diff --git a/test/cmd/aro-hcp-tests/custom-link-tools/cmd_test.go b/test/cmd/aro-hcp-tests/custom-link-tools/cmd_test.go
index 1f956353c5..1b61503524 100644
--- a/test/cmd/aro-hcp-tests/custom-link-tools/cmd_test.go
+++ b/test/cmd/aro-hcp-tests/custom-link-tools/cmd_test.go
@@ -25,7 +25,6 @@ import (
clocktesting "k8s.io/utils/clock/testing"
"github.com/Azure/ARO-HCP/test/util/testutil"
- "github.com/Azure/ARO-HCP/tooling/templatize/pkg/pipeline"
)
func TestGeneratedHTML(t *testing.T) {
@@ -39,31 +38,14 @@ func TestGeneratedHTML(t *testing.T) {
tmpdir := t.TempDir()
opts := Options{
completedOptions: &completedOptions{
- TimingInputDir: "../testdata/output",
- Steps: []pipeline.NodeInfo{
- {
- Identifier: pipeline.Identifier{
- ServiceGroup: "Microsoft.Azure.ARO.HCP.Service.Infra",
- ResourceGroup: "service",
- Step: "cluster",
- },
- Details: &pipeline.ExecutionDetails{
- ARM: &pipeline.ARMExecutionDetails{
- Operations: []pipeline.Operation{
- {
- OperationType: "Create",
- Resource: &pipeline.Resource{
- ResourceType: "Microsoft.ContainerService/managedClusters",
- Name: "hcp-underlay-prow-usw3j688-svc-1",
- ResourceGroup: "hcp-underlay-prow-usw3j688-svc-1",
- },
- },
- },
- },
- },
- },
+ TimingInputDir: "../testdata/output",
+ ADXClusterName: "hcp-dev-us.westus3",
+ ClusterNames: ClusterNames{
+ Service: "hcp-underlay-prow-usw3j688-svc-1",
+ Management: "hcp-underlay-prow-usw3j688-mgmt-1",
},
- OutputDir: tmpdir,
+ GlobalLogStartTime: fakeTime.Add(-6 * time.Hour),
+ OutputDir: tmpdir,
},
}
err = opts.Run(ctx)
diff --git a/test/cmd/aro-hcp-tests/custom-link-tools/options.go b/test/cmd/aro-hcp-tests/custom-link-tools/options.go
index 0e3f463a5e..7e0fa1ecec 100644
--- a/test/cmd/aro-hcp-tests/custom-link-tools/options.go
+++ b/test/cmd/aro-hcp-tests/custom-link-tools/options.go
@@ -35,6 +35,7 @@ import (
"k8s.io/utils/clock"
+ configtypes "github.com/Azure/ARO-Tools/pkg/config/types"
"sigs.k8s.io/yaml"
"github.com/Azure/ARO-HCP/internal/utils"
@@ -47,13 +48,11 @@ var templatesFS embed.FS
var endGracePeriodDuration = 45 * time.Minute
-var (
- serviceClusterStepID = pipeline.Identifier{
- ServiceGroup: "Microsoft.Azure.ARO.HCP.Service.Infra",
- ResourceGroup: "service",
- Step: "cluster",
- }
-)
+// ClusterNames holds the names of the service and management AKS clusters.
+type ClusterNames struct {
+ Service string
+ Management string
+}
func mustReadArtifact(name string) []byte {
ret, err := templatesFS.ReadFile("artifacts/" + name)
@@ -71,13 +70,15 @@ func DefaultOptions() *RawOptions {
func BindOptions(opts *RawOptions, cmd *cobra.Command) error {
cmd.Flags().StringVar(&opts.TimingInputDir, "timing-input", opts.TimingInputDir, "Path to the directory holding timing outputs from an end-to-end test run.")
cmd.Flags().StringVar(&opts.OutputDir, "output", opts.OutputDir, "Path to the directory where html will be written.")
