feat: keyperf collector templates (#3194)

* feat: keyperf collector templates

* feat: address review comments

* feat: add slogx error

cmd/collectors/keyperf/keyperf.go

@@ -140,7 +140,14 @@ func (kp *KeyPerf) loadParamInt(name string, defaultValue int) int {
 }
 
 func (kp *KeyPerf) buildCounters() {
-	for k := range kp.Prop.Metrics {
+	staticCounterDef, err := loadStaticCounterDefinitions(kp.Prop.Object, "cmd/collectors/keyperf/static_counter_definitions.yaml", kp.Logger)
+	if err != nil {
+		// It's acceptable to continue even if there are errors, as the remaining counters will still be processed.
+		// Any counters that require counter metadata will be skipped.
+		kp.Logger.Error("Failed to load static counter definitions", slogx.Err(err))
+	}
+
+	for k, v := range kp.Prop.Metrics {
 		if _, exists := kp.perfProp.counterInfo[k]; !exists {
 			var ctr *counter
 
@@ -170,6 +177,36 @@ func (kp *KeyPerf) buildCounters() {
 					counterType: "delta",
 					unit:        "sec",
 				}
+			default:
+				// look up metric in staticCounterDef
+				if counterDef, exists := staticCounterDef.CounterDefinitions[v.Name]; exists {
+					ctr = &counter{
+						name:        k,
+						counterType: counterDef.Type,
+						denominator: counterDef.BaseCounter,
+					}
+					if counterDef.BaseCounter != "" {
+						// Ensure denominator exists in counterInfo
+						if _, denomExists := kp.perfProp.counterInfo[counterDef.BaseCounter]; !denomExists {
+							var baseCounterType string
+							if baseCounterDef, baseCounterExists := staticCounterDef.CounterDefinitions[counterDef.BaseCounter]; baseCounterExists {
+								baseCounterType = baseCounterDef.Type
+							}
+							if baseCounterType != "" {
+								kp.perfProp.counterInfo[counterDef.BaseCounter] = &counter{
+									name:        counterDef.BaseCounter,
+									counterType: staticCounterDef.CounterDefinitions[counterDef.BaseCounter].Type,
+								}
+								if _, dExists := kp.Prop.Metrics[counterDef.BaseCounter]; !dExists {
+									m := &rest.Metric{Label: "", Name: counterDef.BaseCounter, MetricType: "", Exportable: false}
+									kp.Prop.Metrics[counterDef.BaseCounter] = m
+								}
+							}
+						}
+					}
+				} else {
+					slog.Warn("Skipping metric due to unknown metricType", slog.String("name", k), slog.String("metricType", v.MetricType))
+				}
 			}
 
 			if ctr != nil {
@@ -304,7 +341,8 @@ func (kp *KeyPerf) pollData(
 				orderedDenominatorKeys = append(orderedDenominatorKeys, key)
 			}
 		} else {
-			kp.Logger.Warn("Counter is missing or unable to parse", slog.String("counter", metric.GetName()))
+			kp.Logger.Error("Counter is missing or unable to parse", slog.String("counter", metric.GetName()))
+			metric.SetExportable(false)
 		}
 	}
 
@@ -375,6 +413,8 @@ func (kp *KeyPerf) pollData(
 				slog.String("denominator", counter.denominator),
 				slog.Int("instIndex", instIndex),
 			)
+			skips = curMat.Skip(key)
+			totalSkips += skips
 			continue
 		}
 

cmd/collectors/keyperf/static_counter_definitions.yaml

@@ -0,0 +1,8 @@
+objects:
+  node:
+    counter_definitions:
+      - name: statistics.processor_utilization_raw
+        type: percent
+        base_counter: statistics.processor_utilization_base
+      - name: statistics.processor_utilization_base
+        type: delta

cmd/collectors/keyperf/templating.go

@@ -0,0 +1,65 @@
+package keyperf
+
+import (
+	"gopkg.in/yaml.v3"
+	"log/slog"
+	"os"
+)
+
+type CounterDefinition struct {
+	Name        string `yaml:"name"`
+	Type        string `yaml:"type"`
+	BaseCounter string `yaml:"base_counter,omitempty"`
+}
+
+type ObjectCounters struct {
+	CounterDefinitions map[string]CounterDefinition `yaml:"-"`
+}
+
+type Object struct {
+	CounterDefinitions []CounterDefinition `yaml:"counter_definitions"`
+}
+
+type StaticCounterDefinitions struct {
+	Objects map[string]Object `yaml:"objects"`
+}
+
+func loadStaticCounterDefinitions(object string, filePath string, logger *slog.Logger) (ObjectCounters, error) {
+	var staticDefinitions StaticCounterDefinitions
+	var objectCounters ObjectCounters
+
+	data, err := os.ReadFile(filePath)
+	if err != nil {
+		return objectCounters, err
+	}
+
+	err = yaml.Unmarshal(data, &staticDefinitions)
+	if err != nil {
+		return objectCounters, err
+	}
+
+	if obj, exists := staticDefinitions.Objects[object]; exists {
+		allCounterDefs := make(map[string]CounterDefinition)
+		for _, def := range obj.CounterDefinitions {
+			allCounterDefs[def.Name] = def
+		}
+
+		objectCounters.CounterDefinitions = make(map[string]CounterDefinition)
+		for _, def := range obj.CounterDefinitions {
+			if def.Type == "" {
+				logger.Error("Missing type in counter definition", slog.String("filePath", filePath), slog.String("counterName", def.Name))
+				continue
+			}
+			if def.BaseCounter != "" {
+				if _, baseCounterExists := allCounterDefs[def.BaseCounter]; !baseCounterExists {
+					logger.Error("Base counter definition not found", slog.String("filePath", filePath), slog.String("counterName", def.Name), slog.String("baseCounter", def.BaseCounter))
+					continue
+				}
+			}
+			objectCounters.CounterDefinitions[def.Name] = def
+		}
+		return objectCounters, nil
+	}
+
+	return objectCounters, nil
+}

