feat: added slo duplicates check during validate (#1)

cmd/sloth/commands/validate.go

@@ -27,6 +27,7 @@ type validateCommand struct {
 	sliPluginsPaths      []string
 	sloPeriodWindowsPath string
 	sloPeriod            string
+	ignoreSloDuplicates  bool
 }
 
 // NewValidateCommand returns the validate command.
@@ -40,7 +41,7 @@ func NewValidateCommand(app *kingpin.Application) Command {
 	cmd.Flag("sli-plugins-path", "The path to SLI plugins (can be repeated), if not set it disable plugins support.").Short('p').StringsVar(&c.sliPluginsPaths)
 	cmd.Flag("slo-period-windows-path", "The directory path to custom SLO period windows catalog (replaces default ones).").StringVar(&c.sloPeriodWindowsPath)
 	cmd.Flag("default-slo-period", "The default SLO period windows to be used for the SLOs.").Default("30d").StringVar(&c.sloPeriod)
-
+	cmd.Flag("ignore-slo-duplicates", "Flag to ignore SLO duplicates in specs (service and name used as an SLO/SLI identifier).").Default("false").BoolVar(&c.ignoreSloDuplicates)
 	return c
 }
 
@@ -115,6 +116,7 @@ func (v validateCommand) Run(ctx context.Context, config RootConfig) error {
 	// For every file load the data and start the validation process:
 	validations := []*fileValidation{}
 	totalValidations := 0
+	sloIDs := make(map[string]string, 0)
 	for _, input := range sloPaths {
 		// Get SLO spec data.
 		slxData, err := os.ReadFile(input)
@@ -142,11 +144,21 @@ func (v validateCommand) Run(ctx context.Context, config RootConfig) error {
 			// Match the spec type to know how to validate.
 			switch {
 			case promYAMLLoader.IsSpecType(ctx, dataB):
-				slos, promErr := promYAMLLoader.LoadSpec(ctx, dataB)
+				sloGroup, promErr := promYAMLLoader.LoadSpec(ctx, dataB)
 				if promErr == nil {
-					err := gen.GeneratePrometheus(ctx, *slos, io.Discard)
+					if !v.ignoreSloDuplicates {
+						for _, slo := range sloGroup.SLOs {
+							if sloFile, exists := sloIDs[slo.ID]; !exists {
+								sloIDs[slo.ID] = validation.File
+							} else {
+								validation.Errs = append(validation.Errs, fmt.Errorf("SLO duplicated. SLO{service=%s, name=%s}, ID=%s already exists in a file: %s", slo.Service, slo.Name, slo.ID, sloFile))
+							}
+						}
+					}
+
+					err := gen.GeneratePrometheus(ctx, *sloGroup, io.Discard)
 					if err != nil {
-						validation.Errs = []error{fmt.Errorf("Could not generate Prometheus format rules: %w", err)}
+						validation.Errs = append(validation.Errs, fmt.Errorf("Could not generate Prometheus format rules: %w", err))
 					}
 					continue
 				}
@@ -156,21 +168,41 @@ func (v validateCommand) Run(ctx context.Context, config RootConfig) error {
 			case kubeYAMLLoader.IsSpecType(ctx, dataB):
 				sloGroup, k8sErr := kubeYAMLLoader.LoadSpec(ctx, dataB)
 				if k8sErr == nil {
+					if !v.ignoreSloDuplicates {
+						for _, slo := range sloGroup.SLOs {
+							if sloFile, exists := sloIDs[slo.ID]; !exists {
+								sloIDs[slo.ID] = validation.File
+							} else {
+								validation.Errs = append(validation.Errs, fmt.Errorf("SLO duplicated. SLO{service=%s, name=%s}, ID=%s already exists in a file: %s", slo.Service, slo.Name, slo.ID, sloFile))
+							}
+						}
+					}
+
 					err := gen.GenerateKubernetes(ctx, *sloGroup, io.Discard)
 					if err != nil {
-						validation.Errs = []error{fmt.Errorf("could not generate Kubernetes format rules: %w", err)}
+						validation.Errs = append(validation.Errs, fmt.Errorf("could not generate Kubernetes format rules: %w", err))
 					}
 					continue
 				}
 
 				validation.Errs = []error{fmt.Errorf("Tried loading Kubernetes prometheus SLOs spec, it couldn't: %w", k8sErr)}
 
