def aggregate_target_results(result_id, algorithm):
# Check that we've completed the background tasks at least once. We need
# this data in order to make a configuration recommendation (until we
# implement a sampling technique to generate new training data).
newest_result = Result.objects.get(pk=result_id)
has_pipeline_data = PipelineData.objects.filter(
workload=newest_result.workload).exists()
if not has_pipeline_data or newest_result.session.tuning_session == 'randomly_generate':
if not has_pipeline_data and newest_result.session.tuning_session == 'tuning_session':
LOG.debug(
"Background tasks haven't ran for this workload yet, picking random data."
)
result = Result.objects.filter(pk=result_id)
knobs = SessionKnob.objects.get_knobs_for_session(
newest_result.session)
# generate a config randomly
random_knob_result = gen_random_data(knobs)
agg_data = DataUtil.aggregate_data(result)
agg_data['newest_result_id'] = result_id
agg_data['bad'] = True
agg_data['config_recommend'] = random_knob_result
LOG.debug('%s: Finished generating a random config.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(agg_data, pprint=True))
else:
# Aggregate all knob config results tried by the target so far in this
# tuning session and this tuning workload.
target_results = Result.objects.filter(session=newest_result.session,
dbms=newest_result.dbms,
workload=newest_result.workload)
if len(target_results) == 0:
raise Exception(
'Cannot find any results for session_id={}, dbms_id={}'.format(
newest_result.session, newest_result.dbms))
agg_data = DataUtil.aggregate_data(target_results)
agg_data['newest_result_id'] = result_id
agg_data['bad'] = False
# Clean knob data
cleaned_agg_data = clean_knob_data(agg_data['X_matrix'],
agg_data['X_columnlabels'],
newest_result.session)
agg_data['X_matrix'] = np.array(cleaned_agg_data[0])
agg_data['X_columnlabels'] = np.array(cleaned_agg_data[1])
LOG.debug('%s: Finished aggregating target results.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(agg_data, pprint=True))
return agg_data, algorithm
def map_workload(target_data):
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
dbms = newest_result.dbms.pk
hardware = newest_result.application.hardware.pk
workload_data = PipelineResult.get_latest(
dbms, hardware, PipelineTaskType.WORKLOAD_MAPPING_DATA)
if workload_data is None:
target_data['scores'] = None
return target_data
data_values = JSONUtil.loads(workload_data.value)
X_scaler = np.load(data_values['X_scaler'])
y_scaler = np.load(data_values['y_scaler'])
y_deciles = np.load(data_values['y_deciles'])['deciles']
X_columnlabels = data_values['X_columnlabels']
y_columnlabels = data_values['y_columnlabels']
X_idxs = [
i for i in range(target_data['X_matrix'].shape[1])
if target_data['X_columnlabels'][i] in X_columnlabels
]
y_idxs = [
i for i in range(target_data['y_matrix'].shape[1])
if target_data['y_columnlabels'][i] in y_columnlabels
]
X_target = target_data['X_matrix'][:, X_idxs]
y_target = target_data['y_matrix'][:, y_idxs]
X_target = (X_target - X_scaler['mean']) / X_scaler['scale']
y_target = (y_target - y_scaler['mean']) / y_scaler['scale']
y_binned = np.empty_like(y_target)
for i in range(y_target.shape[1]):
y_binned[:, i] = bin_by_decile(y_target[:, i], y_deciles[i])
scores = {}
for wkld_id, wkld_entry_path in data_values['data'].iteritems():
wkld_entry = np.load(wkld_entry_path)
preds = np.empty_like(y_target)
X_wkld = wkld_entry['X_matrix']
for j in range(y_target.shape[1]):
y_col = wkld_entry['y_matrix'][:, j].reshape(X_wkld.shape[0], 1)
model = GPR()
model.fit(X_wkld, y_col, ridge=0.01)
preds[:, j] = bin_by_decile(
model.predict(X_target).ypreds.ravel(), y_deciles[j])
dists = np.sqrt(np.sum(np.square(np.subtract(preds, y_target)),
axis=1))
scores[wkld_id] = np.mean(dists)
# Find the best (minimum) score
best_score = np.inf
best_wkld_id = None
for wkld_id, similarity_score in scores.iteritems():
if similarity_score < best_score:
best_score = similarity_score
best_wkld_id = wkld_id
target_data['mapped_workload'] = (best_wkld_id, best_score)
target_data['scores'] = scores
return target_data
def aggregate_results():
unique_clusters = WorkloadCluster.objects.all()
unique_clusters = filter(lambda x: x.isdefault is False, unique_clusters)
all_data = {}
all_labels = {}
for cluster in unique_clusters:
results = ResultData.objects.filter(cluster=cluster)
if len(results) < 2:
continue
if cluster.dbms.pk not in all_labels:
knob_labels = np.asarray(
sorted(JSONUtil.loads(results[0].param_data).keys()))
metric_labels = np.asarray(
sorted(JSONUtil.loads(results[0].metric_data).keys()))
all_labels[cluster.dbms.pk] = (knob_labels, metric_labels)
else:
knob_labels, metric_labels = all_labels[cluster.dbms.pk]
entry = DataUtil.aggregate_data(results, knob_labels, metric_labels)
key = (cluster.dbms.pk, cluster.hardware.pk)
if key not in all_data:
all_data[key] = {}
all_data[key][cluster.pk] = entry
ts = now()
tsf = ts.strftime("%Y%m%d-%H%M%S")
for (dbkey, hwkey), cluster_data in all_data.iteritems():
task_name = PipelineTaskType.TYPE_NAMES[
PipelineTaskType.AGGREGATED_DATA].replace(' ', '').upper()
savepaths = {}
for clusterkey, entry in cluster_data.iteritems():
fname = '{}_{}_{}_{}_{}.npz'.format(task_name, dbkey, hwkey,
clusterkey, tsf)
savepath = os.path.join(PIPELINE_DIR, fname)
savepaths[clusterkey] = savepath
np.savez_compressed(savepath, **entry)
value = {'data': savepaths}
new_res = PipelineResult()
new_res.dbms = DBMSCatalog.objects.get(pk=dbkey)
new_res.hardware = Hardware.objects.get(pk=hwkey)
new_res.creation_timestamp = ts
new_res.task_type = PipelineTaskType.AGGREGATED_DATA
new_res.value = JSONUtil.dumps(value)
new_res.save()
def test_util(self):
json_str = \
"""{
"glossary": {
"title": "example glossary",
"GlossDiv": {
"title": "S",
"GlossList": {
"GlossEntry": {
"ID": "SGML",
"SortAs": "SGML",
"GlossTerm": "Standard Generalized Markup Language",
"Acronym": "SGML",
"Abbrev": "ISO 8879:1986",
"GlossDef": {
"para": "A meta-markup language",
