| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "strings" |
| 6 | "testing" |
| 7 | "time" |
| 8 | |
| 9 | "github.com/carlosframework/platform/internal/release" |
| 10 | ) |
| 11 | |
| 12 | func TestRefusalWatchAlertsOncePerRecord(t *testing.T) { |
| 13 | bake := t.TempDir() |
| 14 | now := time.Date(2026, 10, 6, 9, 31, 0, 0, time.UTC) |
| 15 | if err := release.WriteRefusal(bake, release.Refusal{Host: "console.example.com", App: "console", |
| 16 | Version: "v2", RefusedAt: now.Add(-time.Minute), Reason: "exited: status=2"}); err != nil { |
| 17 | t.Fatal(err) |
| 18 | } |
| 19 | var posts []string |
| 20 | sinks := newAlertSinks("https://hook.example", func(_ string, b []byte) error { posts = append(posts, string(b)); return nil }, "", nil, &alertCounters{}) |
| 21 | w := &refusalWatch{} |
| 22 | got := w.tick(context.Background(), bake, sinks, now) |
| 23 | w.tick(context.Background(), bake, sinks, now.Add(time.Minute)) |
| 24 | if len(posts) != 1 || !strings.Contains(posts[0], `"release_refused"`) || !strings.Contains(posts[0], "v2") { |
| 25 | t.Fatalf("posts = %v, want exactly one release_refused alert", posts) |
| 26 | } |
| 27 | if len(got) != 1 || got[0].Host != "console.example.com" { |
| 28 | t.Fatalf("status rows = %+v", got) |
| 29 | } |
| 30 | } |
| 31 | |
| 32 | func TestRefusalWatchSkipsOldRecordsAtStartup(t *testing.T) { |
| 33 | bake := t.TempDir() |
| 34 | now := time.Date(2026, 10, 6, 12, 0, 0, 0, time.UTC) |
| 35 | if err := release.WriteRefusal(bake, release.Refusal{Host: "a.example.com", Version: "v2", |
| 36 | RefusedAt: now.Add(-2 * time.Hour)}); err != nil { |
| 37 | t.Fatal(err) |
| 38 | } |
| 39 | var posts int |
| 40 | sinks := newAlertSinks("https://hook.example", func(string, []byte) error { posts++; return nil }, "", nil, &alertCounters{}) |
| 41 | (&refusalWatch{}).tick(context.Background(), bake, sinks, now) |
| 42 | if posts != 0 { |
| 43 | t.Fatalf("an agent restart re-alerted a refusal from two hours ago") |
| 44 | } |
| 45 | } |
| 46 | |
| 47 | // Final review F6 and ledger Task 8: SNS and the webhook tell a recovered |
| 48 | // rollback from one whose restored build did not come up either, and the |
| 49 | // status row carries the failure for `status --live`. |
| 50 | func TestRefusalWatchReportsTheRecoveryOutcome(t *testing.T) { |
| 51 | now := time.Date(2026, 10, 6, 9, 31, 0, 0, time.UTC) |
| 52 | for _, c := range []struct { |
| 53 | name string |
| 54 | rec release.Refusal |
| 55 | subject string |
| 56 | inMessage string |
| 57 | inBody string |
| 58 | rowErr string |
| 59 | }{ |
| 60 | {"recovered", release.Refusal{RestoredVersion: "v1", Recovered: true}, |
| 61 | "console.example.com: v2 did not start, rolled back to v1", "Retry with: carlos promote --retry v2", |
| 62 | `"recovered":true`, ""}, |
| 63 | {"failed recovery", release.Refusal{RestoredVersion: "v1", RecoveryError: "start console.service: exit status 1"}, |
| 64 | "console.example.com: v2 did not start and the rollback did not come up either", |
| 65 | "The rollback did not come up either: start console.service: exit status 1", |
| 66 | `"recovery_error":"start console.service: exit status 1"`, "start console.service: exit status 1"}, |
| 67 | } { |
| 68 | t.Run(c.name, func(t *testing.T) { |
| 69 | bake := t.TempDir() |
| 70 | r := c.rec |
| 71 | r.Host, r.App, r.Version, r.RefusedAt, r.Reason = "console.example.com", "console", "v2", now.Add(-time.Minute), "exited: status=2" |
| 72 | if err := release.WriteRefusal(bake, r); err != nil { |
| 73 | t.Fatal(err) |
| 74 | } |
