def helper_test_inject_instance_fields(handle, gpuIds):
instances, cis = ensure_instance_ids(handle, gpuIds[0], 1, 1)
firstInstanceId = instances.keys()[0]
lastCIId = cis.keys()[0]
# Set up the watches on these groups
groupId = dcgm_agent.dcgmGroupCreate(handle, dcgm_structs.DCGM_GROUP_EMPTY,
'tien')
fieldGroupId = dcgm_agent.dcgmFieldGroupCreate(
handle, [dcgm_fields.DCGM_FI_DEV_ECC_DBE_VOL_TOTAL], 'kal')
dcgm_agent.dcgmGroupAddEntity(handle, groupId, dcgm_fields.DCGM_FE_GPU,
gpuIds[0])
dcgm_agent.dcgmGroupAddEntity(handle, groupId, dcgm_fields.DCGM_FE_GPU_I,
firstInstanceId)
dcgm_agent.dcgmGroupAddEntity(handle, groupId, dcgm_fields.DCGM_FE_GPU_CI,
lastCIId)
dcgm_agent.dcgmWatchFields(handle, groupId, fieldGroupId, 1, 100, 100)
dcgm_internal_helpers.inject_value(
handle,
gpuIds[0],
dcgm_fields.DCGM_FI_DEV_ECC_DBE_VOL_TOTAL,
2,
5,
isInt=True,
verifyInsertion=True,
entityType=dcgm_fields.DCGM_FE_GPU)
# Read the values to make sure they were stored properly
entities = [
dcgm_structs.c_dcgmGroupEntityPair_t(),
dcgm_structs.c_dcgmGroupEntityPair_t(),
dcgm_structs.c_dcgmGroupEntityPair_t()
]
entities[0].entityGroupId = dcgm_fields.DCGM_FE_GPU_I
entities[0].entityId = firstInstanceId
entities[1].entityGroupId = dcgm_fields.DCGM_FE_GPU_CI
entities[1].entityId = lastCIId
entities[2].entityGroupId = dcgm_fields.DCGM_FE_GPU
entities[2].entityId = gpuIds[0]
fieldIds = [dcgm_fields.DCGM_FI_DEV_ECC_DBE_VOL_TOTAL]
values = dcgm_agent.dcgmEntitiesGetLatestValues(handle, entities, fieldIds,
0)
for v in values:
if v.entityGroupId == dcgm_fields.DCGM_FE_GPU:
assert v.value.i64 == 2, "Failed to inject value 2 for entity %u from group %u" % (
v.entityId, v.entityGroupId)
else:
from dcgm_structs import DCGM_ST_NO_DATA
assert (
v.status == DCGM_ST_NO_DATA
), "Injected meaningless value %u for entity %u from group %u" % (
v.value.i64, v.entityId, v.entityGroupId)
def _WatchFieldGroup(self):
ret = dcgm_agent.dcgmWatchFields(self._handle, self._groupId, self._fieldGroup.fieldGroupId, self._updateFreq, self._maxKeepAge, self._maxKeepSamples)
dcgm_structs._dcgmCheckReturn(ret) #Will throw exception on error
#Force an update of the fields so that we can fetch initial values
ret = dcgm_agent.dcgmUpdateAllFields(self._handle, 1)
dcgm_structs._dcgmCheckReturn(ret) #Will throw exception on error
#initial update will fetch from startTimestamp
self.GetMore()
def WatchFields(self, fieldGroup, updateFreq, maxKeepAge, maxKeepSamples):
ret = dcgm_agent.dcgmWatchFields(self._dcgmHandle.handle, self._groupId, fieldGroup.fieldGroupId, updateFreq, maxKeepAge, maxKeepSamples)
dcgm_structs._dcgmCheckReturn(ret)