"GlossSeeAlso": ["GML", "XML"]
},
"GlossSee": "markup"
}
}
}
}
}"""
compress_str = """{"glossary": {"title": "example glossary",
"GlossDiv": {"title": "S", "GlossList": {"GlossEntry": {"ID": "SGML",
"SortAs": "SGML", "GlossTerm": "Standard Generalized Markup
Language", "Acronym": "SGML", "Abbrev": "ISO 8879:1986", "GlossDef":
{"para": "A meta-markup language", "GlossSeeAlso": ["GML", "XML"]}, "GlossSee":
"markup"}}}}}"""
results = JSONUtil.loads(json_str)
self.assertEqual(list(results.keys())[0], "glossary")
self.assertTrue("title" in list(results["glossary"].keys()))
self.assertTrue("GlossDiv" in list(results["glossary"].keys()))
self.assertEqual(
results["glossary"]["GlossDiv"]["GlossList"]["GlossEntry"]["ID"],
"SGML")
self.assertEqual(
results["glossary"]["GlossDiv"]["GlossList"]["GlossEntry"]
["GlossSee"], "markup")
result_str = "".join(JSONUtil.dumps(results).split())
self.assertEqual(result_str, "".join(compress_str.split()))
def test_util(self):
json_str = \
"""{
"glossary": {
"title": "example glossary",
"GlossDiv": {
"title": "S",
"GlossList": {
"GlossEntry": {
"ID": "SGML",
"SortAs": "SGML",
"GlossTerm": "Standard Generalized Markup Language",
"Acronym": "SGML",
"Abbrev": "ISO 8879:1986",
"GlossDef": {
"para": "A meta-markup language",
"GlossSeeAlso": ["GML", "XML"]
},
"GlossSee": "markup"
}
}
}
}
}"""
compress_str = """{"glossary": {"title": "example glossary",
"GlossDiv": {"title": "S", "GlossList": {"GlossEntry": {"ID": "SGML",
"SortAs": "SGML", "GlossTerm": "Standard Generalized Markup
Language", "Acronym": "SGML", "Abbrev": "ISO 8879:1986", "GlossDef":
{"para": "A meta-markup language", "GlossSeeAlso": ["GML", "XML"]}, "GlossSee":
"markup"}}}}}"""
results = JSONUtil.loads(json_str)
self.assertEqual(list(results.keys())[0], "glossary")
self.assertTrue("title" in list(results["glossary"].keys()))
self.assertTrue("GlossDiv" in list(results["glossary"].keys()))
self.assertEqual(results["glossary"]["GlossDiv"]
["GlossList"]["GlossEntry"]["ID"], "SGML")
self.assertEqual(results["glossary"]["GlossDiv"]
["GlossList"]["GlossEntry"]["GlossSee"], "markup")
result_str = "".join(JSONUtil.dumps(results).split())
self.assertEqual(result_str, "".join(compress_str.split()))
def aggregate_target_results(result_id):
newest_result = Result.objects.get(pk=result_id)
target_results = Result.objects.filter(
application=newest_result.application, dbms=newest_result.dbms)
if len(target_results) == 0:
raise Exception(
'Cannot find any results for app_id={}, dbms_id={}'.format(
newest_result.application, newest_result.dbms))
target_result_datas = [
ResultData.objects.get(result=tres) for tres in target_results
]
knob_labels = np.asarray(
sorted(JSONUtil.loads(target_result_datas[0].param_data).keys()))
metric_labels = np.asarray(
sorted(JSONUtil.loads(target_result_datas[0].metric_data).keys()))
agg_data = DataUtil.aggregate_data(target_result_datas, knob_labels,
metric_labels)
agg_data['newest_result_id'] = result_id
return agg_data
def test_aggregate(self):
workload2 = Result.objects.filter(workload=2)
num_results = Result.objects.filter(workload=2).count()
knobs = list(JSONUtil.loads(workload2[0].knob_data.data).keys())
metrics = list(JSONUtil.loads(workload2[0].metric_data.data).keys())
num_knobs = len(knobs)
num_metrics = len(metrics)
test_result = DataUtil.aggregate_data(workload2)
self.assertTrue('X_matrix' in list(test_result.keys()))
self.assertTrue('y_matrix' in list(test_result.keys()))
self.assertTrue('rowlabels' in list(test_result.keys()))
self.assertTrue('X_columnlabels' in list(test_result.keys()))
self.assertTrue('y_columnlabels' in list(test_result.keys()))
self.assertEqual(test_result['X_columnlabels'], knobs)
self.assertEqual(test_result['y_columnlabels'], metrics)
self.assertEqual(test_result['X_matrix'].shape[0], num_results)
self.assertEqual(test_result['y_matrix'].shape[0], num_results)
self.assertEqual(test_result['X_matrix'].shape[1], num_knobs)
self.assertEqual(test_result['y_matrix'].shape[1], num_metrics)
def test_aggregate(self):
workload2 = Result.objects.filter(workload=2)
num_results = Result.objects.filter(workload=2).count()
knobs = list(JSONUtil.loads(workload2[0].knob_data.data).keys())
metrics = list(JSONUtil.loads(workload2[0].metric_data.data).keys())
num_knobs = len(knobs)
num_metrics = len(metrics)
test_result = DataUtil.aggregate_data(workload2)
self.assertTrue('X_matrix' in list(test_result.keys()))
self.assertTrue('y_matrix' in list(test_result.keys()))
self.assertTrue('rowlabels' in list(test_result.keys()))
self.assertTrue('X_columnlabels' in list(test_result.keys()))
self.assertTrue('y_columnlabels' in list(test_result.keys()))
self.assertEqual(test_result['X_columnlabels'], knobs)
self.assertEqual(test_result['y_columnlabels'], metrics)
self.assertEqual(test_result['X_matrix'].shape[0], num_results)
self.assertEqual(test_result['y_matrix'].shape[0], num_results)
self.assertEqual(test_result['X_matrix'].shape[1], num_knobs)
self.assertEqual(test_result['y_matrix'].shape[1], num_metrics)
def create_and_save_recommendation(recommended_knobs, result, status,
**kwargs):
dbms_id = result.dbms.pk
formatted_knobs = db.parser.format_dbms_knobs(dbms_id, recommended_knobs)
config = db.parser.create_knob_configuration(dbms_id, formatted_knobs)
retval = dict(**kwargs)
retval.update(
status=status,
result_id=result.pk,
recommendation=config,
)
result.next_configuration = JSONUtil.dumps(retval)
result.save()
return retval
def on_success(self, retval, task_id, args, kwargs):
super(ConfigurationRecommendation, self).on_success(retval, task_id, args, kwargs)
result_id = args[0]['newest_result_id']
result = Result.objects.get(pk=result_id)
# Replace result with formatted result
formatted_params = Parser.format_dbms_knobs(result.dbms.pk, retval['recommendation'])
task_meta = TaskMeta.objects.get(task_id=task_id)
retval['recommendation'] = formatted_params
task_meta.result = retval
task_meta.save()
# Create next configuration to try
config = Parser.create_knob_configuration(result.dbms.pk, retval['recommendation'])
