
| 346 | t.Fatalf("printRelayStatus over-claims delivery (%q):\n%s", mustNot, buf.String()) |
| 347 | } |
| 348 | } |
| 349 | } |
| 350 | |
| 351 | // ---- printRelayRotate -------------------------------------------------------- |
| 352 | |
| 353 | // TestPrintRelayRotate: the operator's answer has to carry the retirement |
| 354 | // deadline, because retiring the old master early kills credentials still |
| 355 | // inside their own four-hour lifetime — and nothing else in this system will |
| 356 | // remind them. |
| 357 | func TestPrintRelayRotate(t *testing.T) { |
| 358 | var buf bytes.Buffer |
| 359 | printRelayRotate(&buf, apiRelayRotateReply{ |
| 360 | Region: "eu-west-1", Kid: "k2", Apps: 7, |
| 361 | RetireNoEarlierThan: "2026-09-08T16:00:00Z", |
| 362 | }) |
| 363 | out := buf.String() |
| 364 | for _, want := range []string{"rotated eu-west-1", "kid k2", "7 app(s) re-issued", |
| 365 | "no earlier than 2026-09-08T16:00:00Z", "/etc/carlos/relay-masters"} { |
| 366 | if !strings.Contains(out, want) { |
| 367 | t.Fatalf("printRelayRotate missing %q:\n%s", want, out) |
| 368 | } |
| 369 | } |
| 370 | } |
| 371 | |
amadan