NVIDIA GPU
Public
Commits
86
86
GPU uint `json:"device"`87
87
Memory uint `json:"memory"`88
88
}89
89
90
90
// ECCMode returns current and pending status of ECC.91
91
type ECCMode struct {92
92
Current bool `json:"current"`93
93
Pending bool `json:"pending"`94
94
}95
95
96
-
// New creates a new handler with initialized clients for system and tcp calls.97
-
func New(nvmlRunner nvml.Runner) *Handler {98
-
return &Handler{99
-
// negative indicates no limit100
-
concurrentDeviceDiscoverys: -1,101
-
nvmlRunner: nvmlRunner,102
-
deviceCacheMux: &sync.Mutex{},103
-
deviceCache: make(map[string]nvml.Device),104
-
}105
-
}106
-
107
96
// GetNVMLVersion returns local NVML version.108
97
func (h *Handler) GetNVMLVersion(_ context.Context, _ map[string]string, _ ...string) (any, error) {109
98
version, err := h.nvmlRunner.GetNVMLVersion()110
99
if err != nil {111
100
return "", errs.Wrap(err, "failed to get NVML version")112
101
}113
102
114
103
return version, nil115
104
}116
105
459
448
}460
449
461
450
ecc := ECCErrors{462
451
Corrected: corrected,463
452
Uncorrected: uncorrected,464
453
}465
454
466
455
return ecc, nil467
456
}468
457
469
-
// WithJSONResponse wraps a handler function, marshaling its response470
-
// to a JSON object and returning it as string.471
-
func WithJSONResponse(handler HandlerFunc) HandlerFunc {472
-
return func(473
-
ctx context.Context, metricParams map[string]string, extraParams ...string,474
-
) (any, error) {475
-
res, err := handler(ctx, metricParams, extraParams...)476
-
if err != nil {477
-
return nil, errs.Wrap(err, "failed to receive the result")478
-
}479
-
480
-
jsonRes, err := json.Marshal(res)481
-
if err != nil {482
-
return nil, errs.Wrap(err, "failed to marshal result to JSON")483
-
}484
-
485
-
return string(jsonRes), nil486
-
}487
-
}488
-
489
458
// GetPCIeThroughput retrieves the PCIe receive and transmit throughput for the NVIDIA device in KB/s.490
459
func (h *Handler) GetPCIeThroughput(_ context.Context, metricParams map[string]string, _ ...string) (any, error) {491
460
uuid, ok := metricParams[params.DeviceUUIDParamName]492
461
if !ok {493
462
return nil, errs.New("failed to find param for UUID")494
463
}495
464
496
465
device, err := h.getDeviceByUUID(uuid)497
466
if err != nil {498
467
return nil, errs.Wrap(err, "failed to get device by UUID")706
675
}707
676
708
677
mode := ECCMode{709
678
Current: current,710
679
Pending: pending,711
680
}712
681
713
682
return mode, nil714
683
}715
684
685
+
// WithJSONResponse wraps a handler function, marshaling its response686
+
// to a JSON object and returning it as string.687
+
func WithJSONResponse(handler HandlerFunc) HandlerFunc {688
+
return func(689
+
ctx context.Context, metricParams map[string]string, extraParams ...string,690
+
) (any, error) {691
+
res, err := handler(ctx, metricParams, extraParams...)692
+
if err != nil {693
+
return nil, errs.Wrap(err, "failed to receive the result")694
+
}695
+
696
+
jsonRes, err := json.Marshal(res)697
+
if err != nil {698
+
return nil, errs.Wrap(err, "failed to marshal result to JSON")699
+
}700
+
701
+
return string(jsonRes), nil702
+
}703
+
}704
+
705
+
// New creates a new handler with initialized clients for system and tcp calls.706
+
func New(nvmlRunner nvml.Runner) *Handler {707
+
return &Handler{708
+
// negative indicates no limit709
+
concurrentDeviceDiscoverys: -1,710
+
nvmlRunner: nvmlRunner,711
+
deviceCacheMux: &sync.Mutex{},712
+
deviceCache: make(map[string]nvml.Device),713
+
}714
+
}715
+
716
716
// getDeviceByUUID accesses devices from device cache of runner,717
717
// if device with requested UUID not cached it requests it from NVML.718
718
// In case of success it caches device for future requests, else returns error.719
719
//720
720
//nolint:ireturn721
721
func (h *Handler) getDeviceByUUID(uuid string) (nvml.Device, error) {722
722
h.deviceCacheMux.Lock()723
723
defer h.deviceCacheMux.Unlock()724
724
725
725
device, ok := h.deviceCache[uuid]