retval['recommendation'] = config
result.next_configuration = JSONUtil.dumps(retval)
result.save()
def on_success(self, retval, task_id, args, kwargs):
super(ConfigurationRecommendation, self).on_success(retval, task_id, args, kwargs)
result_id = args[0]['newest_result_id']
result = Result.objects.get(pk=result_id)
# Replace result with formatted result
formatted_params = Parser.format_dbms_knobs(result.dbms.pk, retval['recommendation'])
task_meta = TaskMeta.objects.get(task_id=task_id)
retval['recommendation'] = formatted_params
task_meta.result = retval
task_meta.save()
# Create next configuration to try
config = Parser.create_knob_configuration(result.dbms.pk, retval['recommendation'])
retval['recommendation'] = config
result.next_configuration = JSONUtil.dumps(retval)
result.save()
def configuration_recommendation_ddpg(result_info): # pylint: disable=invalid-name
start_ts = time.time()
LOG.info('Use ddpg to recommend configuration')
result_id = result_info['newest_result_id']
result_list = Result.objects.filter(pk=result_id)
result = result_list.first()
session = result.session
params = JSONUtil.loads(session.hyperparameters)
agg_data = DataUtil.aggregate_data(result_list)
metric_data, _ = clean_metric_data(agg_data['y_matrix'],
agg_data['y_columnlabels'], session)
metric_data = metric_data.flatten()
metric_scalar = MinMaxScaler().fit(metric_data.reshape(1, -1))
normalized_metric_data = metric_scalar.transform(metric_data.reshape(
1, -1))[0]
cleaned_knob_data = clean_knob_data(agg_data['X_matrix'],
agg_data['X_columnlabels'], session)
knob_labels = np.array(cleaned_knob_data[1]).flatten()
knob_num = len(knob_labels)
metric_num = len(metric_data)
ddpg = DDPG(n_actions=knob_num,
n_states=metric_num,
a_hidden_sizes=params['DDPG_ACTOR_HIDDEN_SIZES'],
c_hidden_sizes=params['DDPG_CRITIC_HIDDEN_SIZES'],
use_default=params['DDPG_USE_DEFAULT'])
if session.ddpg_actor_model is not None and session.ddpg_critic_model is not None:
ddpg.set_model(session.ddpg_actor_model, session.ddpg_critic_model)
if session.ddpg_reply_memory is not None:
ddpg.replay_memory.set(session.ddpg_reply_memory)
knob_data = ddpg.choose_action(normalized_metric_data)
knob_bounds = np.vstack(DataUtil.get_knob_bounds(knob_labels, session))
knob_data = MinMaxScaler().fit(knob_bounds).inverse_transform(
knob_data.reshape(1, -1))[0]
conf_map = {k: knob_data[i] for i, k in enumerate(knob_labels)}
conf_map_res = create_and_save_recommendation(recommended_knobs=conf_map,
result=result,
status='good',
info='INFO: ddpg')
save_execution_time(start_ts, "configuration_recommendation_ddpg", result)
return conf_map_res
def on_success(self, retval, task_id, args, kwargs):
super(UpdateTask, self).on_success(retval, task_id, args, kwargs)
result_id = args[0]['newest_result_id']
result = Result.objects.get(pk=result_id)
# Replace result with formatted result
formatted_params = Parser.format_dbms_knobs(result.dbms.pk, retval)
task_meta = TaskMeta.objects.get(task_id=task_id)
task_meta.result = formatted_params
task_meta.save()
# Create next configuration to try
nondefault_params = JSONUtil.loads(result.session.nondefault_settings)
config = Parser.create_knob_configuration(result.dbms.pk,
formatted_params,
nondefault_params)
result.next_configuration = config
result.save()
def on_success(self, retval, task_id, args, kwargs):
super(UpdateTask, self).on_success(retval, task_id, args, kwargs)
result_id = args[0]['newest_result_id']
result = Result.objects.get(pk=result_id)
# Replace result with formatted result
formatted_params = DBMSUtil.format_dbms_params(result.dbms.pk, retval)
task_meta = TaskMeta.objects.get(task_id=task_id)
task_meta.result = formatted_params
task_meta.save()
# Create next configuration to try
nondefault_params = JSONUtil.loads(
result.application.nondefault_settings)
config = DBMSUtil.create_configuration(result.dbms.pk,
formatted_params,
nondefault_params)
path_prefix = MediaUtil.get_result_data_path(result.pk)
path = '{}.next_conf'.format(path_prefix)
with open(path, 'w') as f:
f.write(config)
def run_background_tasks():
# Find all unique workloads that we have data for
unique_workloads = Workload.objects.all()
if len(unique_workloads) == 0:
# No previous workload data yet. Try again later.
return
# Create new entry in PipelineRun table to store the output of each of
# the background tasks
pipeline_run_obj = PipelineRun(start_time=now(), end_time=None)
pipeline_run_obj.save()
for workload in unique_workloads:
# Aggregate the knob & metric data for this workload
knob_data, metric_data = aggregate_data(workload)
# Knob_data and metric_data are 2D numpy arrays. Convert them into a
# JSON-friendly (nested) lists and then save them as new PipelineData
# objects.
knob_data_copy = copy.deepcopy(knob_data)
knob_data_copy['data'] = knob_data_copy['data'].tolist()
knob_data_copy = JSONUtil.dumps(knob_data_copy)
knob_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.KNOB_DATA,
workload=workload,
data=knob_data_copy,
creation_time=now())
knob_entry.save()
metric_data_copy = copy.deepcopy(metric_data)
metric_data_copy['data'] = metric_data_copy['data'].tolist()
metric_data_copy = JSONUtil.dumps(metric_data_copy)
metric_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.METRIC_DATA,
workload=workload,
data=metric_data_copy,
creation_time=now())
metric_entry.save()
# Execute the Workload Characterization task to compute the list of
# pruned metrics for this workload and save them in a new PipelineData
# object.
pruned_metrics = run_workload_characterization(metric_data=metric_data)
pruned_metrics_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.PRUNED_METRICS,
workload=workload,
data=JSONUtil.dumps(pruned_metrics),
creation_time=now())
pruned_metrics_entry.save()
# Use the pruned metrics to filter the metric_data
pruned_metric_idxs = [i for i, metric_name in enumerate(metric_data['columnlabels'])
if metric_name in pruned_metrics]
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]
