Merge pull request #146 from ymtdzzz/feature/mark_error_spans

Mark the error spans in traces and timeline page

tuiexporter/internal/telemetry/cache.go

@@ -1,5 +1,7 @@
 package telemetry
 
+import "go.opentelemetry.io/collector/pdata/ptrace"
+
 // SpanDataMap is a map of span id to span data
 // This is used to quickly look up a span by its id
 type SpanDataMap map[string]*SpanData
@@ -12,37 +14,49 @@ type TraceSpanDataMap map[string][]*SpanData
 // This is used to quickly look up all spans in a trace for a service
 type TraceServiceSpanDataMap map[string]map[string][]*SpanData
 
+// TraceServiceHasErrorMap is a map of trace id and service name to a flag whether
+// the spans have any error status
+type TraceServiceHasErrorMap map[string]map[string]bool
+
 // TraceCache is a cache of trace spans
 type TraceCache struct {
-	spanid2span    SpanDataMap
-	traceid2spans  TraceSpanDataMap
-	tracesvc2spans TraceServiceSpanDataMap
+	spanid2span       SpanDataMap
+	traceid2spans     TraceSpanDataMap
+	tracesvc2spans    TraceServiceSpanDataMap
+	tracesvc2haserror TraceServiceHasErrorMap
 }
 
 // NewTraceCache returns a new trace cache
 func NewTraceCache() *TraceCache {
 	return &TraceCache{
-		spanid2span:    SpanDataMap{},
-		traceid2spans:  TraceSpanDataMap{},
-		tracesvc2spans: TraceServiceSpanDataMap{},
+		spanid2span:       SpanDataMap{},
+		traceid2spans:     TraceSpanDataMap{},
+		tracesvc2spans:    TraceServiceSpanDataMap{},
+		tracesvc2haserror: TraceServiceHasErrorMap{},
 	}
 }
 
 // UpdateCache updates the cache with a new span
 func (c *TraceCache) UpdateCache(sname string, data *SpanData) (newtracesvc bool) {
 	c.spanid2span[data.Span.SpanID().String()] = data
 	traceID := data.Span.TraceID().String()
+	hasError := spanHasError(data.Span)
 	if ts, ok := c.traceid2spans[traceID]; ok {
 		c.traceid2spans[traceID] = append(ts, data)
 		if _, ok := c.tracesvc2spans[traceID][sname]; ok {
 			c.tracesvc2spans[traceID][sname] = append(c.tracesvc2spans[traceID][sname], data)
+			if hasError {
+				c.tracesvc2haserror[traceID][sname] = hasError
+			}
 		} else {
 			c.tracesvc2spans[traceID][sname] = []*SpanData{data}
+			c.tracesvc2haserror[traceID][sname] = hasError
 			newtracesvc = true
 		}
 	} else {
 		c.traceid2spans[traceID] = []*SpanData{data}
 		c.tracesvc2spans[traceID] = map[string][]*SpanData{sname: {data}}
+		c.tracesvc2haserror[traceID] = map[string]bool{sname: hasError}
 		newtracesvc = true
 	}
 
@@ -62,8 +76,10 @@ func (c *TraceCache) DeleteCache(serviceSpans []*SpanData) {
 			}
 		}
 		delete(c.tracesvc2spans[traceID], sname.AsString())
+		delete(c.tracesvc2haserror[traceID], sname.AsString())
 		if len(c.tracesvc2spans[traceID]) == 0 {
 			delete(c.tracesvc2spans, traceID)
+			delete(c.tracesvc2haserror, traceID)
 			// delete spans in traceid2spans only if there are no spans left in tracesvc2spans
 			// for better performance
 			delete(c.traceid2spans, traceID)
@@ -87,6 +103,16 @@ func (c *TraceCache) GetSpansByTraceIDAndSvc(traceID, svc string) ([]*SpanData,
 	return nil, false
 }
 