| 75 | var posts []string |
| 76 | pub := &fakeSNS{} |
| 77 | sinks := newAlertSinks("https://hook.example", func(_ string, b []byte) error { posts = append(posts, string(b)); return nil }, |
| 78 | testTopicARN, pub, &alertCounters{}) |
| 79 | rows := (&refusalWatch{}).tick(context.Background(), bake, sinks, now) |
| 80 | if len(pub.in) != 1 || *pub.in[0].Subject != c.subject { |
| 81 | t.Fatalf("sns = %+v, want one publish with subject %q", pub.in, c.subject) |
| 82 | } |
| 83 | if !strings.Contains(*pub.in[0].Message, c.inMessage) { |
| 84 | t.Fatalf("sns message = %q, want %q in it", *pub.in[0].Message, c.inMessage) |
| 85 | } |
| 86 | if len(posts) != 1 || !strings.Contains(posts[0], c.inBody) { |
| 87 | t.Fatalf("webhook = %v, want %s in the body", posts, c.inBody) |
| 88 | } |
| 89 | if len(rows) != 1 || rows[0].RecoveryError != c.rowErr { |
| 90 | t.Fatalf("status rows = %+v, want recovery_error %q", rows, c.rowErr) |
| 91 | } |
| 92 | }) |
| 93 | } |
| 94 | } |
| 95 | |
| 96 | // D2's refusal is not a rollback: the subject says so and the message points |
| 97 | // at a NEW version shipped with carlos ship --program-change (the refused |
| 98 | // version's manifest is immutable), never --retry. The webhook keeps its envelope kind and |
| 99 | // carries the record's as refusal_kind; the status row carries it too. |
| 100 | func TestRefusalWatchReportsAWrongProgram(t *testing.T) { |
| 101 | bake := t.TempDir() |
| 102 | now := time.Date(2026, 10, 6, 9, 31, 0, 0, time.UTC) |
| 103 | rec := release.Refusal{Host: "console.example.com", App: "console", Version: "v2", RefusedAt: now.Add(-time.Minute), |
| 104 | Reason: "wrong program: example.com/carlos", Kind: release.RefusalWrongProgram, Recovered: true, RestoredVersion: "v1"} |
| 105 | if err := release.WriteRefusal(bake, rec); err != nil { |
| 106 | t.Fatal(err) |
| 107 | } |
| 108 | subject, msg := formatReleaseAlert(rec) |
| 109 | if subject != "console.example.com: v2 refused, wrong program (still on v1)" { |
| 110 | t.Fatalf("subject = %q", subject) |
| 111 | } |
| 112 | if !strings.Contains(msg, "Ship a new version with carlos ship --program-change and promote it, if the change is intended") || strings.Contains(msg, "--retry") { |
| 113 | t.Fatalf("message = %q", msg) |
| 114 | } |
| 115 | var posts []string |
| 116 | sinks := newAlertSinks("https://hook.example", func(_ string, b []byte) error { posts = append(posts, string(b)); return nil }, "", nil, &alertCounters{}) |
| 117 | rows := (&refusalWatch{}).tick(context.Background(), bake, sinks, now) |
| 118 | if len(posts) != 1 || !strings.Contains(posts[0], `"kind":"release_refused"`) || !strings.Contains(posts[0], `"refusal_kind":"wrong-program"`) { |
| 119 | t.Fatalf("posts = %v", posts) |
| 120 | } |
| 121 | if len(rows) != 1 || rows[0].Kind != release.RefusalWrongProgram { |
| 122 | t.Fatalf("rows = %+v", rows) |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | // D3's staged-bytes refusal is not a rollback either (plan 3 ruling 3): the |
| 127 | // box never left its old build. The alert says the staged bytes do not match |
| 128 | // dev's proof, names the build the box is still on, and asks for the exact |
| 129 | // bytes dev proved; never --retry, never a rollback. |
| 130 | func TestRefusalWatchReportsANotProven(t *testing.T) { |
| 131 | bake := t.TempDir() |
| 132 | now := time.Date(2026, 10, 6, 9, 31, 0, 0, time.UTC) |
| 133 | rec := release.Refusal{Host: "console.example.com", App: "console", Version: "v2", RefusedAt: now.Add(-time.Minute), |