+ cmd.Flags().StringVar(&opts.RenderedConfigPath, "rendered-config", opts.RenderedConfigPath, "Path to a rendered config.yaml file for the target environment.")
return nil
}
type RawOptions struct {
- TimingInputDir string
- OutputDir string
+ TimingInputDir string
+ OutputDir string
+ RenderedConfigPath string
}
// validatedOptions is a private wrapper that enforces a call of Validate() before Complete() can be invoked.
@@ -92,9 +93,11 @@ type ValidatedOptions struct {
// completedOptions is a private wrapper that enforces a call of Complete() before config generation can be invoked.
type completedOptions struct {
- TimingInputDir string
- OutputDir string
- Steps []pipeline.NodeInfo
+ TimingInputDir string
+ OutputDir string
+ ClusterNames ClusterNames
+ ADXClusterName string
+ GlobalLogStartTime time.Time
}
type Options struct {
@@ -110,6 +113,7 @@ func (o *RawOptions) Validate() (*ValidatedOptions, error) {
}{
{flag: "timing-input", name: "timing input dir", value: &o.TimingInputDir},
{flag: "output", name: "output dir", value: &o.OutputDir},
+ {flag: "rendered-config", name: "rendered config path", value: &o.RenderedConfigPath},
} {
if item.value == nil || *item.value == "" {
return nil, fmt.Errorf("the %s must be provided with --%s", item.name, item.flag)
@@ -124,43 +128,62 @@ func (o *RawOptions) Validate() (*ValidatedOptions, error) {
}
func (o *ValidatedOptions) Complete(logger logr.Logger) (*Options, error) {
- // we consume steps.yaml (output of templatize and stored for us by the visualization) to determine the cluster name
- // Try to read compressed file first, then fall back to uncompressed
- var stepsYamlBytes []byte
-
- compressedPath := path.Join(o.TimingInputDir, "steps.yaml.gz")
- uncompressedPath := path.Join(o.TimingInputDir, "steps.yaml")
-
- // Try compressed file first
- compressedData, err := os.ReadFile(compressedPath)
- if err == nil {
- gzipReader, err := gzip.NewReader(bytes.NewReader(compressedData))
- if err != nil {
- return nil, fmt.Errorf("failed to create gzip reader for %s: %w", compressedPath, err)
- }
- defer gzipReader.Close()
+ // Read and parse the rendered config.yaml.
+ configData, err := os.ReadFile(o.RenderedConfigPath)
+ if err != nil {
+ return nil, fmt.Errorf("failed to read rendered config %s: %w", o.RenderedConfigPath, err)
+ }
+ var cfg configtypes.Configuration
+ if err := yaml.Unmarshal(configData, &cfg); err != nil {
+ return nil, fmt.Errorf("failed to unmarshal rendered config %s: %w", o.RenderedConfigPath, err)
+ }
- stepsYamlBytes, err = io.ReadAll(gzipReader)
- if err != nil {
- return nil, fmt.Errorf("failed to decompress %s: %w", compressedPath, err)
- }
- } else {
- // Fall back to uncompressed file
- stepsYamlBytes, err = os.ReadFile(uncompressedPath)
- if err != nil {
- return nil, utils.TrackError(err)
- }
+ // Extract ADX cluster name: kustoName + "." + region (e.g. "hcp-int-uk.uksouth").
+ kustoNameAny, err := cfg.GetByPath("kusto.kustoName")
+ if err != nil {
+ return nil, fmt.Errorf("failed to get kusto.kustoName from config: %w", err)
+ }
+ kustoName, ok := kustoNameAny.(string)
+ if !ok {
+ return nil, fmt.Errorf("kusto.kustoName is not a string in config")
+ }
+ regionAny, err := cfg.GetByPath("region")
+ if err != nil {
+ return nil, fmt.Errorf("failed to get region from config: %w", err)
}
+ region, ok := regionAny.(string)
+ if !ok {
+ return nil, fmt.Errorf("region is not a string in config")
+ }
+ adxClusterName := kustoName + "." + region
- var steps []pipeline.NodeInfo
- if err := yaml.Unmarshal(stepsYamlBytes, &steps); err != nil {
- return nil, fmt.Errorf("failed to unmarshal timing input file: %w", err)
+ // Extract cluster names.