}
# Execute the Knob Identification task to compute an ordered list of knobs
# ranked by their impact on the DBMS's performance. Save them in a new
# PipelineData object.
ranked_knobs = run_knob_identification(knob_data=knob_data,
metric_data=pruned_metric_data,
dbms=workload.dbms)
ranked_knobs_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.RANKED_KNOBS,
workload=workload,
data=JSONUtil.dumps(ranked_knobs),
creation_time=now())
ranked_knobs_entry.save()
# Set the end_timestamp to the current time to indicate that we are done running
# the background tasks
pipeline_run_obj.end_time = now()
pipeline_run_obj.save()
def run_background_tasks():
# Find all unique workloads that we have data for
unique_workloads = Workload.objects.all()
if len(unique_workloads) == 0:
# No previous workload data yet. Try again later.
return
# Create new entry in PipelineRun table to store the output of each of
# the background tasks
pipeline_run_obj = PipelineRun(start_time=now(), end_time=None)
pipeline_run_obj.save()
for workload in unique_workloads:
# Aggregate the knob & metric data for this workload
knob_data, metric_data = aggregate_data(workload)
# Knob_data and metric_data are 2D numpy arrays. Convert them into a
# JSON-friendly (nested) lists and then save them as new PipelineData
# objects.
knob_data_copy = copy.deepcopy(knob_data)
knob_data_copy['data'] = knob_data_copy['data'].tolist()
knob_data_copy = JSONUtil.dumps(knob_data_copy)
knob_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.KNOB_DATA,
workload=workload,
data=knob_data_copy,
creation_time=now())
knob_entry.save()
metric_data_copy = copy.deepcopy(metric_data)
metric_data_copy['data'] = metric_data_copy['data'].tolist()
metric_data_copy = JSONUtil.dumps(metric_data_copy)
metric_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.METRIC_DATA,
workload=workload,
data=metric_data_copy,
creation_time=now())
metric_entry.save()
# Execute the Workload Characterization task to compute the list of
# pruned metrics for this workload and save them in a new PipelineData
# object.
pruned_metrics = run_workload_characterization(metric_data=metric_data)
pruned_metrics_entry = PipelineData(
pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.PRUNED_METRICS,
workload=workload,
data=JSONUtil.dumps(pruned_metrics),
creation_time=now())
pruned_metrics_entry.save()
# Use the pruned metrics to filter the metric_data
pruned_metric_idxs = [
i for i, metric_name in enumerate(metric_data['columnlabels'])
if metric_name in pruned_metrics
]
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]
}
# Execute the Knob Identification task to compute an ordered list of knobs
# ranked by their impact on the DBMS's performance. Save them in a new
# PipelineData object.
ranked_knobs = run_knob_identification(knob_data=knob_data,
metric_data=pruned_metric_data,
dbms=workload.dbms)
ranked_knobs_entry = PipelineData(
pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.RANKED_KNOBS,
workload=workload,
data=JSONUtil.dumps(ranked_knobs),
creation_time=now())
ranked_knobs_entry.save()
# Set the end_timestamp to the current time to indicate that we are done running
# the background tasks
pipeline_run_obj.end_time = now()
pipeline_run_obj.save()
def map_workload(map_workload_input):
target_data, algorithm = map_workload_input
if target_data['bad']:
assert target_data is not None
target_data['pipeline_run'] = None
LOG.debug('%s: Skipping workload mapping.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(target_data, pprint=True))
return target_data, algorithm
# Get the latest version of pipeline data that's been computed so far.
latest_pipeline_run = PipelineRun.objects.get_latest()
assert latest_pipeline_run is not None
target_data['pipeline_run'] = latest_pipeline_run.pk
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
target_workload = newest_result.workload
X_columnlabels = np.array(target_data['X_columnlabels'])
y_columnlabels = np.array(target_data['y_columnlabels'])
# Find all pipeline data belonging to the latest version with the same
# DBMS and hardware as the target
pipeline_data = PipelineData.objects.filter(
pipeline_run=latest_pipeline_run,
workload__dbms=target_workload.dbms,
workload__hardware=target_workload.hardware)
# FIXME (dva): we should also compute the global (i.e., overall) ranked_knobs
# and pruned metrics but we just use those from the first workload for now
initialized = False
global_ranked_knobs = None
global_pruned_metrics = None
ranked_knob_idxs = None
pruned_metric_idxs = None
# Compute workload mapping data for each unique workload
unique_workloads = pipeline_data.values_list('workload',
flat=True).distinct()
assert len(unique_workloads) > 0
workload_data = {}
for unique_workload in unique_workloads:
workload_obj = Workload.objects.get(pk=unique_workload)
wkld_results = Result.objects.filter(workload=workload_obj)
if wkld_results.exists() is False:
# delete the workload
workload_obj.delete()
continue
# Load knob & metric data for this workload
knob_data = load_data_helper(pipeline_data, unique_workload,
PipelineTaskType.KNOB_DATA)
knob_data["data"], knob_data["columnlabels"] = clean_knob_data(
knob_data["data"], knob_data["columnlabels"],
newest_result.session)
metric_data = load_data_helper(pipeline_data, unique_workload,
PipelineTaskType.METRIC_DATA)
X_matrix = np.array(knob_data["data"])
y_matrix = np.array(metric_data["data"])
rowlabels = np.array(knob_data["rowlabels"])
assert np.array_equal(rowlabels, metric_data["rowlabels"])
if not initialized:
# For now set ranked knobs & pruned metrics to be those computed
# for the first workload
global_ranked_knobs = load_data_helper(
pipeline_data, unique_workload,
PipelineTaskType.RANKED_KNOBS)[:IMPORTANT_KNOB_NUMBER]
global_pruned_metrics = load_data_helper(
pipeline_data, unique_workload,
PipelineTaskType.PRUNED_METRICS)
ranked_knob_idxs = [
i for i in range(X_matrix.shape[1])
if X_columnlabels[i] in global_ranked_knobs
]
pruned_metric_idxs = [
i for i in range(y_matrix.shape[1])
if y_columnlabels[i] in global_pruned_metrics
]
# Filter X & y columnlabels by top ranked_knobs & pruned_metrics
X_columnlabels = X_columnlabels[ranked_knob_idxs]
y_columnlabels = y_columnlabels[pruned_metric_idxs]
initialized = True
# Filter X & y matrices by top ranked_knobs & pruned_metrics
X_matrix = X_matrix[:, ranked_knob_idxs]
y_matrix = y_matrix[:, pruned_metric_idxs]
# Combine duplicate rows (rows with same knob settings)
X_matrix, y_matrix, rowlabels = DataUtil.combine_duplicate_rows(
X_matrix, y_matrix, rowlabels)
workload_data[unique_workload] = {
'X_matrix': X_matrix,
'y_matrix': y_matrix,
'rowlabels': rowlabels,
}
assert len(workload_data) > 0
# Stack all X & y matrices for preprocessing
Xs = np.vstack(
[entry['X_matrix'] for entry in list(workload_data.values())])
ys = np.vstack(
[entry['y_matrix'] for entry in list(workload_data.values())])
# Scale the X & y values, then compute the deciles for each column in y
X_scaler = StandardScaler(copy=False)
X_scaler.fit(Xs)
y_scaler = StandardScaler(copy=False)
y_scaler.fit_transform(ys)
y_binner = Bin(bin_start=1, axis=0)
y_binner.fit(ys)