| 134 | Reason: "not proven: staged sha256 aaaa, dev proved bbbb", Kind: release.RefusalNotProven, Recovered: true, RestoredVersion: "v1"} |
| 135 | if err := release.WriteRefusal(bake, rec); err != nil { |
| 136 | t.Fatal(err) |
| 137 | } |
| 138 | subject, msg := formatReleaseAlert(rec) |
| 139 | if subject != "console.example.com: v2 refused, staged bytes do not match dev's proof (still on v1)" { |
| 140 | t.Fatalf("subject = %q", subject) |
| 141 | } |
| 142 | for _, want := range []string{"staged sha256 aaaa, dev proved bbbb", "still on v1", "Roll the exact bytes dev proved"} { |
| 143 | if !strings.Contains(msg, want) { |
| 144 | t.Fatalf("message = %q, want %q in it", msg, want) |
| 145 | } |
| 146 | } |
| 147 | for _, never := range []string{"--retry", "rolled back", "rollback"} { |
| 148 | if strings.Contains(msg, never) || strings.Contains(subject, never) { |
| 149 | t.Fatalf("subject %q, message %q: %q has no place in a not-proven alert", subject, msg, never) |
| 150 | } |
| 151 | } |
| 152 | var posts []string |
| 153 | sinks := newAlertSinks("https://hook.example", func(_ string, b []byte) error { posts = append(posts, string(b)); return nil }, "", nil, &alertCounters{}) |
| 154 | rows := (&refusalWatch{}).tick(context.Background(), bake, sinks, now) |
| 155 | if len(posts) != 1 || !strings.Contains(posts[0], `"refusal_kind":"not-proven"`) { |
| 156 | t.Fatalf("posts = %v", posts) |
| 157 | } |
| 158 | if len(rows) != 1 || rows[0].Kind != release.RefusalNotProven || rows[0].RecoveryError != "" { |
| 159 | t.Fatalf("rows = %+v", rows) |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | // Final review finding 3: held refusals moved to their own namespace, and |
| 164 | // status and alerts still surface them as they did when they shared |
| 165 | // adoption's directory: one row each, beside an adoption refusal for the |
| 166 | // same host, and one alert each, keyed apart from adoption's. |
| 167 | func TestRefusalWatchReadsHeldRefusals(t *testing.T) { |
| 168 | bake := t.TempDir() |
| 169 | now := time.Date(2026, 10, 7, 9, 0, 0, 0, time.UTC) |
| 170 | host := "held.example.com" |
| 171 | if err := release.WriteRefusal(bake, release.Refusal{Host: host, App: "console", Version: "v1", |
| 172 | RefusedAt: now.Add(-time.Minute), Reason: "exited: status=2"}); err != nil { |
| 173 | t.Fatal(err) |
| 174 | } |
| 175 | if err := release.WriteRefusal(release.HeldRefusalBase(bake), release.Refusal{Host: host, App: "console", Version: "v2", |
| 176 | RefusedAt: now.Add(-2 * time.Minute), Reason: "not proven: staged sha256 bb", Kind: release.RefusalNotProven, |
| 177 | Recovered: true, MoveOp: "mv_abcdefghijklmnopqrstuvwxyz"}); err != nil { |
| 178 | t.Fatal(err) |
| 179 | } |
| 180 | var posts []string |
| 181 | sinks := newAlertSinks("https://hook.example", func(_ string, b []byte) error { posts = append(posts, string(b)); return nil }, "", nil, &alertCounters{}) |
| 182 | w := &refusalWatch{} |
| 183 | rows := w.tick(context.Background(), bake, sinks, now) |
| 184 | w.tick(context.Background(), bake, sinks, now.Add(time.Minute)) |
| 185 | var versions []string |
| 186 | for _, r := range rows { |
| 187 | versions = append(versions, r.Host+"@"+r.Version) |
| 188 | } |
| 189 | if strings.Join(versions, ",") != host+"@v1,"+host+"@v2" { |
| 190 | t.Fatalf("status rows %v, want the adoption refusal and the held one", versions) |
| 191 | } |
| 192 | if len(posts) != 2 { |
| 193 | t.Fatalf("%d alerts, want one per record: %v", len(posts), posts) |
| 194 | } |
| 195 | } |
| 196 | |
amadan