+ svcNameAny, err := cfg.GetByPath("svc.aks.name")
+ if err != nil {
+ return nil, fmt.Errorf("failed to get svc.aks.name from config: %w", err)
+ }
+ svcName, ok := svcNameAny.(string)
+ if !ok {
+ return nil, fmt.Errorf("svc.aks.name is not a string in config")
}
+ mgmtNameAny, err := cfg.GetByPath("mgmt.aks.name")
+ if err != nil {
+ return nil, fmt.Errorf("failed to get mgmt.aks.name from config: %w", err)
+ }
+ mgmtName, ok := mgmtNameAny.(string)
+ if !ok {
+ return nil, fmt.Errorf("mgmt.aks.name is not a string in config")
+ }
+
+ globalLogStartTime := computeGlobalLogStartTime(logger, o.TimingInputDir)
+
return &Options{
completedOptions: &completedOptions{
- Steps: steps,
- OutputDir: o.OutputDir,
- TimingInputDir: o.TimingInputDir,
+ ClusterNames: ClusterNames{Service: svcName, Management: mgmtName},
+ ADXClusterName: adxClusterName,
+ GlobalLogStartTime: globalLogStartTime,
+ OutputDir: o.OutputDir,
+ TimingInputDir: o.TimingInputDir,
},
}, nil
}
@@ -184,6 +207,7 @@ type QueryInfo struct {
EndTime string
ClusterName string
Database string
+ ADXClusterName string
}
type TimingInfo struct {
@@ -202,7 +226,7 @@ func createQueryURL(templatePath string, info QueryInfo) string {
currURL := url.URL{
Scheme: "https",
Host: "dataexplorer.azure.com",
- Path: fmt.Sprintf("clusters/hcp-dev-us.westus3/databases/%s", info.Database),
+ Path: fmt.Sprintf("clusters/%s/databases/%s", info.ADXClusterName, info.Database),
}
urlQuery := currURL.Query()
template, err := template.New("custom-link-tools").Parse(string(mustReadArtifact(templatePath)))
@@ -267,12 +291,14 @@ func (o Options) Run(ctx context.Context) error {
Database: "HostedControlPlaneLogs",
StartTime: timing.StartTime.Format(time.RFC3339),
EndTime: timing.EndTime.Format(time.RFC3339),
+ ADXClusterName: o.ADXClusterName,
}),
createLinkForTest("Service Logs", "service-logs.kql.tmpl", QueryInfo{
ResourceGroupName: rg,
Database: "ServiceLogs",
StartTime: timing.StartTime.Format(time.RFC3339),
EndTime: timing.EndTime.Format(time.RFC3339),
+ ADXClusterName: o.ADXClusterName,
}),
},
Database: "HostedControlPlaneLogs",
@@ -296,7 +322,10 @@ func (o Options) Run(ctx context.Context) error {
return utils.TrackError(err)
}
- serviceLogLinks, err := getServiceLogLinks(o.Steps)
+ endTime := localClock.Now().Add(1 * time.Hour)
+ serviceLogLinks, err := getServiceLogLinksFromClusterNamesWithWindow(
+ o.ClusterNames, o.ADXClusterName, o.GlobalLogStartTime, endTime,
+ )
if err != nil {
return utils.TrackError(err)
}
@@ -413,44 +442,102 @@ func loadAllTestTimingInfo(timingInputDir string) (map[string]TimingInfo, error)
var localClock clock.PassiveClock = clock.RealClock{}
-func getServiceLogLinks(steps []pipeline.NodeInfo) ([]LinkDetails, error) {
- allLinks := []LinkDetails{}
+// computeGlobalLogStartTime determines the start time for global component log links
+// using a three-tier fallback: steps.yaml, timing-metadata files, or 6 hours ago.
+func computeGlobalLogStartTime(logger logr.Logger, timingInputDir string) time.Time {
+ if t, ok := extractStartTimeFromStepsYaml(logger, timingInputDir); ok {
+ return t
+ }
+ if t, ok := extractEarliestStartTimeFromTimingMetadata(logger, timingInputDir); ok {
+ return t
+ }
+ return localClock.Now().Add(-6 * time.Hour)
+}
+
+// extractStartTimeFromStepsYaml reads steps.yaml(.gz) and returns the earliest StartedAt time.