del Xs
del ys
# Filter the target's X & y data by the ranked knobs & pruned metrics.
X_target = target_data['X_matrix'][:, ranked_knob_idxs]
y_target = target_data['y_matrix'][:, pruned_metric_idxs]
# Now standardize the target's data and bin it by the deciles we just
# calculated
X_target = X_scaler.transform(X_target)
y_target = y_scaler.transform(y_target)
y_target = y_binner.transform(y_target)
scores = {}
for workload_id, workload_entry in list(workload_data.items()):
predictions = np.empty_like(y_target)
X_workload = workload_entry['X_matrix']
X_scaled = X_scaler.transform(X_workload)
y_workload = workload_entry['y_matrix']
y_scaled = y_scaler.transform(y_workload)
for j, y_col in enumerate(y_scaled.T):
# Using this workload's data, train a Gaussian process model
# and then predict the performance of each metric for each of
# the knob configurations attempted so far by the target.
y_col = y_col.reshape(-1, 1)
model = GPRNP(length_scale=DEFAULT_LENGTH_SCALE,
magnitude=DEFAULT_MAGNITUDE,
max_train_size=MAX_TRAIN_SIZE,
batch_size=BATCH_SIZE)
model.fit(X_scaled, y_col, ridge=DEFAULT_RIDGE)
predictions[:, j] = model.predict(X_target).ypreds.ravel()
# Bin each of the predicted metric columns by deciles and then
# compute the score (i.e., distance) between the target workload
# and each of the known workloads
predictions = y_binner.transform(predictions)
dists = np.sqrt(
np.sum(np.square(np.subtract(predictions, y_target)), axis=1))
scores[workload_id] = np.mean(dists)
# Find the best (minimum) score
best_score = np.inf
best_workload_id = None
best_workload_name = None
scores_info = {}
for workload_id, similarity_score in list(scores.items()):
workload_name = Workload.objects.get(pk=workload_id).name
if similarity_score < best_score:
best_score = similarity_score
best_workload_id = workload_id
best_workload_name = workload_name
scores_info[workload_id] = (workload_name, similarity_score)
target_data.update(mapped_workload=(best_workload_id, best_workload_name,
best_score),
scores=scores_info)
LOG.debug('%s: Finished mapping the workload.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(target_data, pprint=True))
return target_data, algorithm
def configuration_recommendation(recommendation_input):
target_data, algorithm = recommendation_input
LOG.info('configuration_recommendation called')
if target_data['bad'] is True:
target_data_res = dict(
status='bad',
result_id=target_data['newest_result_id'],
info='WARNING: no training data, the config is generated randomly',
recommendation=target_data['config_recommend'],
pipeline_run=target_data['pipeline_run'])
LOG.debug('%s: Skipping configuration recommendation.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(target_data, pprint=True))
return target_data_res
# Load mapped workload data
mapped_workload_id = target_data['mapped_workload'][0]
latest_pipeline_run = PipelineRun.objects.get(
pk=target_data['pipeline_run'])
mapped_workload = Workload.objects.get(pk=mapped_workload_id)
workload_knob_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.KNOB_DATA)
workload_knob_data = JSONUtil.loads(workload_knob_data.data)
workload_metric_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.METRIC_DATA)
workload_metric_data = JSONUtil.loads(workload_metric_data.data)
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
cleaned_workload_knob_data = clean_knob_data(
workload_knob_data["data"], workload_knob_data["columnlabels"],
newest_result.session)
X_workload = np.array(cleaned_workload_knob_data[0])
X_columnlabels = np.array(cleaned_workload_knob_data[1])
y_workload = np.array(workload_metric_data['data'])
y_columnlabels = np.array(workload_metric_data['columnlabels'])
rowlabels_workload = np.array(workload_metric_data['rowlabels'])
# Target workload data
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
X_target = target_data['X_matrix']
y_target = target_data['y_matrix']
rowlabels_target = np.array(target_data['rowlabels'])
if not np.array_equal(X_columnlabels, target_data['X_columnlabels']):
raise Exception(('The workload and target data should have '
'identical X columnlabels (sorted knob names)'))
if not np.array_equal(y_columnlabels, target_data['y_columnlabels']):
raise Exception(('The workload and target data should have '
'identical y columnlabels (sorted metric names)'))
# Filter Xs by top 10 ranked knobs
ranked_knobs = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.RANKED_KNOBS)
ranked_knobs = JSONUtil.loads(ranked_knobs.data)[:IMPORTANT_KNOB_NUMBER]
ranked_knob_idxs = [
i for i, cl in enumerate(X_columnlabels) if cl in ranked_knobs
]
X_workload = X_workload[:, ranked_knob_idxs]
X_target = X_target[:, ranked_knob_idxs]
X_columnlabels = X_columnlabels[ranked_knob_idxs]
# Filter ys by current target objective metric
target_objective = newest_result.session.target_objective
target_obj_idx = [
i for i, cl in enumerate(y_columnlabels) if cl == target_objective
]
if len(target_obj_idx) == 0:
raise Exception(('Could not find target objective in metrics '
'(target_obj={})').format(target_objective))
elif len(target_obj_idx) > 1:
raise Exception(
('Found {} instances of target objective in '
'metrics (target_obj={})').format(len(target_obj_idx),
target_objective))
metric_meta = db.target_objectives.get_metric_metadata(
newest_result.session.dbms.pk, newest_result.session.target_objective)
lessisbetter = metric_meta[
target_objective].improvement == db.target_objectives.LESS_IS_BETTER
y_workload = y_workload[:, target_obj_idx]
y_target = y_target[:, target_obj_idx]
y_columnlabels = y_columnlabels[target_obj_idx]
# Combine duplicate rows in the target/workload data (separately)
X_workload, y_workload, rowlabels_workload = DataUtil.combine_duplicate_rows(
X_workload, y_workload, rowlabels_workload)
X_target, y_target, rowlabels_target = DataUtil.combine_duplicate_rows(
X_target, y_target, rowlabels_target)
# Delete any rows that appear in both the workload data and the target
# data from the workload data
dups_filter = np.ones(X_workload.shape[0], dtype=bool)
target_row_tups = [tuple(row) for row in X_target]
for i, row in enumerate(X_workload):
if tuple(row) in target_row_tups:
dups_filter[i] = False
X_workload = X_workload[dups_filter, :]
y_workload = y_workload[dups_filter, :]
rowlabels_workload = rowlabels_workload[dups_filter]
# Combine target & workload Xs for preprocessing
X_matrix = np.vstack([X_target, X_workload])
# Dummy encode categorial variables
categorical_info = DataUtil.dummy_encoder_helper(X_columnlabels,
mapped_workload.dbms)
dummy_encoder = DummyEncoder(categorical_info['n_values'],
categorical_info['categorical_features'],
categorical_info['cat_columnlabels'],
categorical_info['noncat_columnlabels'])
X_matrix = dummy_encoder.fit_transform(X_matrix)
# below two variables are needed for correctly determing max/min on dummies
binary_index_set = set(categorical_info['binary_vars'])
total_dummies = dummy_encoder.total_dummies()
# Scale to N(0, 1)
X_scaler = StandardScaler()
X_scaled = X_scaler.fit_transform(X_matrix)
