| 1 | package edge |
| 2 | |
| 3 | // The acceptance test issue #53 names: edge + fake remote host over a REAL |
| 4 | // reverse channel on loopback. Round-trip, header survival, kill-mid-flight, |
| 5 | // re-dial, and the reconnect bound (asserted, not narrated). |
| 6 | // |
| 7 | // The fake remote host is not a hand-rolled dialer: it runs |
| 8 | // internal/tunnelclient.Loop, the exact loop `carlos agent` runs on a Pi |
| 9 | // (internal/host/tunnel_client.go is a thin env-reading wrapper around it). |
| 10 | // That is the whole point of the extraction — a reconnect bound measured |
| 11 | // against a test double would prove nothing about the fleet. |
| 12 | |
| 13 | import ( |
| 14 | "context" |
| 15 | "fmt" |
| 16 | "io" |
| 17 | "net/http" |
| 18 | "net/http/httptest" |
| 19 | "net/url" |
| 20 | "strings" |
| 21 | "sync" |
| 22 | "testing" |
| 23 | "time" |
| 24 | |
| 25 | "github.com/carlosframework/platform/internal/registry" |
| 26 | "github.com/carlosframework/platform/internal/tunnelclient" |
| 27 | ) |
| 28 | |
| 29 | // e2eReconnectBound is the promise: a channel that dies is back and serving |
| 30 | // within this. The client sleeps one jittered backoff step (250-375ms on a |
| 31 | // first loss, 500-750ms once it has doubled) before re-dialing, so the real |
| 32 | // figure is well under a second — the bound is deliberately loose enough that |
| 33 | // a loaded CI box cannot flake it, and tight enough that a regression to the |
| 34 | // 2s cap compounding, or to no reconnect at all, fails here. |
| 35 | const e2eReconnectBound = 5 * time.Second |
| 36 | |
| 37 | func TestTunnelEndToEnd(t *testing.T) { |
| 38 | reg := tunnelRoute(t, registry.Route{ |
| 39 | Host: "demo.example", Kind: "instance", Status: "ready", Tunnel: "pi-box", |
| 40 | }) |
| 41 | e, err := New(reg, nil) |
| 42 | if err != nil { |
| 43 | t.Fatal(err) |
| 44 | } |
| 45 | // The edge listens on loopback; both app traffic and the tunnel |
| 46 | // registration ride this ONE listener — the no-new-port property. |
| 47 | srv := httptest.NewServer(e) |
| 48 | defer srv.Close() |
| 49 | |
| 50 | // The hub's reserved hostname has to be the name the host actually dials, |
| 51 | // because the dialer sends the URL's host as its Host header and the edge |
| 52 | // routes registration on that. On a loopback edge that name is the |
| 53 | // listener's own address; in production it is the operator's |
| 54 | // CARLOS_TUNNEL_HOST, resolved by DNS to the same 443. |
| 55 | u, err := url.Parse(srv.URL) |
| 56 | if err != nil { |
| 57 | t.Fatal(err) |
| 58 | } |
| 59 | hub := NewTunnelHub(u.Hostname(), map[string]string{"pi-box": "sekrit"}) |
| 60 | e.EnableTunnels(hub) |
| 61 | if err := e.Reload(); err != nil { |
| 62 | t.Fatal(err) |
| 63 | } |
| 64 | |
| 65 | // Fake remote host: its "local edge" answers with a header + body, and |
| 66 | // echoes Host and X-Forwarded-For so the test can prove header discipline |
| 67 | // held all the way across the channel. /slow parks until released — that |
| 68 | // is what makes the kill below genuinely mid-flight rather than a sleep |
| 69 | // and a hope. |
| 70 | inflight := make(chan struct{}, 1) |
| 71 | release := make(chan struct{}) |
| 72 | defer close(release) |
| 73 | appHandler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 74 | if r.URL.Path == "/slow" { |
| 75 | select { |
| 76 | case inflight <- struct{}{}: |
| 77 | default: |
| 78 | } |
| 79 | // Either the test lets it finish, or the stream dies under it — |
| 80 | // which is what a real handler sees when the channel drops, and |
| 81 | // what keeps this goroutine from outliving the connection. |
| 82 | select { |
| 83 | case <-release: |
| 84 | case <-r.Context().Done(): |
| 85 | } |
| 86 | } |
| 87 | w.Header().Set("X-App-Header", "survived") |
| 88 | w.Header().Set("Carlos-Purge", "/should-be-stripped") // edge must strip |
| 89 | fmt.Fprintf(w, "host=%s xff=%s", r.Host, r.Header.Get("X-Forwarded-For")) |
| 90 | }) |
| 91 | stop := startFakeRemoteHost(t, srv.URL, "pi-box", "sekrit", appHandler) |
| 92 | defer stop() |
| 93 | |
| 94 | waitForTunnelState(t, hub, "pi-box", "connected", e2eReconnectBound) |
| 95 | |
| 96 | do := func(path string) (*http.Response, string) { |
| 97 | t.Helper() |
| 98 | req, err := http.NewRequest("GET", srv.URL+path, nil) |
| 99 | if err != nil { |
| 100 | t.Fatal(err) |
| 101 | } |
| 102 | req.Host = "demo.example" |
| 103 | resp, err := http.DefaultClient.Do(req) |
| 104 | if err != nil { |
| 105 | t.Fatalf("request through tunnel: %v", err) |
| 106 | } |
| 107 | body, _ := io.ReadAll(resp.Body) |
