CP-33636: add DCGM GPU metric collection and transformation

.golangci.yaml

@@ -51,7 +51,7 @@ linters:
         - name: atomic
         - name: add-constant
           arguments:
-            - allowFloats: 0.0,0.,1.0,1.,2.0,2.,3.0,3.
+            - allowFloats: 0.0,0.,1.0,1.,2.0,2.,3.0,3.,100.0
               allowInts: 0,1,2,3,10,8,16,32,64,100,128,192,256,512,1024,2048,4096,8192,16384,32768,65536
               allowStrs: '""'
               maxLitCount: "3"

Makefile

@@ -289,7 +289,7 @@ GO_BINARY_DIRS = \
 	$(NULL)
 
 GO_COMMAND_PACKAGE_DIRS = \
-	$(foreach parent_dir,$(GO_BINARY_DIRS),$(foreach src_dir,$(wildcard $(parent_dir)/*/),$(patsubst %/,%,$(src_dir)))) \
+	$(patsubst %/,%,$(filter %/,$(foreach parent_dir,$(GO_BINARY_DIRS),$(wildcard $(parent_dir)/*/)))) \
 	$(NULL)
 
 GO_BINARIES = \

app/domain/metric_collector.go

@@ -30,6 +30,7 @@ import (
 	"github.com/rs/zerolog/log"
 
 	config "github.com/cloudzero/cloudzero-agent/app/config/gator"
+	"github.com/cloudzero/cloudzero-agent/app/domain/transform"
 	"github.com/cloudzero/cloudzero-agent/app/types"
 )
 
@@ -113,6 +114,9 @@ type MetricCollector struct {
 	// filter implements metric classification logic to separate cost from observability metrics.
 	filter *MetricFilter
 
+	// transformer handles vendor-specific metric transformation (e.g., DCGM GPU metrics).
+	transformer types.MetricTransformer
+
 	// clock provides time abstraction for testing and consistent timestamping.
 	clock types.TimeProvider
 
@@ -144,6 +148,7 @@ func NewMetricCollector(s *config.Settings, clock types.TimeProvider, costStore
 		costStore:          costStore,
 		observabilityStore: observabilityStore,
 		filter:             filter,
+		transformer:        transform.NewMetricTransformer(),
 		clock:              clock,
 		cancelFunc:         cancel,
 	}
@@ -192,6 +197,27 @@ func (d *MetricCollector) PutMetrics(ctx context.Context, contentType, encodingT
 		return nil, fmt.Errorf("unsupported content type: %s", contentType)
 	}
 
+	// Log complete DCGM metrics for debugging GPU transformation
+	for _, metric := range metrics {
+		if strings.HasPrefix(metric.MetricName, "DCGM_FI_DEV_") {
+			log.Ctx(ctx).Info().
+				Str("metricName", metric.MetricName).
+				Str("value", metric.Value).
+				Str("nodeName", metric.NodeName).
+				Interface("labels", metric.Labels).
+				Time("timestamp", metric.TimeStamp).
+				Str("clusterName", metric.ClusterName).
+				Str("cloudAccountID", metric.CloudAccountID).
+				Msg("DCGM metric received")
+		}
+	}
+
+	// Transform vendor-specific metrics (e.g., DCGM GPU metrics) before filtering
+	metrics, err = d.transformer.Transform(ctx, metrics)
+	if err != nil {
+		return stats, fmt.Errorf("failed to transform metrics: %w", err)
+	}
+
 	costMetrics, observabilityMetrics, droppedMetrics := d.filter.Filter(metrics)
 
 	metricsReceived.WithLabelValues().Add(float64(len(metrics)))

app/domain/transform/catalog/catalog.go

@@ -0,0 +1,65 @@
+// SPDX-FileCopyrightText: Copyright (c) 2016-2025, CloudZero, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+// Package catalog provides a catalog-based metric transformer that routes
+// metrics to registered specialized transformers.
+//
+// The catalog transformer orchestrates multiple specialized transformers to
+// provide automatic routing based on metric characteristics.
+package catalog
+
+import (
+	"context"
+
+	"github.com/cloudzero/cloudzero-agent/app/types"
+)
+
+// Transformer implements types.MetricTransformer using a catalog of specialized
+// transformers.
+//
+// Each transformer in the catalog processes all metrics sequentially.
+// Transformers identify which metrics they can handle and transform those while
+// passing through others unchanged.
+type Transformer struct {
+	transformers []types.MetricTransformer
+}
+
+// NewTransformer creates a new catalog transformer with the provided
+// specialized transformers.
+//
+// Transformers are applied sequentially - each transformer receives all metrics
+// and decides which ones to transform based on implementation-specific logic
+// (e.g., metric name patterns).
+func NewTransformer(transformers ...types.MetricTransformer) *Transformer {
+	return &Transformer{
+		transformers: transformers,
+	}
+}
+
+// Transform processes metrics by routing them sequentially through specialized
+// transformers.
+//
+// Processing flow:
+//  1. Pass metrics through first transformer
+//  2. Pass results through second transformer
+//  3. Continue until all transformers have processed the metrics
+//
+// This implements the types.MetricTransformer interface.
+func (t *Transformer) Transform(ctx context.Context, metrics []types.Metric) ([]types.Metric, error) {
+	if len(t.transformers) == 0 {
+		return metrics, nil
+	}
+
+	// Process through each transformer in sequence
+	result := metrics
+	var err error
+
+	for _, transformer := range t.transformers {
+		result, err = transformer.Transform(ctx, result)
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	return result, nil
+}