if y_target.shape[0] < 5: # FIXME
# FIXME (dva): if there are fewer than 5 target results so far
# then scale the y values (metrics) using the workload's
# y_scaler. I'm not sure if 5 is the right cutoff.
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_matrix = np.vstack([y_target, y_workload])
y_scaled = y_workload_scaler.fit_transform(y_matrix)
else:
# FIXME (dva): otherwise try to compute a separate y_scaler for
# the target and scale them separately.
try:
y_target_scaler = StandardScaler()
y_workload_scaler = StandardScaler()
y_target_scaled = y_target_scaler.fit_transform(y_target)
y_workload_scaled = y_workload_scaler.fit_transform(y_workload)
y_scaled = np.vstack([y_target_scaled, y_workload_scaled])
except ValueError:
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_scaled = y_workload_scaler.fit_transform(y_target)
# Set up constraint helper
constraint_helper = ParamConstraintHelper(
scaler=X_scaler,
encoder=dummy_encoder,
binary_vars=categorical_info['binary_vars'],
init_flip_prob=INIT_FLIP_PROB,
flip_prob_decay=FLIP_PROB_DECAY)
# FIXME (dva): check if these are good values for the ridge
# ridge = np.empty(X_scaled.shape[0])
# ridge[:X_target.shape[0]] = 0.01
# ridge[X_target.shape[0]:] = 0.1
# FIXME: we should generate more samples and use a smarter sampling
# technique
num_samples = NUM_SAMPLES
X_samples = np.empty((num_samples, X_scaled.shape[1]))
X_min = np.empty(X_scaled.shape[1])
X_max = np.empty(X_scaled.shape[1])
X_scaler_matrix = np.zeros([1, X_scaled.shape[1]])
session_knobs = SessionKnob.objects.get_knobs_for_session(
newest_result.session)
# Set min/max for knob values
for i in range(X_scaled.shape[1]):
if i < total_dummies or i in binary_index_set:
col_min = 0
col_max = 1
else:
col_min = X_scaled[:, i].min()
col_max = X_scaled[:, i].max()
for knob in session_knobs:
if X_columnlabels[i] == knob["name"]:
X_scaler_matrix[0][i] = knob["minval"]
col_min = X_scaler.transform(X_scaler_matrix)[0][i]
X_scaler_matrix[0][i] = knob["maxval"]
col_max = X_scaler.transform(X_scaler_matrix)[0][i]
X_min[i] = col_min
X_max[i] = col_max
X_samples[:, i] = np.random.rand(num_samples) * (col_max -
col_min) + col_min
# Maximize the throughput, moreisbetter
# Use gradient descent to minimize -throughput
if not lessisbetter:
y_scaled = -y_scaled
q = queue.PriorityQueue()
for x in range(0, y_scaled.shape[0]):
q.put((y_scaled[x][0], x))
i = 0
while i < TOP_NUM_CONFIG:
try:
item = q.get_nowait()
# Tensorflow get broken if we use the training data points as
# starting points for GPRGD. We add a small bias for the
# starting points. GPR_EPS default value is 0.001
# if the starting point is X_max, we minus a small bias to
# make sure it is within the range.
dist = sum(np.square(X_max - X_scaled[item[1]]))
if dist < 0.001:
X_samples = np.vstack(
(X_samples, X_scaled[item[1]] - abs(GPR_EPS)))
else:
X_samples = np.vstack(
(X_samples, X_scaled[item[1]] + abs(GPR_EPS)))
i = i + 1
except queue.Empty:
break
session = newest_result.session
res = None
if algorithm == AlgorithmType.DNN:
# neural network model
model_nn = NeuralNet(n_input=X_samples.shape[1],
batch_size=X_samples.shape[0],
explore_iters=DNN_EXPLORE_ITER,
noise_scale_begin=DNN_NOISE_SCALE_BEGIN,
noise_scale_end=DNN_NOISE_SCALE_END,
debug=DNN_DEBUG,
debug_interval=DNN_DEBUG_INTERVAL)
if session.dnn_model is not None:
model_nn.set_weights_bin(session.dnn_model)
model_nn.fit(X_scaled, y_scaled, fit_epochs=DNN_TRAIN_ITER)
res = model_nn.recommend(X_samples,
X_min,
X_max,
explore=DNN_EXPLORE,
recommend_epochs=MAX_ITER)
session.dnn_model = model_nn.get_weights_bin()
session.save()
elif algorithm == AlgorithmType.GPR:
# default gpr model
model = GPRGD(length_scale=DEFAULT_LENGTH_SCALE,
magnitude=DEFAULT_MAGNITUDE,
max_train_size=MAX_TRAIN_SIZE,
batch_size=BATCH_SIZE,
num_threads=NUM_THREADS,
learning_rate=DEFAULT_LEARNING_RATE,
epsilon=DEFAULT_EPSILON,
max_iter=MAX_ITER,
sigma_multiplier=DEFAULT_SIGMA_MULTIPLIER,
mu_multiplier=DEFAULT_MU_MULTIPLIER)
model.fit(X_scaled, y_scaled, X_min, X_max, ridge=DEFAULT_RIDGE)
res = model.predict(X_samples, constraint_helper=constraint_helper)
best_config_idx = np.argmin(res.minl.ravel())
best_config = res.minl_conf[best_config_idx, :]
best_config = X_scaler.inverse_transform(best_config)
# Decode one-hot encoding into categorical knobs
best_config = dummy_encoder.inverse_transform(best_config)
# Although we have max/min limits in the GPRGD training session, it may
# lose some precisions. e.g. 0.99..99 >= 1.0 may be True on the scaled data,
# when we inversely transform the scaled data, the different becomes much larger
# and cannot be ignored. Here we check the range on the original data
# directly, and make sure the recommended config lies within the range
X_min_inv = X_scaler.inverse_transform(X_min)
X_max_inv = X_scaler.inverse_transform(X_max)
best_config = np.minimum(best_config, X_max_inv)
best_config = np.maximum(best_config, X_min_inv)
conf_map = {k: best_config[i] for i, k in enumerate(X_columnlabels)}
conf_map_res = dict(status='good',
result_id=target_data['newest_result_id'],
recommendation=conf_map,
info='INFO: training data size is {}'.format(
X_scaled.shape[0]),
pipeline_run=latest_pipeline_run.pk)
LOG.debug('%s: Finished selecting the next config.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(conf_map_res, pprint=True))
return conf_map_res
def configuration_recommendation(recommendation_input):
target_data, algorithm = recommendation_input
LOG.info('configuration_recommendation called')
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
session = newest_result.session
params = JSONUtil.loads(session.hyperparameters)
if target_data['bad'] is True:
target_data_res = create_and_save_recommendation(
recommended_knobs=target_data['config_recommend'],
result=newest_result,
status='bad',
info='WARNING: no training data, the config is generated randomly',
pipeline_run=target_data['pipeline_run'])
LOG.debug('%s: Skipping configuration recommendation.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(target_data, pprint=True))
return target_data_res
X_columnlabels, X_scaler, X_scaled, y_scaled, X_max, X_min,\
dummy_encoder, constraint_helper = combine_workload(target_data)
# FIXME: we should generate more samples and use a smarter sampling
# technique
num_samples = params['NUM_SAMPLES']
X_samples = np.empty((num_samples, X_scaled.shape[1]))
for i in range(X_scaled.shape[1]):
X_samples[:, i] = np.random.rand(num_samples) * (X_max[i] -
X_min[i]) + X_min[i]
q = queue.PriorityQueue()
for x in range(0, y_scaled.shape[0]):
q.put((y_scaled[x][0], x))
i = 0
while i < params['TOP_NUM_CONFIG']:
try:
item = q.get_nowait()
