Add TraceTTL to cassandra schema spec

fixes #1101

Signed-off-by: Moritz Johner <beller.moritz@googlemail.com>

deploy/crds/jaegertracing.io_jaegers_crd.yaml

@@ -6227,6 +6227,8 @@ spec:
                       type: string
                     timeout:
                       type: string
+                    traceTTL:
+                      type: string
                     ttlSecondsAfterFinished:
                       format: int32
                       type: integer

pkg/apis/jaegertracing/v1/jaeger_types.go

@@ -431,16 +431,33 @@ type JaegerCassandraCreateSchemaSpec struct {
 	// +optional
 	Enabled *bool `json:"enabled,omitempty"`
 
+	// Image specifies the docker image to use to create the cassandra schema.
+	// The Image is used by a Kubernetes Job, defaults to the image provided through the cli flag "jaeger-cassandra-schema-image" (default: jaegertracing/jaeger-cassandra-schema).
+	// See here for the jaeger-provided image: https://github.com/jaegertracing/jaeger/tree/master/plugin/storage/cassandra
 	// +optional
 	Image string `json:"image,omitempty"`
 
+	// Datacenter is a collection of racks in the cassandra topology.
+	// defaults to "test"
 	// +optional
 	Datacenter string `json:"datacenter,omitempty"`
 
+	// Mode controls the replication factor of your cassandra schema.
+	// Set it to "prod" (which is the default) to use the NetworkTopologyStrategy with a replication factor of 2, effectively meaning
+	// that at least 3 nodes are required in the cassandra cluster.
+	// When set to "test" the schema uses the SimpleStrategy with a replication factor of 1. You never want to do this in a production setup.
 	// +optional
 	Mode string `json:"mode,omitempty"`
 
-	// we parse it with time.ParseDuration
+	// TraceTTL sets the TTL for your trace data
+	// +optional
+	TraceTTL string `json:"traceTTL,omitempty"`
+
+	// Timeout controls the Job deadline, it defaults to 1 day.
+	// specify it with a value which can be parsed by time.ParseDuration, e.g. 24h or 120m.
+	// If the job does not succeed within that duration it transitions into a permanent error state.
+	// See https://github.com/jaegertracing/jaeger-kubernetes/issues/32 and
+	// https://github.com/jaegertracing/jaeger-kubernetes/pull/125
 	// +optional
 	Timeout string `json:"timeout,omitempty"`
 

pkg/storage/cassandra_dependencies.go

@@ -1,6 +1,7 @@
 package storage
 
 import (
+	"fmt"
 	"time"
 
 	batchv1 "k8s.io/api/batch/v1"
@@ -72,6 +73,21 @@ func cassandraDeps(jaeger *v1.Jaeger) []batchv1.Job {
 	podTimeoutSeconds := int64(320)
 	podTimeout := &podTimeoutSeconds
 
+	// TTL for trace data, in seconds (default: 172800, 2 days)
+	// see: https://github.com/jaegertracing/jaeger/blob/master/plugin/storage/cassandra/schema/create.sh
+	traceTTLSeconds := "172800"
+	if jaeger.Spec.Storage.CassandraCreateSchema.TraceTTL != "" {
+		dur, err := time.ParseDuration(jaeger.Spec.Storage.CassandraCreateSchema.TraceTTL)
+		if err != nil {
+			jaeger.Logger().
+				WithError(err).
+				WithField("timeout", jaeger.Spec.Storage.CassandraCreateSchema.TraceTTL).
+				Error("Failed to parse cassandraCreateSchema.traceTTL to time.duration. Using the default.")
+		} else {
+			traceTTLSeconds = fmt.Sprintf("%.0f", dur.Seconds())
+		}
+	}
+
 	if jaeger.Spec.Storage.CassandraCreateSchema.Timeout != "" {
 		dur, err := time.ParseDuration(jaeger.Spec.Storage.CassandraCreateSchema.Timeout)
 		if err == nil {
@@ -89,7 +105,7 @@ func cassandraDeps(jaeger *v1.Jaeger) []batchv1.Job {
 
 	truncatedName := util.Truncate("%s-cassandra-schema-job", 63, jaeger.Name)
 	return []batchv1.Job{
-		batchv1.Job{
+		{
 			TypeMeta: metav1.TypeMeta{
 				APIVersion: "batch/v1",
 				Kind:       "Job",
@@ -101,7 +117,7 @@ func cassandraDeps(jaeger *v1.Jaeger) []batchv1.Job {
 				Namespace: jaeger.Namespace,
 				Labels:    util.Labels(truncatedName, "cronjob-cassandra-schema", *jaeger),
 				OwnerReferences: []metav1.OwnerReference{
-					metav1.OwnerReference{
+					{
 						APIVersion: jaeger.APIVersion,
 						Kind:       jaeger.Kind,
 						Name:       jaeger.Name,
@@ -130,6 +146,9 @@ func cassandraDeps(jaeger *v1.Jaeger) []batchv1.Job {
 							}, {
 								Name:  "DATACENTER",
 								Value: jaeger.Spec.Storage.CassandraCreateSchema.Datacenter,
+							}, {
+								Name:  "TRACE_TTL",
+								Value: traceTTLSeconds,
 							}, {
 								Name:  "KEYSPACE",
 								Value: keyspace,

pkg/storage/cassandra_dependencies_test.go

@@ -22,6 +22,40 @@ func TestCassandraCustomImage(t *testing.T) {
 	assert.Equal(t, "mynamespace/image:version", b[0].Spec.Template.Spec.Containers[0].Image)
 }
 
+func TestCassandraCustomTraceTTL(t *testing.T) {
+	jaeger := v1.NewJaeger(types.NamespacedName{Name: "my-instance"})
+	jaeger.Spec.Storage.CassandraCreateSchema.TraceTTL = "168h" // 7d
+
+	b := cassandraDeps(jaeger)
+	assert.Len(t, b, 1)
+	assert.Len(t, b[0].Spec.Template.Spec.Containers, 1)
+	foundValue := ""
+	for _, e := range b[0].Spec.Template.Spec.Containers[0].Env {
+		if e.Name == "TRACE_TTL" {
+			foundValue = e.Value
+		}
+	}
+	assert.Equal(t, "604800", foundValue, "unexpected TRACE_TTL environment var value")
+}
+
+func TestCassandraCustomTraceTTLParseError(t *testing.T) {
+	jaeger := v1.NewJaeger(types.NamespacedName{Name: "my-instance"})
+	// this does not work. time.ParseDuration can not handle "days"
+	// TRACE_TTL should fallback to default value
+	jaeger.Spec.Storage.CassandraCreateSchema.TraceTTL = "7d"
+
+	b := cassandraDeps(jaeger)
+	assert.Len(t, b, 1)
+	assert.Len(t, b[0].Spec.Template.Spec.Containers, 1)
+	foundValue := ""
+	for _, e := range b[0].Spec.Template.Spec.Containers[0].Env {
+		if e.Name == "TRACE_TTL" {
+			foundValue = e.Value
+		}
+	}
+	assert.Equal(t, "172800", foundValue, "unexpected TRACE_TTL environment var value")
+}
+
 func TestDefaultImage(t *testing.T) {
 	viper.Set("jaeger-cassandra-schema-image", "jaegertracing/theimage")
 	defer viper.Reset()