feat(cloudevents-server): improve notify info (#98)

## Improvements

1. avoid notify for failed task run when it's created by a pipeline run.
2. add more details about failed tasks and step details for failed
pipeline run.

## Tests

- [x] Tested with real lark IM.

## Effects

<img width="592" alt="image"
src="https://github.com/PingCAP-QE/ee-apps/assets/2574558/2b33b33b-d5c9-4437-ae89-8c5ce07f5547">
<img width="597" alt="image"
src="https://github.com/PingCAP-QE/ee-apps/assets/2574558/b7f75fed-6dcd-45bb-a490-858ff318b021">

Signed-off-by: wuhuizuo <wuhuizuo@126.com>

cloudevents-server/pkg/events/custom/tekton/handler.go

@@ -2,8 +2,17 @@ package tekton
 
 import (
 	"encoding/json"
+	"fmt"
+	"strings"
+	"time"
 
+	cloudevents "github.com/cloudevents/sdk-go/v2"
 	larksdk "github.com/larksuite/oapi-sdk-go/v3"
+	"github.com/tektoncd/pipeline/pkg/apis/pipeline/v1alpha1"
+	"github.com/tektoncd/pipeline/pkg/apis/pipeline/v1beta1"
+	tektoncloudevent "github.com/tektoncd/pipeline/pkg/reconciler/events/cloudevent"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"knative.dev/pkg/apis"
 
 	"github.com/PingCAP-QE/ee-apps/cloudevents-server/pkg/config"
 	"github.com/PingCAP-QE/ee-apps/cloudevents-server/pkg/events/handler"
@@ -36,3 +45,143 @@ func getTriggerUser(run AnnotationsGetter) string {
 