cmd/collectors/rest/rest.go

@@ -630,6 +630,8 @@ func (r *Rest) HandleResults(mat *matrix.Matrix, result []gjson.Result, prop *pr
 						slogx.Err(err),
 						slog.String("name", metric.Name),
 					)
+				} else {
+					metr.SetExportable(metric.Exportable)
 				}
 			}
 			f := instanceData.Get(metric.Name)

cmd/tools/template/template_test.go

@@ -19,7 +19,7 @@ import (
 
 const toConf = "../../../conf"
 
-var allTemplatesButEms = []string{"rest", "restperf", "storagegrid", "zapi", "zapiperf"}
+var allTemplatesButEms = []string{"rest", "restperf", "storagegrid", "zapi", "zapiperf", "keyperf"}
 
 // validates each template file name:
 //   - ends with yaml

conf/keyperf/9.15.0/aggr.yaml

@@ -0,0 +1,30 @@
+name:                     Aggregate
+query:                    api/storage/aggregates
+object:                   aggr
+
+counters:
+  - ^^uuid                                => uuid
+  - ^name                                 => aggr
+  - ^node.name                            => node
+  - ^statistics.status                    => status
+  - statistics.iops_raw.other             => other_ops
+  - statistics.iops_raw.read              => read_ops
+  - statistics.iops_raw.total             => total_ops
+  - statistics.iops_raw.write             => write_ops
+  - statistics.latency_raw.other          => other_latency
+  - statistics.latency_raw.read           => read_latency
+  - statistics.latency_raw.total          => total_latency
+  - statistics.latency_raw.write          => write_latency
+  - statistics.throughput_raw.other       => other_data
+  - statistics.throughput_raw.read        => read_data
+  - statistics.throughput_raw.total       => total_data
+  - statistics.throughput_raw.write       => write_data
+  - statistics.timestamp(timestamp)       => timestamp
+  - hidden_fields:
+      - statistics
+
+
+export_options:
+  instance_keys:
+    - aggr
+    - node

conf/keyperf/9.15.0/cifs_vserver.yaml

@@ -0,0 +1,27 @@
+name:                     CIFSvserver
+query:                    api/protocols/cifs/services
+object:                   svm_cifs
+
+counters:
+  - ^^svm.uuid                            => uuid
+  - ^statistics.status                    => status
+  - ^svm.name                             => svm
+  - statistics.iops_raw.other             => other_ops
+  - statistics.iops_raw.read              => read_ops
+  - statistics.iops_raw.total             => total_ops
+  - statistics.iops_raw.write             => write_ops
+  - statistics.latency_raw.other          => other_latency
+  - statistics.latency_raw.read           => read_latency
+  - statistics.latency_raw.total          => total_latency
+  - statistics.latency_raw.write          => write_latency
+  - statistics.throughput_raw.other       => other_data
+  - statistics.throughput_raw.read        => read_data
+  - statistics.throughput_raw.write       => write_data
+  - statistics.timestamp(timestamp)       => timestamp
+  - hidden_fields:
+      - statistics
+
+
+export_options:
+  instance_keys:
+    - svm

conf/keyperf/9.15.0/cluster.yaml

@@ -0,0 +1,27 @@
+name:                     Cluster
+query:                    api/cluster
+object:                   cluster
+
+counters:
+  - ^^uuid                                => uuid
+  - ^statistics.status                    => status
+  - statistics.iops_raw.other             => other_ops
+  - statistics.iops_raw.read              => read_ops
+  - statistics.iops_raw.total             => total_ops
+  - statistics.iops_raw.write             => write_ops
+  - statistics.latency_raw.other          => other_latency
+  - statistics.latency_raw.read           => read_latency
+  - statistics.latency_raw.total          => total_latency
+  - statistics.latency_raw.write          => write_latency
+  - statistics.throughput_raw.other       => other_data
+  - statistics.throughput_raw.read        => read_data
+  - statistics.throughput_raw.total       => total_data
+  - statistics.throughput_raw.write       => write_data
+  - statistics.timestamp(timestamp)       => timestamp
+  - hidden_fields:
+      - statistics
+
+
+export_options:
+  instance_keys:
+    - uuid