 			case openSLOYAMLLoader.IsSpecType(ctx, dataB):
-				slos, openSLOErr := openSLOYAMLLoader.LoadSpec(ctx, dataB)
+				sloGroup, openSLOErr := openSLOYAMLLoader.LoadSpec(ctx, dataB)
 				if openSLOErr == nil {
-					err := gen.GenerateOpenSLO(ctx, *slos, io.Discard)
+					if !v.ignoreSloDuplicates {
+						for _, slo := range sloGroup.SLOs {
+							if sloFile, exists := sloIDs[slo.ID]; !exists {
+								sloIDs[slo.ID] = validation.File
+							} else {
+								validation.Errs = append(validation.Errs, fmt.Errorf("SLO duplicated. SLO{service=%s, name=%s}, ID=%s already exists in a file: %s", slo.Service, slo.Name, slo.ID, sloFile))
+							}
+						}
+					}
+
+					err := gen.GenerateOpenSLO(ctx, *sloGroup, io.Discard)
 					if err != nil {
-						validation.Errs = []error{fmt.Errorf("Could not generate OpenSLO format rules: %w", err)}
+						validation.Errs = append(validation.Errs, fmt.Errorf("Could not generate OpenSLO format rules: %w", err))
 					}
 					continue
 				}

test/integration/prometheus/testdata/validate/bad/duplicates/bad-prom-multi-duplicates.yaml

@@ -0,0 +1,87 @@
+---
+version: "prometheus/v1"
+service: "svc01"
+labels:
+  global01k1: global01v1
+slos:
+  - name: "slo1"
+    objective: 99.9
+    description: "This is SLO 01."
+    labels:
+      global02k1: global02v1
+    sli:
+      events:
+        error_query: sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+        total_query: sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+    alerting:
+      name: myServiceAlert
+      labels:
+        alert01k1: "alert01v1"
+      annotations:
+        alert02k1: "alert02k2"
+      pageAlert:
+        labels:
+          alert03k1: "alert03v1"
+      ticketAlert:
+        labels:
+          alert04k1: "alert04v1"
+  - name: "slo02"
+    objective: 95
+    description: "This is SLO 02."
+    labels:
+      global03k1: global03v1
+    sli:
+      raw:
+        error_ratio_query: |
+          sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+          /
+          sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+    alerting:
+      page_alert:
+        disable: true
+      ticket_alert:
+        disable: true
+
+---
+version: "prometheus/v1"
+service: "svc01"
+labels:
+  global01k1: global01v1
+slos:
+  - name: "slo1" # duplicate
+    objective: 99.9
+    description: "This is SLO 01."
+    labels:
+      global02k1: global02v1
+    sli:
+      events:
+        error_query: sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+        total_query: sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+    alerting:
+      name: myServiceAlert
+      labels:
+        alert01k1: "alert01v1"
+      annotations:
+        alert02k1: "alert02k2"
+      pageAlert:
+        labels:
+          alert03k1: "alert03v1"
+      ticketAlert:
+        labels:
+          alert04k1: "alert04v1"
+  - name: "slo02" # duplicate
+    objective: 95
+    description: "This is SLO 02."
+    labels:
+      global03k1: global03v1
+    sli:
+      raw:
+        error_ratio_query: |
+          sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+          /
+          sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+    alerting:
+      page_alert:
+        disable: true
+      ticket_alert:
+        disable: true

test/integration/prometheus/testdata/validate/bad/duplicates/k8s/bad-k8s-1-duplicate.yaml

@@ -0,0 +1,47 @@
+apiVersion: sloth.slok.dev/v1
+kind: PrometheusServiceLevel
+metadata:
+  name: svc
+  namespace: test-ns
+spec:
+  service: "svc01"
+  labels:
+    global01k1: global01v1
+  slos:
+    - name: "slo1"
+      objective: 99.9
+      description: "This is SLO 01."
+      labels:
+        global02k1: global02v1
+      sli:
+        events:
+          errorQuery: sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+          totalQuery: sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+      alerting:
+        name: myServiceAlert
+        labels:
+          alert01k1: "alert01v1"
+        annotations:
+          alert02k1: "alert02k2"
+        pageAlert:
+          labels:
+            alert03k1: "alert03v1"
+        ticketAlert:
+          labels:
+            alert04k1: "alert04v1"
+    - name: "slo02"
+      objective: 95
+      description: "This is SLO 02."
+      labels:
+        global03k1: global03v1
+      sli:
+        raw:
+          errorRatioQuery: |
+            sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+            /
+            sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+      alerting:
+        pageAlert:
+          disable: true
+        ticketAlert:
+          disable: true

test/integration/prometheus/testdata/validate/bad/duplicates/k8s/bad-k8s-1-original.yaml