# Tensorflow get broken if we use the training data points as
# starting points for GPRGD. We add a small bias for the
# starting points. GPR_EPS default value is 0.001
# if the starting point is X_max, we minus a small bias to
# make sure it is within the range.
dist = sum(np.square(X_max - X_scaled[item[1]]))
if dist < 0.001:
X_samples = np.vstack(
(X_samples, X_scaled[item[1]] - abs(params['GPR_EPS'])))
else:
X_samples = np.vstack(
(X_samples, X_scaled[item[1]] + abs(params['GPR_EPS'])))
i = i + 1
except queue.Empty:
break
res = None
if algorithm == AlgorithmType.DNN:
# neural network model
model_nn = NeuralNet(n_input=X_samples.shape[1],
batch_size=X_samples.shape[0],
explore_iters=params['DNN_EXPLORE_ITER'],
noise_scale_begin=params['DNN_NOISE_SCALE_BEGIN'],
noise_scale_end=params['DNN_NOISE_SCALE_END'],
debug=params['DNN_DEBUG'],
debug_interval=params['DNN_DEBUG_INTERVAL'])
if session.dnn_model is not None:
model_nn.set_weights_bin(session.dnn_model)
model_nn.fit(X_scaled, y_scaled, fit_epochs=params['DNN_TRAIN_ITER'])
res = model_nn.recommend(X_samples,
X_min,
X_max,
explore=params['DNN_EXPLORE'],
recommend_epochs=params['DNN_GD_ITER'])
session.dnn_model = model_nn.get_weights_bin()
session.save()
elif algorithm == AlgorithmType.GPR:
# default gpr model
if params['GPR_USE_GPFLOW']:
model_kwargs = {}
model_kwargs['model_learning_rate'] = params[
'GPR_HP_LEARNING_RATE']
model_kwargs['model_maxiter'] = params['GPR_HP_MAX_ITER']
opt_kwargs = {}
opt_kwargs['learning_rate'] = params['GPR_LEARNING_RATE']
opt_kwargs['maxiter'] = params['GPR_MAX_ITER']
opt_kwargs['bounds'] = [X_min, X_max]
opt_kwargs['debug'] = params['GPR_DEBUG']
opt_kwargs['ucb_beta'] = ucb.get_ucb_beta(
params['GPR_UCB_BETA'],
scale=params['GPR_UCB_SCALE'],
t=i + 1.,
ndim=X_scaled.shape[1])
tf.reset_default_graph()
graph = tf.get_default_graph()
gpflow.reset_default_session(graph=graph)
m = gpr_models.create_model(params['GPR_MODEL_NAME'],
X=X_scaled,
y=y_scaled,
**model_kwargs)
res = tf_optimize(m.model, X_samples, **opt_kwargs)
else:
model = GPRGD(length_scale=params['GPR_LENGTH_SCALE'],
magnitude=params['GPR_MAGNITUDE'],
max_train_size=params['GPR_MAX_TRAIN_SIZE'],
batch_size=params['GPR_BATCH_SIZE'],
num_threads=params['TF_NUM_THREADS'],
learning_rate=params['GPR_LEARNING_RATE'],
epsilon=params['GPR_EPSILON'],
max_iter=params['GPR_MAX_ITER'],
sigma_multiplier=params['GPR_SIGMA_MULTIPLIER'],
mu_multiplier=params['GPR_MU_MULTIPLIER'],
ridge=params['GPR_RIDGE'])
model.fit(X_scaled, y_scaled, X_min, X_max)
res = model.predict(X_samples, constraint_helper=constraint_helper)
best_config_idx = np.argmin(res.minl.ravel())
best_config = res.minl_conf[best_config_idx, :]
best_config = X_scaler.inverse_transform(best_config)
if ENABLE_DUMMY_ENCODER:
# Decode one-hot encoding into categorical knobs
best_config = dummy_encoder.inverse_transform(best_config)
# Although we have max/min limits in the GPRGD training session, it may
# lose some precisions. e.g. 0.99..99 >= 1.0 may be True on the scaled data,
# when we inversely transform the scaled data, the different becomes much larger
# and cannot be ignored. Here we check the range on the original data
# directly, and make sure the recommended config lies within the range
X_min_inv = X_scaler.inverse_transform(X_min)
X_max_inv = X_scaler.inverse_transform(X_max)
best_config = np.minimum(best_config, X_max_inv)
best_config = np.maximum(best_config, X_min_inv)
conf_map = {k: best_config[i] for i, k in enumerate(X_columnlabels)}
conf_map_res = create_and_save_recommendation(
recommended_knobs=conf_map,
result=newest_result,
status='good',
info='INFO: training data size is {}'.format(X_scaled.shape[0]),
pipeline_run=target_data['pipeline_run'])
LOG.debug('%s: Finished selecting the next config.\n\ndata=%s\n',
AlgorithmType.name(algorithm),
JSONUtil.dumps(conf_map_res, pprint=True))
return conf_map_res
def create_workload_mapping_data():
agg_datas = PipelineResult.objects.filter(
task_type=PipelineTaskType.AGGREGATED_DATA)
dbmss = set([ad.dbms.pk for ad in agg_datas])
hardwares = set([ad.hardware.pk for ad in agg_datas])
for dbms_id, hw_id in itertools.product(dbmss, hardwares):
data = PipelineResult.get_latest(dbms_id, hw_id,
PipelineTaskType.AGGREGATED_DATA)
file_info = JSONUtil.loads(data.value)
cluster_data = OrderedDict()
for cluster, path in file_info['data'].iteritems():
compressed_data = np.load(path)
X_matrix = compressed_data['X_matrix']
y_matrix = compressed_data['y_matrix']
X_columnlabels = compressed_data['X_columnlabels']
y_columnlabels = compressed_data['y_columnlabels']
rowlabels = compressed_data['rowlabels']
# Filter metrics and knobs
ranked_knobs = JSONUtil.loads(
PipelineResult.get_latest(
dbms_id, hw_id,
PipelineTaskType.RANKED_KNOBS).value)[:10] # FIXME
pruned_metrics = JSONUtil.loads(
PipelineResult.get_latest(
dbms_id, hw_id, PipelineTaskType.PRUNED_METRICS).value)
knob_idxs = [
i for i in range(X_matrix.shape[1])
