[MM-64244] Add websocket disconnect reason metric (#31032)

We've recently spent some effort improving websocket reconnection logic. With this commit, I've augmented the websocket reconnect metric to include a disconnect reason. This will help us measure the impact of these changes in production.

server/channels/api4/websocket.go

@@ -5,6 +5,7 @@ package api4
 
 import (
 	"net/http"
+	"strconv"
 
 	"github.com/gorilla/websocket"
 
@@ -15,11 +16,39 @@ import (
 )
 
 const (
-	connectionIDParam   = "connection_id"
-	sequenceNumberParam = "sequence_number"
-	postedAckParam      = "posted_ack"
+	connectionIDParam      = "connection_id"
+	sequenceNumberParam    = "sequence_number"
+	postedAckParam         = "posted_ack"
+	disconnectErrCodeParam = "disconnect_err_code"
+
+	clientPingTimeoutErrCode      = 4000
+	clientSequenceMismatchErrCode = 4001
 )
 
+// validateDisconnectErrCode ensures the specified disconnect error code
+// is a valid websocket close code
+func validateDisconnectErrCode(errCode string) bool {
+	if errCode == "" {
+		return false
+	}
+
+	// Ensure the disconnect code is a standard close code
+	code, err := strconv.Atoi(errCode)
+	if err != nil {
+		return false
+	}
+
+	// We only support the standard close codes between
+	// 1000 and 1016, and a few custom application codes
+	if (code < 1000 || code > 1016) &&
+		code != clientPingTimeoutErrCode &&
+		code != clientSequenceMismatchErrCode {
+		return false
+	}
+
+	return true
+}
+
 func (api *API) InitWebSocket() {
 	// Optionally supports a trailing slash
 	api.BaseRoutes.APIRoot.Handle("/{websocket:websocket(?:\\/)?}", api.APIHandlerTrustRequester(connectWebSocket)).Methods(http.MethodGet)
@@ -53,6 +82,12 @@ func connectWebSocket(c *Context, w http.ResponseWriter, r *http.Request) {
 		RemoteAddress: c.AppContext.IPAddress(),
 		XForwardedFor: c.AppContext.XForwardedFor(),
 	}
+
+	disconnectErrCode := r.URL.Query().Get(disconnectErrCodeParam)
+	if codeValid := validateDisconnectErrCode(disconnectErrCode); codeValid {
+		cfg.DisconnectErrCode = disconnectErrCode
+	}
+
 	// The WebSocket upgrade request coming from mobile is missing the
 	// user agent so we need to fallback on the session's metadata.
 	if c.AppContext.Session().IsMobileApp() {

server/channels/api4/websocket_test.go

@@ -465,3 +465,74 @@ func TestWebSocketUpgrade(t *testing.T) {
 	require.NoError(t, th.TestLogger.Flush())
 	testlib.AssertLog(t, buffer, mlog.LvlDebug.Name, "URL Blocked because of CORS. Url: ")
 }
+
+func TestValidateDisconnectErrCode(t *testing.T) {
+	testCases := []struct {
+		name    string
+		errCode string
+		valid   bool
+	}{
+		{
+			name:    "empty string",
+			errCode: "",
+			valid:   false,
+		},
+		{
+			name:    "non-numeric string",
+			errCode: "not-a-number",
+			valid:   false,
+		},
+		{
+			name:    "valid standard close code - 1000",
+			errCode: "1000",
+			valid:   true,
+		},
+		{
+			name:    "valid standard close code - 1001",
+			errCode: "1001",
+			valid:   true,
+		},
+		{
+			name:    "valid standard close code - 1015",
+			errCode: "1015",
+			valid:   true,
+		},
+		{
+			name:    "valid standard close code - 1016",
+			errCode: "1016",
+			valid:   true,
+		},
+		{
+			name:    "out of range (too low)",
+			errCode: "999",
+			valid:   false,
+		},
+		{
+			name:    "out of range (too high)",
+			errCode: "1017",
+			valid:   false,
+		},
+		{
+			name:    "valid custom code - client ping timeout",
+			errCode: "4000",
+			valid:   true,
+		},
+		{
+			name:    "valid custom code - client sequence mismatch",
+			errCode: "4001",
+			valid:   true,
+		},
+		{
+			name:    "invalid custom code",
+			errCode: "5000",
+			valid:   false,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			result := validateDisconnectErrCode(tc.errCode)
+			require.Equal(t, tc.valid, result)
+		})
+	}
+}