|
8 | 8 | "fmt" |
9 | 9 | "time" |
10 | 10 |
|
11 | | - "github.com/NVIDIA/KAI-scheduler/pkg/common/constants" |
| 11 | + commonconstants "github.com/NVIDIA/KAI-scheduler/pkg/common/constants" |
12 | 12 | "github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/common_info" |
13 | 13 | "github.com/NVIDIA/KAI-scheduler/pkg/scheduler/api/queue_info" |
14 | 14 | "github.com/NVIDIA/KAI-scheduler/pkg/scheduler/cache/usagedb/api" |
@@ -76,40 +76,17 @@ func (p *PrometheusClient) GetResourceUsage() (*queue_info.ClusterUsage, error) |
76 | 76 |
|
77 | 77 | usage := queue_info.NewClusterUsage() |
78 | 78 |
|
79 | | - // get gpu usage per queue |
80 | | - gpuUsage, err := p.queryResourceUsage(ctx, p.allocationMetricsMap["gpu"]) |
81 | | - if err != nil { |
82 | | - return nil, fmt.Errorf("error querying gpu capacity and usage: %v", err) |
83 | | - } |
84 | | - for queueID, queueGPUUsage := range gpuUsage { |
85 | | - if _, exists := usage.Queues[queueID]; !exists { |
86 | | - usage.Queues[queueID] = queue_info.QueueUsage{} |
| 79 | + for _, resource := range []v1.ResourceName{commonconstants.GpuResource, v1.ResourceCPU, v1.ResourceMemory} { |
| 80 | + resourceUsage, err := p.queryResourceUsage(ctx, p.allocationMetricsMap[string(resource)]) |
| 81 | + if err != nil { |
| 82 | + return nil, fmt.Errorf("error querying %s and usage: %v", resource, err) |
87 | 83 | } |
88 | | - usage.Queues[queueID][constants.GpuResource] = queueGPUUsage |
89 | | - } |
90 | | - |
91 | | - // get cpu usage per queue |
92 | | - cpuUsage, err := p.queryResourceUsage(ctx, p.allocationMetricsMap["cpu"]) |
93 | | - if err != nil { |
94 | | - return nil, fmt.Errorf("error querying cpu capacity and usage: %v", err) |
95 | | - } |
96 | | - for queueID, queueCPUUsage := range cpuUsage { |
97 | | - if _, exists := usage.Queues[queueID]; !exists { |
98 | | - usage.Queues[queueID] = queue_info.QueueUsage{} |
99 | | - } |
100 | | - usage.Queues[queueID][v1.ResourceCPU] = queueCPUUsage |
101 | | - } |
102 | | - |
103 | | - // get memory usage per queue |
104 | | - memoryUsage, err := p.queryResourceUsage(ctx, p.allocationMetricsMap["memory"]) |
105 | | - if err != nil { |
106 | | - return nil, fmt.Errorf("error querying memory capacity and usage: %v", err) |
107 | | - } |
108 | | - for queueID, queueMemoryUsage := range memoryUsage { |
109 | | - if _, exists := usage.Queues[queueID]; !exists { |
110 | | - usage.Queues[queueID] = queue_info.QueueUsage{} |
| 84 | + for queueID, queueResourceUsage := range resourceUsage { |
| 85 | + if _, exists := usage.Queues[queueID]; !exists { |
| 86 | + usage.Queues[queueID] = queue_info.QueueUsage{} |
| 87 | + } |
| 88 | + usage.Queues[queueID][resource] = queueResourceUsage |
111 | 89 | } |
112 | | - usage.Queues[queueID][v1.ResourceMemory] = queueMemoryUsage |
113 | 90 | } |
114 | 91 |
|
115 | 92 | return usage, nil |
|
0 commit comments