Merge pull request #21 from axoflow/workaround-escaping-bug

syslog-ng-ctl: workaround invalid escaping in syslog-ng/AxoSyslog metrics

Author: MrAnno

pkg/syslog-ng-ctl/stats_prometheus.go

@@ -154,6 +154,38 @@ func transformEventDelayMetric(delayMetric *io_prometheus_client.MetricFamily, d
 	delayMetric.Type = io_prometheus_client.MetricType_GAUGE.Enum()
 }
 
+// Workaround for a bug in older syslog-ng/AxoSyslog versions where the output of STATS PROMETHEUS was overescaped.
+// Escapes \ as \\ everywhere except for the allowed sequences: \\, \n, \"
+func sanitizeBuggyFormat(output string) string {
+	var fixedOutput strings.Builder
+
+	length := len(output)
+	for i := 0; i < length; i++ {
+		c := output[i]
+
+		if c != '\\' {
+			fixedOutput.WriteByte(c)
+			continue
+		}
+
+		if i+1 >= length {
+			fixedOutput.WriteString(`\\`)
+			break
+		}
+
+		if next := output[i+1]; next == '\\' || next == 'n' || next == '"' {
+			fixedOutput.WriteByte(c)
+			fixedOutput.WriteByte(next)
+			i++
+			continue
+		}
+
+		fixedOutput.WriteString(`\\`)
+	}
+
+	return fixedOutput.String()
+}
+
 func StatsPrometheus(ctx context.Context, cc ControlChannel, lastMetricQueryTime *time.Time) ([]*io_prometheus_client.MetricFamily, error) {
 	rsp, err := cc.SendCommand(ctx, "STATS PROMETHEUS")
 	if err != nil {
@@ -169,6 +201,7 @@ func StatsPrometheus(ctx context.Context, cc ControlChannel, lastMetricQueryTime
 		return maps.Values(mfs), err
 	}
 
+	rsp = sanitizeBuggyFormat(rsp)
 	mfs, err = new(expfmt.TextParser).TextToMetricFamilies(strings.NewReader(rsp))
 
 	var delayMetric *io_prometheus_client.MetricFamily

pkg/syslog-ng-ctl/stats_prometheus_test.go

@@ -297,6 +297,34 @@ func TestStatsPrometheus(t *testing.T) {
 	}
 	sortMetricFamilies(expectedDelayMetrics)
 
+	expectedEscapeMetrics := []*io_prometheus_client.MetricFamily{
+		{
+			Name: amp("syslogng_classified_output_events_total"),
+			Type: io_prometheus_client.MetricType_COUNTER.Enum(),
+			Metric: []*io_prometheus_client.Metric{
+				{
+					Label: []*io_prometheus_client.LabelPair{
+						newLabel("app", "MSWinEventLog\\t1\\tSecurity\\t921448325\\tFri"),
+						newLabel("source", "s_critical_hosts_515"),
+					},
+					Counter: &io_prometheus_client.Counter{
+						Value: amp(1.0),
+					},
+				},
+				{
+					Label: []*io_prometheus_client.LabelPair{
+						newLabel("app", "\\a\\t\n\"\\xfa\\"),
+						newLabel("source", "s_unescaped_bug"),
+					},
+					Counter: &io_prometheus_client.Counter{
+						Value: amp(1.0),
+					},
+				},
+			},
+		},
+	}
+	sortMetricFamilies(expectedEscapeMetrics)
+
 	testCases := map[string]struct {
 		cc       ControlChannel
 		expected []*io_prometheus_client.MetricFamily
@@ -322,6 +350,13 @@ func TestStatsPrometheus(t *testing.T) {
 			}),
 			expected: expectedDelayMetrics,
 		},
+		"syslog-ng stats prometheus label escaping": {
+			cc: ControlChannelFunc(func(_ context.Context, cmd string) (rsp string, err error) {
+				require.Equal(t, "STATS PROMETHEUS", cmd)
+				return PROMETHEUS_ESCAPE_METRICS_OUTPUT, nil
+			}),
+			expected: expectedEscapeMetrics,
+		},
 	}
 
 	for name, testCase := range testCases {
@@ -460,6 +495,10 @@ syslogng_output_event_delay_sample_age_seconds{driver="http",url="http://localho
 syslogng_output_event_delay_sample_age_seconds{transport="tcp",address="localhost:5555",driver="afsocket",id="#anon-destination0#0"} 31
 `
 
+const PROMETHEUS_ESCAPE_METRICS_OUTPUT = `syslogng_classified_output_events_total{app="MSWinEventLog\\t1\\tSecurity\\t921448325\\tFri",source="s_critical_hosts_515"} 1
+syslogng_classified_output_events_total{app="\a\t\n\"\xfa\\",source="s_unescaped_bug"} 1
+`
+
 func metricFamiliesToText(mfs []*io_prometheus_client.MetricFamily) string {
 	var buf strings.Builder
 	for _, mf := range mfs {