 	return contextData[eventContextAnnotationInnerKeyUser]
 }
+
+// <dashboard base url>/#/namespaces/<namespace>/<run-type>s/<run-name>
+// source: /apis///namespaces/<namespace>//<run-name>
+// https://tekton.abc.com/tekton/apis/tekton.dev/v1beta1/namespaces/ee-cd/pipelineruns/auto-compose-multi-arch-image-run-g5hqv
+//
+//	"source": "/apis///namespaces/ee-cd//build-package-tikv-tikv-linux-9bn55-build-binaries",
+func newDetailURL(etype, source, baseURL string) string {
+	words := strings.Split(source, "/")
+	runName := words[len(words)-1]
+	runType := words[len(words)-2]
+	runNamespace := words[len(words)-3]
+
+	if runType == "" {
+		runType = strings.Split(etype, ".")[3] + "s"
+	}
+
+	return fmt.Sprintf("%s/#/namespaces/%s/%s/%s", baseURL, runNamespace, runType, runName)
+}
+
+func extractLarkInfosFromEvent(event cloudevents.Event, baseURL string) (*cardMessageInfos, error) {
+	var data tektoncloudevent.TektonCloudEventData
+	if err := event.DataAs(&data); err != nil {
+		return nil, err
+	}
+
+	ret := cardMessageInfos{
+		Title:         newLarkTitle(event.Type(), event.Subject()),
+		TitleTemplate: larkCardHeaderTemplates[tektoncloudevent.TektonEventType(event.Type())],
+		ViewURL:       newDetailURL(event.Type(), event.Source(), baseURL),
+	}
+
+	switch {
+	case data.PipelineRun != nil:
+		fillInfosWithPipelineRun(data.PipelineRun, &ret)
+		if event.Type() == string(tektoncloudevent.PipelineRunFailedEventV1) {
+			ret.FailedTasks = getFailedTasks(data.PipelineRun)
+		}
+	case data.TaskRun != nil:
+		fillInfosWithTaskRun(data.TaskRun, &ret)
+		if event.Type() == string(tektoncloudevent.TaskRunFailedEventV1) {
+			ret.StepStatuses = getStepStatuses(&data.TaskRun.Status)
+		}
+	case data.Run != nil:
+		fillInfosWithCustomRun(data.Run, &ret)
+	}
+
+	return &ret, nil
+}
+
+func fillInfosWithCustomRun(data *v1alpha1.Run, ret *cardMessageInfos) {
+	fillTimeFileds(ret, data.Status.StartTime, data.Status.CompletionTime)
+
+	for _, p := range data.Spec.Params {
+		v, _ := p.Value.MarshalJSON()
+		ret.Params = append(ret.Params, [2]string{p.Name, string(v)})
+	}
+	if results := data.Status.Results; len(results) > 0 {
+		var parts []string
+		for _, r := range results {
+			parts = append(parts, fmt.Sprintf("**%s**:", r.Name), r.Value, "---")
+			ret.Results = append(ret.Results, [2]string{r.Name, r.Value})
+
+		}
+	}
+}
+
+func fillInfosWithTaskRun(data *v1beta1.TaskRun, ret *cardMessageInfos) {
+	fillTimeFileds(ret, data.Status.StartTime, data.Status.CompletionTime)
+
+	for _, p := range data.Spec.Params {
+		v, _ := p.Value.MarshalJSON()
+		ret.Params = append(ret.Params, [2]string{p.Name, string(v)})
+	}
+	if data.Status.GetCondition(apis.ConditionSucceeded).IsFalse() {
+		ret.RerunURL = fmt.Sprintf("tkn -n %s task start %s --use-taskrun %s",
+			data.Namespace, data.Spec.TaskRef.Name, data.Name)
+	}
+	if results := data.Status.TaskRunResults; len(results) > 0 {
+		for _, r := range results {
+			v, _ := r.Value.MarshalJSON()
+			ret.Results = append(ret.Results, [2]string{r.Name, string(v)})
+		}
+	}
+
+	getStepStatuses(&data.Status)
+}
+
+func fillInfosWithPipelineRun(data *v1beta1.PipelineRun, ret *cardMessageInfos) {
+	fillTimeFileds(ret, data.Status.StartTime, data.Status.CompletionTime)
+
+	for _, p := range data.Spec.Params {
+		v, _ := p.Value.MarshalJSON()
+		ret.Params = append(ret.Params, [2]string{p.Name, string(v)})
+	}
+	if data.Status.GetCondition(apis.ConditionSucceeded).IsFalse() {
+		ret.RerunURL = fmt.Sprintf("tkn -n %s pipeline start %s --use-pipelinerun %s",
+			data.Namespace, data.Spec.PipelineRef.Name, data.Name)
+	}
+	if results := data.Status.PipelineResults; len(results) > 0 {
+		for _, r := range results {
+			ret.Results = append(ret.Results, [2]string{r.Name, r.Value.StringVal})
+		}
+	}
+
+}
+
+func getFailedTasks(data *v1beta1.PipelineRun) map[string][][2]string {
+	ret := make(map[string][][2]string)
+	for _, v := range data.Status.TaskRuns {
+		succeededCondition := v.Status.GetCondition(apis.ConditionSucceeded)
+		if !succeededCondition.IsTrue() {
+			ret[v.PipelineTaskName] = getStepStatuses(v.Status)
+		}
+	}
+
+	return ret
+}
+
+func getStepStatuses(status *v1beta1.TaskRunStatus) [][2]string {
+	var ret [][2]string
+	for _, s := range status.Steps {
+		if s.Terminated != nil {
+			ret = append(ret, [2]string{s.Name, s.Terminated.Reason})
+		}
+	}
+
+	return ret
+}
+
+func fillTimeFileds(ret *cardMessageInfos, startTime, endTime *metav1.Time) {
+	if startTime != nil {
+		ret.StartTime = startTime.Format(time.RFC3339)
+	}
+	if endTime != nil {
+		ret.EndTime = endTime.Format(time.RFC3339)
+	}
+	if startTime != nil && endTime != nil {
+		ret.TimeCost = time.Duration(endTime.UnixNano() - startTime.UnixNano()).String()
+	}
+}

cloudevents-server/pkg/events/custom/tekton/handler_pipelinerun.go

@@ -59,7 +59,10 @@ func (h *pipelineRunHandler) Handle(event cloudevents.Event) cloudevents.Result
 
 		return sendLarkMessages(h.LarkClient, receivers, event, h.DashboardBaseURL)
 	default:
-		log.Debug().Str("ce-type", event.Type()).Msg("skip notifing for the event type.")
+		log.Debug().
+			Str("handler", "pipelineRunHandler").
+			Str("ce-type", event.Type()).
+			Msg("skip notifing for the event type.")
 		return cloudevents.ResultACK
 	}
 }

cloudevents-server/pkg/events/custom/tekton/handler_taskrun.go

@@ -8,6 +8,7 @@ import (
 	cloudevents "github.com/cloudevents/sdk-go/v2"
 	lark "github.com/larksuite/oapi-sdk-go/v3"
 	"github.com/rs/zerolog/log"
+	"github.com/tektoncd/pipeline/pkg/apis/pipeline"
 	tektoncloudevent "github.com/tektoncd/pipeline/pkg/reconciler/events/cloudevent"
 )
 
