def test_dcgm_reader_default(handle):
# pylint: disable=undefined-variable
dr = DcgmReader()
dr.SetHandle(handle)
latest = dr.GetLatestGpuValuesAsFieldNameDict()
for gpuId in latest:
# latest data might be less than the list, because blank values aren't included
# Defined in DcgmReader
# pylint: disable=undefined-variable
assert len(latest[gpuId]) <= len(defaultFieldIds)
# Make sure we get strings
for key in latest[gpuId]:
assert isinstance(key, str)
sample = dr.GetLatestGpuValuesAsFieldIdDict()
for gpuId in sample:
# Defined in DcgmReader
# pylint: disable=undefined-variable
assert len(sample[gpuId]) <= len(defaultFieldIds)
# Make sure we get valid integer field ids
for fieldId in sample[gpuId]:
assert isinstance(fieldId, int)
assert dcgm_fields.DcgmFieldGetById(fieldId) != None
def GetFieldById(self, fieldId):
'''
Get a field's metadata by its dcgm_fields.DCGM_FI_* field ID
fieldId: dcgm_fields.DCGM_FI_* field ID of the field
Returns a dcgm_fields.c_dcgm_field_meta_t struct on success or None on error.
'''
return dcgm_fields.DcgmFieldGetById(fieldId)
def helper_field_has_variable_size(fieldId):
'''
Returns True if a field has a variable memory size per record. False if it doesn't.
'''
if fieldId == dcgm_fields.DCGM_FI_DEV_GPU_UTIL_SAMPLES or \
fieldId == dcgm_fields.DCGM_FI_DEV_MEM_COPY_UTIL_SAMPLES or \
fieldId == dcgm_fields.DCGM_FI_DEV_GRAPHICS_PIDS or \
fieldId == dcgm_fields.DCGM_FI_DEV_COMPUTE_PIDS:
return True
fieldMeta = dcgm_fields.DcgmFieldGetById(fieldId)
if fieldMeta.fieldType == dcgm_fields.DCGM_FT_BINARY:
return True
else:
return False
def _GetLatestGpuErrorSamples(self):
numErrors = 0
nowStr = time.strftime("%m/%d/%Y %H:%M:%S")
self._gpuWatcher.GetMore()
for entityGroupId in self._gpuWatcher.values.keys():
for entityId in self._gpuWatcher.values[entityGroupId]:
for fieldId in self._gpuWatcher.values[entityGroupId][entityId]:
for value in self._gpuWatcher.values[entityGroupId][entityId][fieldId].values:
if not value.isBlank and value.value > 0:
fieldMeta = dcgm_fields.DcgmFieldGetById(fieldId)
print "%s: Got error for GPU %d, field Id %s, value %d" % (nowStr, entityId, fieldMeta.tag, int(value.value))
numErrors += 1
self._gpuWatcher.EmptyValues()
if numErrors == 0:
print "%s: No GPU errors." % nowStr
# See the License for the specific language governing permissions and
# limitations under the License.
import pydcgm
import dcgm_fields
import dcgm_agent_internal
import dcgm_structs
import time
dcgmHandle = pydcgm.DcgmHandle(ipAddress="127.0.0.1")
dcgmSystem = dcgmHandle.GetSystem()
dcgmGroup = dcgmSystem.GetDefaultGroup()
#Discover which fieldIds are valid
g_fieldTags = {}
for fieldId in range(1, dcgm_fields.DCGM_FI_MAX_FIELDS):
fieldMeta = dcgm_fields.DcgmFieldGetById(fieldId)
if fieldMeta is None:
continue
g_fieldTags[fieldId] = fieldMeta.tag
#print("Found field tags: " + str(g_fieldTags))
fieldIds = sorted(g_fieldTags.keys())
gpuIds = dcgmGroup.GetGpuIds()
totalSampleCount = 0
cycleCount = 0
