def testValidImportantKnobs(self):
workloads = Workload.objects.all()
wkld_results = Result.objects.filter(workload=workloads[0])
knob_data, metric_data = aggregate_data(wkld_results)
# instead of doing actual metric pruning by factor analysis / clustering,
# just randomly select 5 nonconstant metrics
nonconst_metric_columnlabels = []
for col, cl in zip(metric_data['data'].T, metric_data['columnlabels']):
if np.any(col != col[0]):
nonconst_metric_columnlabels.append(cl)
num_metrics = min(5, len(nonconst_metric_columnlabels))
selected_columnlabels = np.random.choice(nonconst_metric_columnlabels,
num_metrics,
replace=False)
pruned_metric_idxs = [
i for i, metric_name in enumerate(metric_data['columnlabels'])
if metric_name in selected_columnlabels
]
pruned_metric_data = {
'data':
metric_data['data'][:, pruned_metric_idxs],
'rowlabels':
copy.deepcopy(metric_data['rowlabels']),
'columnlabels':
[metric_data['columnlabels'][i] for i in pruned_metric_idxs]
}
# run knob_identification using knob_data and fake pruned metrics
ranked_knobs = run_knob_identification(knob_data, pruned_metric_data,
workloads[0].dbms)
for k in ranked_knobs:
self.assertIn(k, knob_data['columnlabels'])
def testValidPrunedMetrics(self):
workloads = Workload.objects.all()
wkld_results = Result.objects.filter(workload=workloads[0])
metric_data = aggregate_data(wkld_results)[1]
pruned_metrics = run_workload_characterization(metric_data)
for m in pruned_metrics:
self.assertIn(m, metric_data['columnlabels'])
def testValidWorkload(self):
workloads = Workload.objects.all()
valid_workload = workloads[0]
dicts = aggregate_data(valid_workload)
keys = ['data', 'rowlabels', 'columnlabels']
for d in dicts:
for k in keys:
self.assertIn(k, d)
def testValidWorkload(self):
workloads = Workload.objects.all()
valid_workload = workloads[0]
wkld_results = Result.objects.filter(workload=valid_workload)
dicts = aggregate_data(wkld_results)
keys = ['data', 'rowlabels', 'columnlabels']
for d in dicts:
for k in keys:
self.assertIn(k, d)