@@ -0,0 +1,47 @@
+apiVersion: sloth.slok.dev/v1
+kind: PrometheusServiceLevel
+metadata:
+  name: svc
+  namespace: test-ns
+spec:
+  service: "svc01"
+  labels:
+    global01k1: global01v1
+  slos:
+    - name: "slo1"
+      objective: 99.9
+      description: "This is SLO 01."
+      labels:
+        global02k1: global02v1
+      sli:
+        events:
+          errorQuery: sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+          totalQuery: sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+      alerting:
+        name: myServiceAlert
+        labels:
+          alert01k1: "alert01v1"
+        annotations:
+          alert02k1: "alert02k2"
+        pageAlert:
+          labels:
+            alert03k1: "alert03v1"
+        ticketAlert:
+          labels:
+            alert04k1: "alert04v1"
+    - name: "slo02"
+      objective: 95
+      description: "This is SLO 02."
+      labels:
+        global03k1: global03v1
+      sli:
+        raw:
+          errorRatioQuery: |
+            sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+            /
+            sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+      alerting:
+        pageAlert:
+          disable: true
+        ticketAlert:
+          disable: true

test/integration/prometheus/testdata/validate/bad/duplicates/openslo/bad-openslo-1-duplicate.yaml

@@ -0,0 +1,23 @@
+apiVersion: openslo/v1alpha
+kind: SLO
+metadata:
+  name: slo1
+  displayName: Integration test SLO1
+spec:
+  service: svc01
+  description: "this is SLO1."
+  budgetingMethod: Occurrences
+  objectives:
+    - ratioMetrics:
+        good:
+          source: prometheus
+          queryType: promql
+          query: sum(rate(http_request_duration_seconds_count{job="myservice",code!~"(5..|429)"}[{{.window}}]))
+        total:
+          source: prometheus
+          queryType: promql
+          query: sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+      target: 0.999
+  timeWindows:
+    - count: 30
+      unit: Day

test/integration/prometheus/testdata/validate/bad/duplicates/openslo/bad-openslo-1-original.yaml

@@ -0,0 +1,23 @@
+apiVersion: openslo/v1alpha
+kind: SLO
+metadata:
+  name: slo1
+  displayName: Integration test SLO1
+spec:
+  service: svc01
+  description: "this is SLO1."
+  budgetingMethod: Occurrences
+  objectives:
+    - ratioMetrics:
+        good:
+          source: prometheus
+          queryType: promql
+          query: sum(rate(http_request_duration_seconds_count{job="myservice",code!~"(5..|429)"}[{{.window}}]))
+        total:
+          source: prometheus
+          queryType: promql
+          query: sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+      target: 0.999
+  timeWindows:
+    - count: 30
+      unit: Day

test/integration/prometheus/testdata/validate/bad/duplicates/prom/bad-prom-1-duplicate.yaml

@@ -0,0 +1,42 @@
+version: "prometheus/v1"
+service: "svc01"
+labels:
+  global01k1: global01v1
+slos:
+  - name: "slo1"
+    objective: 99.9
+    description: "This is SLO 01."
+    labels:
+      global02k1: global02v1
+    sli:
+      events:
+        error_query: sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+        total_query: sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+    alerting:
+      name: myServiceAlert
+      labels:
+        alert01k1: "alert01v1"
+      annotations:
+        alert02k1: "alert02k2"
+      pageAlert:
+        labels:
+          alert03k1: "alert03v1"
+      ticketAlert:
+        labels:
+          alert04k1: "alert04v1"
+  - name: "slo02"
+    objective: 95
+    description: "This is SLO 02."
+    labels:
+      global03k1: global03v1
+    sli:
+      raw:
+        error_ratio_query: |
+          sum(rate(http_request_duration_seconds_count{job="myservice",code=~"(5..|429)"}[{{.window}}]))
+          /
+          sum(rate(http_request_duration_seconds_count{job="myservice"}[{{.window}}]))
+    alerting:
+      page_alert:
+        disable: true
+      ticket_alert:
+        disable: true