| 108 | resp.Body.Close() |
| 109 | return resp, string(body) |
| 110 | } |
| 111 | |
| 112 | resp, body := do("/") |
| 113 | if resp.StatusCode != 200 || resp.Header.Get("X-App-Header") != "survived" { |
| 114 | t.Fatalf("round-trip: %d %v", resp.StatusCode, resp.Header) |
| 115 | } |
| 116 | if resp.Header.Get("Carlos-Purge") != "" { |
| 117 | t.Fatal("Carlos-Purge leaked to the client — purge discipline broken over tunnel") |
| 118 | } |
| 119 | if !strings.Contains(body, "host=demo.example") || !strings.Contains(body, "xff=127.") { |
| 120 | t.Fatalf("host/XFF discipline broken: %q", body) |
| 121 | } |
| 122 | |
| 123 | // Kill the channel out from under the host with a request PARKED inside |
| 124 | // the remote handler, then assert three things: the in-flight request |
| 125 | // fails fast rather than hanging, the host re-dials, and the next request |
| 126 | // succeeds — all inside the bound. |
| 127 | slowDone := make(chan int, 1) |
| 128 | go func() { |
| 129 | req, err := http.NewRequest("GET", srv.URL+"/slow", nil) |
| 130 | if err != nil { |
| 131 | slowDone <- -1 |
| 132 | return |
| 133 | } |
| 134 | req.Host = "demo.example" |
| 135 | resp, err := http.DefaultClient.Do(req) |
| 136 | if err != nil { |
| 137 | slowDone <- 0 // the client saw the failure directly |
| 138 | return |
| 139 | } |
| 140 | resp.Body.Close() |
| 141 | slowDone <- resp.StatusCode |
| 142 | }() |
| 143 | select { |
| 144 | case <-inflight: |
| 145 | case <-time.After(e2eReconnectBound): |
| 146 | t.Fatal("/slow never reached the remote host's handler") |
| 147 | } |
| 148 | |
| 149 | // Unregister BEFORE closing, so the ordering is deterministic rather than |
| 150 | // a race with the host's re-registration: once the entry is gone, State is |
| 151 | // "" until the NEW channel lands, and a re-register can never be clobbered |
| 152 | // by this teardown. |
| 153 | hub.mu.Lock() |
| 154 | dead := hub.conns["pi-box"] |
| 155 | delete(hub.conns, "pi-box") |
| 156 | hub.mu.Unlock() |
| 157 | if dead == nil { |
| 158 | t.Fatal("no live channel to kill — the hub lost it before the test did") |
| 159 | } |
| 160 | start := time.Now() |
| 161 | dead.raw.Close() |
| 162 | |
| 163 | select { |
| 164 | case code := <-slowDone: |
| 165 | if code < 0 { |
| 166 | t.Fatal("could not build the in-flight request") |
| 167 | } |
| 168 | if code == http.StatusOK { |
| 169 | t.Fatal("the in-flight request succeeded after its channel died — the edge served bytes over a dead channel") |
| 170 | } |
| 171 | case <-time.After(e2eReconnectBound): |
| 172 | t.Fatal("in-flight request neither completed nor failed after its channel died — it hung") |
| 173 | } |
| 174 | |
| 175 | waitForTunnelState(t, hub, "pi-box", "connected", e2eReconnectBound) |
| 176 | if resp, _ := do("/"); resp.StatusCode != 200 { |
| 177 | t.Fatalf("post-reconnect request: %d", resp.StatusCode) |
| 178 | } |
| 179 | if d := time.Since(start); d > e2eReconnectBound { |
| 180 | t.Fatalf("reconnect took %s, bound is %s", d, e2eReconnectBound) |
| 181 | } else { |
| 182 | t.Logf("reconnected and served in %s (bound %s)", d, e2eReconnectBound) |
| 183 | } |
| 184 | } |
| 185 | |
| 186 | // startFakeRemoteHost runs the production host-side loop against edgeURL and |
| 187 | // returns a stop func that cancels it and WAITS for it to exit — a loop still |
| 188 | // dialing after its test returned would log into the next test's output and |
| 189 | // hold a goroutine for the rest of the run. |
| 190 | func startFakeRemoteHost(t *testing.T, edgeURL, name, token string, handler http.Handler) func() { |
| 191 | t.Helper() |
| 192 | ctx, cancel := context.WithCancel(context.Background()) |
| 193 | var wg sync.WaitGroup |
| 194 | wg.Add(1) |
| 195 | go func() { |
| 196 | defer wg.Done() |
| 197 | tunnelclient.Loop(ctx, tunnelclient.Config{ |
| 198 | URL: edgeURL, Name: name, Token: token, Handler: handler, |
| 199 | Logf: func(format string, args ...any) { t.Logf("fake host: "+format, args...) }, |
| 200 | }) |
| 201 | }() |
| 202 | return func() { |
| 203 | cancel() |
| 204 | wg.Wait() // no t.Logf after the test returns |
| 205 | } |
| 206 | } |
| 207 | |
| 208 | func waitForTunnelState(t *testing.T, hub *TunnelHub, name, want string, within time.Duration) { |
| 209 | t.Helper() |
| 210 | deadline := time.Now().Add(within) |
| 211 | for time.Now().Before(deadline) { |
| 212 | if hub.State(name) == want { |
| 213 | return |
| 214 | } |
| 215 | time.Sleep(20 * time.Millisecond) |
| 216 | } |
| 217 | t.Fatalf("tunnel %s never reached state %q within %s", name, want, within) |
| 218 | } |
| 219 | |
amadan