+// Returns false on any error (soft failure).
+func extractStartTimeFromStepsYaml(logger logr.Logger, timingInputDir string) (time.Time, bool) {
+ var stepsYamlBytes []byte
+
+ compressedPath := path.Join(timingInputDir, "steps.yaml.gz")
+ uncompressedPath := path.Join(timingInputDir, "steps.yaml")
+
+ compressedData, err := os.ReadFile(compressedPath)
+ if err == nil {
+ gzipReader, gzErr := gzip.NewReader(bytes.NewReader(compressedData))
+ if gzErr != nil {
+ logger.V(1).Info("failed to create gzip reader for steps.yaml.gz, skipping", "error", gzErr)
+ return time.Time{}, false
+ }
+ defer gzipReader.Close()
+
+ stepsYamlBytes, err = io.ReadAll(gzipReader)
+ if err != nil {
+ logger.V(1).Info("failed to decompress steps.yaml.gz, skipping", "error", err)
+ return time.Time{}, false
+ }
+ } else {
+ stepsYamlBytes, err = os.ReadFile(uncompressedPath)
+ if err != nil {
+ logger.V(1).Info("steps.yaml not found, skipping", "error", err)
+ return time.Time{}, false
+ }
+ }
+
+ var steps []pipeline.NodeInfo
+ if err := yaml.Unmarshal(stepsYamlBytes, &steps); err != nil {
+ logger.V(1).Info("failed to unmarshal steps.yaml, skipping", "error", err)
+ return time.Time{}, false
+ }
- earliestStartTime := time.Time{}
- allClusterNames := []string{}
+ earliest := time.Time{}
for _, step := range steps {
if len(step.Info.StartedAt) > 0 {
startTime, err := time.Parse(time.RFC3339, step.Info.StartedAt)
if err != nil {
- return nil, fmt.Errorf("failed to parse started at: %w", err)
- }
- if earliestStartTime.IsZero() || startTime.Before(earliestStartTime) {
- earliestStartTime = startTime
+ logger.V(1).Info("failed to parse StartedAt in steps.yaml, skipping", "error", err)
+ return time.Time{}, false
}
- }
-
- // we're looking for the service cluster's step to make a query for backend and frontend
- // forming like this so that we can easily add more steps (like the management cluster) that we want queries for
- if step.Identifier == serviceClusterStepID {
- if step.Details != nil && step.Details.ARM != nil {
- for _, operation := range step.Details.ARM.Operations {
- allClusterNames = append(allClusterNames, locateAllServiceClusters(operation)...)
- }
+ if earliest.IsZero() || startTime.Before(earliest) {
+ earliest = startTime
}
}
}
- if earliestStartTime.IsZero() {
- earliestStartTime = localClock.Now().Add(-6 * time.Hour) // lots longer than default timeouts, but still shorter than forever
+ if earliest.IsZero() {
+ logger.V(1).Info("no StartedAt times found in steps.yaml")
+ return time.Time{}, false
}
+ return earliest, true
+}
- if len(allClusterNames) != 1 {
- return nil, fmt.Errorf("expecting only one service cluster, found %d: %s", len(allClusterNames), strings.Join(allClusterNames, ", "))
+// extractEarliestStartTimeFromTimingMetadata returns the earliest StartTime from timing-metadata files.
+// Returns false on any error (soft failure).
+func extractEarliestStartTimeFromTimingMetadata(logger logr.Logger, timingInputDir string) (time.Time, bool) {
+ timingInfo, err := loadAllTestTimingInfo(timingInputDir)
+ if err != nil {
+ logger.V(1).Info("failed to load timing metadata, skipping", "error", err)
+ return time.Time{}, false
+ }
+ if len(timingInfo) == 0 {
+ logger.V(1).Info("no timing metadata found")
+ return time.Time{}, false
}
- endTime := localClock.Now().Add(1 * time.Hour) // we need to include all cleanup, this is a good bet.