@@ -34,6 +35,11 @@ func (h *taskRunHandler) Handle(event cloudevents.Event) cloudevents.Result {
 
 	switch event.Type() {
 	case string(tektoncloudevent.TaskRunFailedEventV1):
+		// Skip notify the taskrun when it created by a pipelineRun.
+		if data.TaskRun.Labels[pipeline.PipelineRunLabelKey] != "" {
+			break
+		}
+
 		var receivers []string
 		// send notify to the trigger user if it's existed, else send to the receivers configurated by type.
 		if receiver := getTriggerUser(data.TaskRun); receiver != "" {
@@ -48,8 +54,11 @@ func (h *taskRunHandler) Handle(event cloudevents.Event) cloudevents.Result {
 			Msg("send notification for the event type.")
 
 		return sendLarkMessages(h.LarkClient, receivers, event, h.DashboardBaseURL)
-	default:
-		log.Debug().Str("ce-type", event.Type()).Msg("skip notifing for the event type.")
-		return cloudevents.ResultACK
 	}
+
+	log.Debug().
+		Str("handler", "taskRunHandler").
+		Str("ce-type", event.Type()).
+		Msg("skip notifing for the event type.")
+	return cloudevents.ResultACK
 }

cloudevents-server/pkg/events/custom/tekton/lark.go

@@ -9,7 +9,6 @@ import (
 	"regexp"
 	"strings"
 	"text/template"
-	"time"
 
 	"github.com/Masterminds/sprig/v3"
 	cloudevents "github.com/cloudevents/sdk-go/v2"
@@ -19,8 +18,6 @@ import (
 	"github.com/rs/zerolog/log"
 	tektoncloudevent "github.com/tektoncd/pipeline/pkg/reconciler/events/cloudevent"
 	"gopkg.in/yaml.v3"
-	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
-	"knative.dev/pkg/apis"
 
 	_ "embed"
 )
@@ -115,84 +112,6 @@ func newLarkMessages(receivers []string, event cloudevents.Event, detailBaseUrl
 	return reqs, nil
 }
 
-func extractLarkInfosFromEvent(event cloudevents.Event, baseURL string) (*cardMessageInfos, error) {
-	var data tektoncloudevent.TektonCloudEventData
-	if err := event.DataAs(&data); err != nil {
-		return nil, err
-	}
-
-	ret := cardMessageInfos{
-		Title:         newLarkTitle(event.Type(), event.Subject()),
-		TitleTemplate: larkCardHeaderTemplates[tektoncloudevent.TektonEventType(event.Type())],
-		ViewURL:       newDetailURL(event.Type(), event.Source(), baseURL),
-	}
-
-	var startTime, endTime *metav1.Time
-	switch {
-	case data.PipelineRun != nil:
-		startTime = data.PipelineRun.Status.StartTime
-		endTime = data.PipelineRun.Status.CompletionTime
-		for _, p := range data.PipelineRun.Spec.Params {
-			v, _ := p.Value.MarshalJSON()
-			ret.Params = append(ret.Params, [2]string{p.Name, string(v)})
-		}
-		if data.PipelineRun.Status.GetCondition(apis.ConditionSucceeded).IsFalse() {
-			ret.RerunURL = fmt.Sprintf("tkn -n %s pipeline start %s --use-pipelinerun %s",
-				data.PipelineRun.Namespace, data.PipelineRun.Spec.PipelineRef.Name, data.PipelineRun.Name)
-		}
-		if results := data.PipelineRun.Status.PipelineResults; len(results) > 0 {
-			for _, r := range results {
-				ret.Results = append(ret.Results, [2]string{r.Name, r.Value.StringVal})
-			}
-		}
-	case data.TaskRun != nil:
-		for _, p := range data.TaskRun.Spec.Params {
-			v, _ := p.Value.MarshalJSON()
-			ret.Params = append(ret.Params, [2]string{p.Name, string(v)})
-		}
-		startTime = data.TaskRun.Status.StartTime
-		endTime = data.TaskRun.Status.CompletionTime
-		if data.TaskRun.Status.GetCondition(apis.ConditionSucceeded).IsFalse() {
-			ret.RerunURL = fmt.Sprintf("tkn -n %s task start %s --use-taskrun %s",
-				data.TaskRun.Namespace, data.TaskRun.Spec.TaskRef.Name, data.TaskRun.Name)
-		}
-
-		if results := data.TaskRun.Status.TaskRunResults; len(results) > 0 {
-			for _, r := range results {
-				v, _ := r.Value.MarshalJSON()
-				ret.Results = append(ret.Results, [2]string{r.Name, string(v)})
-			}
-		}
-	case data.Run != nil:
-		startTime = data.Run.Status.StartTime
-		endTime = data.Run.Status.CompletionTime
-		for _, p := range data.Run.Spec.Params {
-			v, _ := p.Value.MarshalJSON()
-			ret.Params = append(ret.Params, [2]string{p.Name, string(v)})
-		}
-		if results := data.Run.Status.Results; len(results) > 0 {
-			var parts []string
-			for _, r := range results {
-				parts = append(parts, fmt.Sprintf("**%s**:", r.Name), r.Value, "---")
-				ret.Results = append(ret.Results, [2]string{r.Name, r.Value})
-
-			}
-		}
-	}
-
-	if startTime != nil {
-		ret.StartTime = startTime.Format(time.RFC3339)
-	}
-	if endTime != nil {
-		ret.EndTime = endTime.Format(time.RFC3339)
-	}
-	if startTime != nil && endTime != nil {
-		ret.TimeCost = time.Duration(endTime.UnixNano() - startTime.UnixNano()).String()
-	}
-
-	return &ret, nil
-}
-
 func newLarkCardWithGoTemplate(event cloudevents.Event, baseURL string) (string, error) {
 	infos, err := extractLarkInfosFromEvent(event, baseURL)
 	if err != nil {
@@ -232,21 +151,3 @@ func newLarkTitle(etype, subject string) string {
 