+// HasErrorByTraceIDAndSvc returns the flag whether the spans have any errors
+func (c *TraceCache) HasErrorByTraceIDAndSvc(traceID, svc string) (bool, bool) {
+	if spans, ok := c.tracesvc2haserror[traceID]; ok {
+		if haserr, ok := spans[svc]; ok {
+			return haserr, ok
+		}
+	}
+	return false, false
+}
+
 // GetSpanByID returns a span by its id
 func (c *TraceCache) GetSpanByID(spanID string) (*SpanData, bool) {
 	span, ok := c.spanid2span[spanID]
@@ -99,6 +125,10 @@ func (c *TraceCache) flush() {
 	c.tracesvc2spans = TraceServiceSpanDataMap{}
 }
 
+func spanHasError(span *ptrace.Span) bool {
+	return span.Status().Code() == ptrace.StatusCodeError
+}
+
 // TraceLogDataMap is a map of trace id to a slice of logs
 // This is used to quickly look up all logs in a trace
 type TraceLogDataMap map[string][]*LogData

tuiexporter/internal/tui/component/page.go

@@ -152,7 +152,7 @@ func (p *TUIPages) createTracePage(store *telemetry.Store) *tview.Flex {
 	table := tview.NewTable().
 		SetBorders(false).
 		SetSelectable(true, false).
-		SetContent(NewSpanDataForTable(store.GetFilteredSvcSpans())).
+		SetContent(NewSpanDataForTable(store.GetTraceCache(), store.GetFilteredSvcSpans())).
 		SetSelectedFunc(func(row, _ int) {
 			p.showTimelineByRow(store, row-1)
 		}).

tuiexporter/internal/tui/component/timeline.go

@@ -9,6 +9,7 @@ import (
 	"github.com/gdamore/tcell/v2"
 	"github.com/rivo/tview"
 	"github.com/ymtdzzz/otel-tui/tuiexporter/internal/telemetry"
+	"go.opentelemetry.io/collector/pdata/ptrace"
 )
 
 const (
@@ -372,7 +373,11 @@ func newSpanTree(traceID string, cache *telemetry.TraceCache) (rootNodes []*span
 		st, en := span.Span.StartTimestamp().AsTime().Sub(start), span.Span.EndTimestamp().AsTime().Sub(start)
 		d := en - st
 		node.box = createSpan(colorMemo[sname.AsString()], duration, st, en)
-		node.label = fmt.Sprintf("%s %s", span.Span.Name(), d.String())
+		if span.Span.Status().Code() == ptrace.StatusCodeError {
+			node.label = fmt.Sprintf("[!] %s %s", span.Span.Name(), d.String())
+		} else {
+			node.label = fmt.Sprintf("%s %s", span.Span.Name(), d.String())
+		}
 
 		parent := span.Span.ParentSpanID().String()
 		_, parentExists := cache.GetSpanByID(parent)
@@ -484,7 +489,13 @@ func getSpanInfoTree(commands *tview.TextView, span *telemetry.SpanData, title s
 	smessageNode := tview.NewTreeNode(fmt.Sprintf("message: %s", smessage))
 	status.AddChild(smessageNode)
 	scode := span.Span.Status().Code()
-	scodeNode := tview.NewTreeNode(fmt.Sprintf("code: %s", scode))
+	scodeText := ""
+	if scode == ptrace.StatusCodeError {
+		scodeText = fmt.Sprintf("code: %s ⚠️", scode)
+	} else {
+		scodeText = fmt.Sprintf("code: %s", scode)
+	}
+	scodeNode := tview.NewTreeNode(scodeText)
 	status.AddChild(scodeNode)
 	root.AddChild(status)
 

tuiexporter/internal/tui/component/trace.go

@@ -10,7 +10,8 @@ import (
 	"go.opentelemetry.io/collector/pdata/pcommon"
 )
 
-var spanTableHeader = [4]string{
+var spanTableHeader = [5]string{
+	" ", // Error indicator
 	"Trace ID",
 	"Service Name",
 	"Received At",
@@ -20,13 +21,15 @@ var spanTableHeader = [4]string{
 // SpanDataForTable is a wrapper for spans to be displayed in a table.
 type SpanDataForTable struct {
 	tview.TableContentReadOnly
-	spans *telemetry.SvcSpans
+	tcache *telemetry.TraceCache
+	spans  *telemetry.SvcSpans
 }
 