+ earliest := time.Time{}
+ for _, info := range timingInfo {
+ if earliest.IsZero() || info.StartTime.Before(earliest) {
+ earliest = info.StartTime
+ }
+ }
+ return earliest, true
+}
- // Define all components and their log query templates
- components := []struct {
+// getServiceLogLinksFromClusterNamesWithWindow generates service log links for all components
+// using the provided cluster names and time window.
+// Frontend, Backend, Cluster Service, and Maestro logs use the service cluster.
+// Hypershift and ACM logs use the management cluster.
+func getServiceLogLinksFromClusterNamesWithWindow(clusterNames ClusterNames, adxClusterName string, startTime, endTime time.Time) ([]LinkDetails, error) {
+ svcComponents := []struct {
component string
template string
}{
@@ -458,39 +545,38 @@ func getServiceLogLinks(steps []pipeline.NodeInfo) ([]LinkDetails, error) {
{"Frontend Logs", "frontend-logs.kql.tmpl"},
{"Clusters Service Logs", "clusters-service-logs.kql.tmpl"},
{"Maestro Logs", "maestro-logs.kql.tmpl"},
+ }
+
+ mgmtComponents := []struct {
+ component string
+ template string
+ }{
{"Hypershift Logs", "hypershift-logs.kql.tmpl"},
{"ACM Logs", "acm-logs.kql.tmpl"},
}
- // Generate links for each component and cluster
- for _, clusterName := range allClusterNames {
- for _, comp := range components {
- allLinks = append(allLinks, createLinkForTest(comp.component, comp.template, QueryInfo{
- ResourceGroupName: clusterName,
- Database: "ServiceLogs",
- ClusterName: clusterName,
- StartTime: earliestStartTime.Format(time.RFC3339),
- EndTime: endTime.Format(time.RFC3339),
- }))
- }
+ var links []LinkDetails
+ for _, comp := range svcComponents {
+ links = append(links, createLinkForTest(comp.component, comp.template, QueryInfo{
+ ResourceGroupName: clusterNames.Service,
+ Database: "ServiceLogs",
+ ClusterName: clusterNames.Service,
+ ADXClusterName: adxClusterName,
+ StartTime: startTime.Format(time.RFC3339),
+ EndTime: endTime.Format(time.RFC3339),
+ }))
}
- return allLinks, nil
-}
-
-func locateAllServiceClusters(operation pipeline.Operation) []string {
- allClusterNames := []string{}
- for _, currChild := range operation.Children {
- currClusterNames := locateAllServiceClusters(currChild)
- if len(currClusterNames) > 0 {
- allClusterNames = append(allClusterNames, currClusterNames...)
- }
- }
- if operation.Resource == nil {
- return allClusterNames
+ for _, comp := range mgmtComponents {
+ links = append(links, createLinkForTest(comp.component, comp.template, QueryInfo{
+ ResourceGroupName: clusterNames.Management,
+ Database: "ServiceLogs",
+ ClusterName: clusterNames.Management,
+ ADXClusterName: adxClusterName,
+ StartTime: startTime.Format(time.RFC3339),
+ EndTime: endTime.Format(time.RFC3339),
+ }))
}
- if operation.OperationType == "Create" && operation.Resource.ResourceType == "Microsoft.ContainerService/managedClusters" {
- allClusterNames = append(allClusterNames, operation.Resource.Name)
- }
- return allClusterNames
+
+ return links, nil
}
diff --git a/test/cmd/testdata/zz_fixture_TestGeneratedHTMLcustom_link_tools.html b/test/cmd/testdata/zz_fixture_TestGeneratedHTMLcustom_link_tools.html
index 68666c2ef0..ea8c9597cb 100644
--- a/test/cmd/testdata/zz_fixture_TestGeneratedHTMLcustom_link_tools.html
+++ b/test/cmd/testdata/zz_fixture_TestGeneratedHTMLcustom_link_tools.html
@@ -56,13 +56,13 @@
-
+
Hypershift Logs
-
+
ACM Logs