 	return fmt.Sprintf("%s [%s] %s is %s ", larkCardHeaderEmojis[tektoncloudevent.TektonEventType(etype)], runType, subject, runState)
 }
-
-// <dashboard base url>/#/namespaces/<namespace>/<run-type>s/<run-name>
-// source: /apis///namespaces/<namespace>//<run-name>
-// https://tekton.abc.com/tekton/apis/tekton.dev/v1beta1/namespaces/ee-cd/pipelineruns/auto-compose-multi-arch-image-run-g5hqv
-//
-//	"source": "/apis///namespaces/ee-cd//build-package-tikv-tikv-linux-9bn55-build-binaries",
-func newDetailURL(etype, source, baseURL string) string {
-	words := strings.Split(source, "/")
-	runName := words[len(words)-1]
-	runType := words[len(words)-2]
-	runNamespace := words[len(words)-3]
-
-	if runType == "" {
-		runType = strings.Split(etype, ".")[3] + "s"
-	}
-
-	return fmt.Sprintf("%s/#/namespaces/%s/%s/%s", baseURL, runNamespace, runType, runName)
-}

cloudevents-server/pkg/events/custom/tekton/lark_templates/tekton-run-notify.yaml.tmpl

@@ -32,6 +32,29 @@ elements:
       {{- end }}
   {{- end }}
 
+  {{- with .FailedTasks }}
+  - tag: hr
+  - tag: markdown
+    content: |-
+      **Failed tasks:**
+      {{- range $t, $ss := . }}
+      - {{ $t }}:
+      {{- range $ss }}
+        1. {{ index . 0 }}: {{ index . 1 }}
+      {{- end }}
+      {{- end }}
+  {{- end }}
+
+  {{- with .StepStatuses }}
+  - tag: hr
+  - tag: markdown
+    content: |-
+      **Steps:**
+      {{- range . }}
+      1. {{ index . 0 }}: {{ index . 1 }}
+      {{- end }}
+  {{- end }}
+
   {{- with .Results }}
   - tag: hr
   - tag: markdown

cloudevents-server/pkg/events/custom/tekton/types.go

@@ -13,8 +13,10 @@ type cardMessageInfos struct {
 	StartTime     string
 	EndTime       string
 	TimeCost      string
-	Params        [][2]string // key-value pairs.
-	Results       [][2]string // Key-Value pairs.
+	Params        [][2]string            // key-value pairs.
+	Results       [][2]string            // Key-Value pairs.
+	StepStatuses  [][2]string            // name-status pairs.
+	FailedTasks   map[string][][2]string // task id => step statuses.
 }
 
 var larkCardHeaderTemplates = map[tektoncloudevent.TektonEventType]string{