 // NewSpanDataForTable creates a new SpanDataForTable.
-func NewSpanDataForTable(spans *telemetry.SvcSpans) SpanDataForTable {
+func NewSpanDataForTable(tcache *telemetry.TraceCache, spans *telemetry.SvcSpans) SpanDataForTable {
 	return SpanDataForTable{
-		spans: spans,
+		tcache: tcache,
+		spans:  spans,
 	}
 }
 
@@ -37,7 +40,7 @@ func (s SpanDataForTable) GetCell(row, column int) *tview.TableCell {
 		return getHeaderCell(spanTableHeader[:], column)
 	}
 	if row > 0 && row <= len(*s.spans) {
-		return getCellFromSpan((*s.spans)[row-1], column)
+		return s.getCellFromSpan((*s.spans)[row-1], column)
 	}
 	return tview.NewTableCell("N/A")
 }
@@ -50,38 +53,48 @@ func (s SpanDataForTable) GetColumnCount() int {
 	return len(spanTableHeader)
 }
 
-func getHeaderCell(header []string, column int) *tview.TableCell {
-	cell := tview.NewTableCell("N/A").
-		SetSelectable(false).
-		SetTextColor(tcell.ColorYellow)
-	if column >= len(header) {
-		return cell
-	}
-	cell.SetText(header[column])
-
-	return cell
-}
-
 // getCellFromSpan returns a table cell for the given span and column.
-func getCellFromSpan(span *telemetry.SpanData, column int) *tview.TableCell {
+func (s SpanDataForTable) getCellFromSpan(span *telemetry.SpanData, column int) *tview.TableCell {
 	text := "N/A"
 
 	switch column {
 	case 0:
-		text = span.Span.TraceID().String()
-	case 1:
-		if serviceName, ok := span.ResourceSpan.Resource().Attributes().Get("service.name"); ok {
-			text = serviceName.AsString()
+		if s.tcache == nil {
+			return tview.NewTableCell("")
 		}
+		text = ""
+		if sname, ok := span.ResourceSpan.Resource().Attributes().Get("service.name"); ok {
+			if haserr, ok := s.tcache.HasErrorByTraceIDAndSvc(span.Span.TraceID().String(), sname.AsString()); ok && haserr {
+				text = "[!]"
+			}
+		}
+	case 1:
+		text = span.Span.TraceID().String()
 	case 2:
-		text = span.ReceivedAt.Local().Format("2006-01-02 15:04:05")
+		if sname, ok := span.ResourceSpan.Resource().Attributes().Get("service.name"); ok {
+			text = sname.AsString()
+		}
 	case 3:
+		text = span.ReceivedAt.Local().Format("2006-01-02 15:04:05")
+	case 4:
 		text = span.Span.Name()
 	}
 
 	return tview.NewTableCell(text)
 }
 
+func getHeaderCell(header []string, column int) *tview.TableCell {
+	cell := tview.NewTableCell("N/A").
+		SetSelectable(false).
+		SetTextColor(tcell.ColorYellow)
+	if column >= len(header) {
+		return cell
+	}
+	cell.SetText(header[column])
+
+	return cell
+}
+
 func getTraceInfoTree(commands *tview.TextView, spans []*telemetry.SpanData) *tview.TreeView {
 	if len(spans) == 0 {
 		return tview.NewTreeView()

tuiexporter/internal/tui/component/trace_test.go

@@ -10,51 +10,69 @@ import (
 	"github.com/stretchr/testify/assert"
 	"github.com/ymtdzzz/otel-tui/tuiexporter/internal/telemetry"
 	"github.com/ymtdzzz/otel-tui/tuiexporter/internal/test"
+	"go.opentelemetry.io/collector/pdata/ptrace"
 )
 