conf/keyperf/9.15.0/iscsi_svm.yaml

@@ -0,0 +1,27 @@
+name:                     ISCSISvm
+query:                    api/protocols/san/iscsi/services
+object:                   iscsi_svm
+
+counters:
+  - ^^svm.name                            => svm
+  - ^statistics.status                    => status
+  - ^target.name                          => target
+  - statistics.iops_raw.other             => other_ops
+  - statistics.iops_raw.read              => read_ops
+  - statistics.iops_raw.total             => total_ops
+  - statistics.iops_raw.write             => write_ops
+  - statistics.latency_raw.other          => other_latency
+  - statistics.latency_raw.read           => read_latency
+  - statistics.latency_raw.total          => total_latency
+  - statistics.latency_raw.write          => write_latency
+  - statistics.throughput_raw.read        => read_data
+  - statistics.throughput_raw.total       => total_data
+  - statistics.throughput_raw.write       => write_data
+  - statistics.timestamp(timestamp)       => timestamp
+  - hidden_fields:
+      - statistics
+
+export_options:
+  instance_keys:
+    - svm
+    - target

conf/keyperf/9.15.0/lif.yaml

@@ -0,0 +1,25 @@
+name:                     LIF
+query:                    api/network/ip/interfaces
+object:                   lif
+
+counters:
+  - ^^uuid
+  - ^location.node.name                   => node
+  - ^location.port.name                   => port
+  - ^name                                 => lif
+  - ^statistics.status                    => status
+  - ^svm.name                             => svm
+  - statistics.throughput_raw.read        => sent_data
+  - statistics.throughput_raw.total       => total_data
+  - statistics.throughput_raw.write       => recv_data
+  - statistics.timestamp(timestamp)       => timestamp
+  - hidden_fields:
+      - statistics
+
+
+export_options:
+  instance_keys:
+    - lif
+    - node
+    - port
+    - svm

conf/keyperf/9.15.0/lun.yaml

@@ -0,0 +1,38 @@
+name:                     Lun
+query:                    api/storage/luns
+object:                   lun
+
+counters:
+  - ^^uuid                                => uuid
+  - ^location.volume.name                 => volume
+  - ^name                                 => path
+  - ^statistics.status                    => status
+  - ^svm.name                             => svm
+  - statistics.iops_raw.other             => other_ops
+  - statistics.iops_raw.read              => read_ops
+  - statistics.iops_raw.total             => total_ops
+  - statistics.iops_raw.write             => write_ops
+  - statistics.latency_raw.other          => other_latency
+  - statistics.latency_raw.read           => avg_read_latency
+  - statistics.latency_raw.total          => total_latency
+  - statistics.latency_raw.write          => avg_write_latency
+  - statistics.throughput_raw.other       => other_data
+  - statistics.throughput_raw.read        => read_data
+  - statistics.throughput_raw.total       => total_data
+  - statistics.throughput_raw.write       => write_data
+  - statistics.timestamp(timestamp)       => timestamp
+  - hidden_fields:
+      - statistics
+
+
+plugins:
+  LabelAgent:
+    split_regex:
+      - path `^/[^/]+/([^/]+)(?:/.*?|)/([^/]+)$` volume,lun
+      - path `^([^/]+)$` lun
+
+export_options:
+  instance_keys:
+    - lun
+    - svm
+    - volume

conf/keyperf/9.15.0/namespace.yaml

@@ -0,0 +1,36 @@
+name:                     Namespace
+query:                    api/storage/namespaces
+object:                   namespace
+
+counters:
+  - ^^uuid                                => uuid
+  - ^name                                 => path
+  - ^statistics.status                    => status
+  - ^svm.name                             => svm
+  - statistics.iops_raw.other             => other_ops
+  - statistics.iops_raw.read              => read_ops
+  - statistics.iops_raw.total             => total_ops
+  - statistics.iops_raw.write             => write_ops
+  - statistics.latency_raw.other          => avg_other_latency
+  - statistics.latency_raw.read           => avg_read_latency
+  - statistics.latency_raw.total          => avg_total_latency
+  - statistics.latency_raw.write          => avg_write_latency
+  - statistics.throughput_raw.other       => other_data
+  - statistics.throughput_raw.read        => read_data
+  - statistics.throughput_raw.total       => total_data
+  - statistics.throughput_raw.write       => write_data
+  - statistics.timestamp(timestamp)       => timestamp
+  - hidden_fields:
+      - statistics
+
+plugins:
+  LabelAgent:
+    split:
+      - path `/` ,,volume,namespace
+
+export_options:
+  instance_keys:
+    - namespace
+    - path
+    - svm
+    - volume