 func TestSpanDataForTable(t *testing.T) {
 	// traceid: 1
 	//  └- resource: test-service-1
 	//  | └- scope: test-scope-1-1
-	//  | | └- span: span-1-1-1
+	//  | | └- span: span-1-1-1 (code: Error)
 	//  | | └- span: span-1-1-2
 	//  | └- scope: test-scope-1-2
 	//  |   └- span: span-1-2-3
 	//  └- resource: test-service-2
 	//    └- scope: test-scope-2-1
-	//      └- span: span-2-1-1
+	//      └- span: span-2-1-1 (code: OK)
 	// traceid: 2
 	//  └- resource: test-service-1
 	//    └- scope: test-scope-1-1
-	//      └- span: span-1-1-1
+	//      └- span: span-1-1-1 (code: Unset)
 	_, testdata1 := test.GenerateOTLPTracesPayload(t, 1, 2, []int{2, 1}, [][]int{{2, 1}, {1}})
 	_, testdata2 := test.GenerateOTLPTracesPayload(t, 2, 1, []int{1}, [][]int{{1}})
 	receivedAt := time.Date(2024, 3, 30, 12, 30, 15, 0, time.UTC)
-	svcspans := &telemetry.SvcSpans{
-		&telemetry.SpanData{
+	testdata1.Spans[0].Status().SetCode(ptrace.StatusCodeError)
+	testdata1.Spans[3].Status().SetCode(ptrace.StatusCodeOk)
+	testdata2.Spans[0].Status().SetCode(ptrace.StatusCodeUnset)
+	svc1sds := []*telemetry.SpanData{
+		{
 			Span:         testdata1.Spans[0],
 			ResourceSpan: testdata1.RSpans[0],
 			ReceivedAt:   receivedAt,
 		}, // trace 1, span-1-1-1
-		&telemetry.SpanData{
+		{
 			Span:         testdata1.Spans[3],
 			ResourceSpan: testdata1.RSpans[1],
 			ReceivedAt:   receivedAt,
 		}, // trace 1, span-2-1-1
-		&telemetry.SpanData{
+	}
+	svc2sds := []*telemetry.SpanData{
+		{
 			Span:         testdata2.Spans[0],
 			ResourceSpan: testdata2.RSpans[0],
 			ReceivedAt:   receivedAt,
 		}, // trace 2, span-1-1-1
 	}
-	sdftable := NewSpanDataForTable(svcspans)
+	svcspans := &telemetry.SvcSpans{
+		svc1sds[0],
+		svc1sds[1],
+		svc2sds[0],
+	}
+	tcache := telemetry.NewTraceCache()
+	for _, sd := range svc1sds {
+		tcache.UpdateCache("test-service-1", sd)
+	}
+	for _, sd := range svc2sds {
+		tcache.UpdateCache("test-service-2", sd)
+	}
+	sdftable := NewSpanDataForTable(tcache, svcspans)
 
 	t.Run("GetRowCount", func(t *testing.T) {
 		assert.Equal(t, 4, sdftable.GetRowCount()) // including header row
 	})
 
 	t.Run("GetColumnCount", func(t *testing.T) {
-		assert.Equal(t, 4, sdftable.GetColumnCount())
+		assert.Equal(t, 5, sdftable.GetColumnCount())
 	})
 
 	t.Run("GetCell", func(t *testing.T) {
@@ -67,37 +85,55 @@ func TestSpanDataForTable(t *testing.T) {
 			{
 				name:   "invalid row",
 				row:    3,
-				column: 0,
+				column: 1,
 				want:   "N/A",
 			},
 			{
 				name:   "invalid column",
 				row:    0,
-				column: 4,
+				column: 5,
 				want:   "N/A",
 			},
 			{
-				name:   "trace ID trace 1 span-1-1-1",
+				name:   "has error trace 1 span-1-1-1",
 				row:    0,
 				column: 0,
+				want:   "[!]",
+			},
+			{
+				name:   "has no errors (OK) trace 1 span-2-1-1",
+				row:    1,
+				column: 0,
+				want:   "",
+			},
+			{
+				name:   "has no errors (Unset) trace 2 span-1-1-1",
+				row:    2,
+				column: 0,
+				want:   "",
+			},
+			{
+				name:   "trace ID trace 1 span-1-1-1",
+				row:    0,
+				column: 1,
 				want:   "01000000000000000000000000000000",
 			},
 			{
 				name:   "service name trace 1 span-2-1-1",
 				row:    1,
-				column: 1,
+				column: 2,
 				want:   "test-service-2",
 			},
 			{
 				name:   "received at trace 2 span-1-1-1",
 				row:    2,
-				column: 2,
+				column: 3,
 				want:   receivedAt.Local().Format("2006-01-02 15:04:05"),
 			},
 			{
 				name:   "span name trace 2 span-1-1-1",
 				row:    2,
-				column: 3,
+				column: 4,
 				want:   "span-0-0-0",
 			},
 		}