if X_columnlabels[i] in ranked_knobs
]
metric_idxs = [
i for i in range(y_matrix.shape[1])
if y_columnlabels[i] in pruned_metrics
]
X_matrix = X_matrix[:, knob_idxs]
X_columnlabels = X_columnlabels[knob_idxs]
y_matrix = y_matrix[:, metric_idxs]
y_columnlabels = y_columnlabels[metric_idxs]
# Combine duplicate rows
X_matrix, y_matrix, rowlabels = DataUtil.combine_duplicate_rows(
X_matrix, y_matrix, rowlabels)
cluster_data[cluster] = {
'X_matrix': X_matrix,
'y_matrix': y_matrix,
'X_columnlabels': X_columnlabels,
'y_columnlabels': y_columnlabels,
'rowlabels': rowlabels,
}
Xs = np.vstack([entry['X_matrix'] for entry in cluster_data.values()])
ys = np.vstack([entry['y_matrix'] for entry in cluster_data.values()])
X_scaler = StandardScaler(copy=False)
X_scaler.fit(Xs)
y_scaler = StandardScaler(copy=False)
y_scaler.fit_transform(ys)
y_binner = Bin(axis=0)
y_binner.fit(ys)
del Xs
del ys
task_name = PipelineTaskType.TYPE_NAMES[
PipelineTaskType.WORKLOAD_MAPPING_DATA].replace(' ', '').upper()
timestamp = data.creation_timestamp
tsf = timestamp.strftime("%Y%m%d-%H%M%S")
savepaths = {}
for cluster, entry in cluster_data.iteritems():
X_scaler.transform(entry['X_matrix'])
y_scaler.transform(entry['y_matrix'])
fname = '{}_{}_{}_{}_{}.npz'.format(task_name, dbms_id, hw_id,
cluster, tsf)
savepath = os.path.join(PIPELINE_DIR, fname)
savepaths[cluster] = savepath
np.savez_compressed(savepath, **entry)
X_scaler_path = os.path.join(
PIPELINE_DIR,
'{}_XSCALER_{}_{}_{}.npz'.format(task_name, dbms_id, hw_id, tsf))
np.savez_compressed(X_scaler_path,
mean=X_scaler.mean_,
scale=X_scaler.scale_)
y_scaler_path = os.path.join(
PIPELINE_DIR,
'{}_YSCALER_{}_{}_{}.npz'.format(task_name, dbms_id, hw_id, tsf))
np.savez_compressed(y_scaler_path,
mean=y_scaler.mean_,
scale=y_scaler.scale_)
y_deciles_path = os.path.join(
PIPELINE_DIR,
'{}_YDECILES_{}_{}_{}.npz'.format(task_name, dbms_id, hw_id, tsf))
np.savez_compressed(y_deciles_path, deciles=y_binner.deciles_)
value = {
'data': savepaths,
'X_scaler': X_scaler_path,
'y_scaler': y_scaler_path,
'y_deciles': y_deciles_path,
'X_columnlabels':
cluster_data.values()[0]['X_columnlabels'].tolist(),
'y_columnlabels':
cluster_data.values()[0]['y_columnlabels'].tolist(),
}
new_res = PipelineResult()
new_res.dbms = DBMSCatalog.objects.get(pk=dbms_id)
new_res.hardware = Hardware.objects.get(pk=hw_id)
new_res.creation_timestamp = timestamp
new_res.task_type = PipelineTaskType.WORKLOAD_MAPPING_DATA
new_res.value = JSONUtil.dumps(value, pprint=True)
new_res.save()
def run_background_tasks():
start_ts = time.time()
LOG.info("Starting background tasks...")
# Find modified and not modified workloads, we only have to calculate for the
# modified workloads.
modified_workloads = Workload.objects.filter(
status=WorkloadStatusType.MODIFIED)
num_modified = modified_workloads.count()
non_modified_workloads = Workload.objects.filter(
status=WorkloadStatusType.PROCESSED)
non_modified_workloads = list(
non_modified_workloads.values_list('pk', flat=True))
last_pipeline_run = PipelineRun.objects.get_latest()
LOG.debug("Workloads: # modified: %s, # processed: %s, # total: %s",
num_modified, len(non_modified_workloads),
Workload.objects.all().count())
if num_modified == 0:
# No previous workload data yet. Try again later.
LOG.info("No modified workload data yet. Ending background tasks.")
return
# Create new entry in PipelineRun table to store the output of each of
# the background tasks
pipeline_run_obj = PipelineRun(start_time=now(), end_time=None)
pipeline_run_obj.save()
for i, workload in enumerate(modified_workloads):
workload.status = WorkloadStatusType.PROCESSING
workload.save()
wkld_results = Result.objects.filter(workload=workload)
num_wkld_results = wkld_results.count()
workload_name = '{}@{}.{}'.format(workload.dbms.key,
workload.project.name, workload.name)
LOG.info("Starting workload %s (%s/%s, # results: %s)...",
workload_name, i + 1, num_modified, num_wkld_results)
if num_wkld_results == 0:
# delete the workload
LOG.info("Deleting workload %s because it has no results.",
workload_name)
workload.delete()
continue
if num_wkld_results < MIN_WORKLOAD_RESULTS_COUNT:
# Check that there are enough results in the workload
LOG.info(
"Not enough results in workload %s (# results: %s, # required: %s).",
workload_name, num_wkld_results, MIN_WORKLOAD_RESULTS_COUNT)
workload.status = WorkloadStatusType.PROCESSED
workload.save()
continue
LOG.info("Aggregating data for workload %s...", workload_name)
# Aggregate the knob & metric data for this workload
knob_data, metric_data = aggregate_data(wkld_results)
LOG.debug(
"Aggregated knob data: rowlabels=%s, columnlabels=%s, data=%s.",
len(knob_data['rowlabels']), len(knob_data['columnlabels']),
knob_data['data'].shape)
LOG.debug(
"Aggregated metric data: rowlabels=%s, columnlabels=%s, data=%s.",
len(metric_data['rowlabels']), len(metric_data['columnlabels']),
metric_data['data'].shape)
LOG.info("Done aggregating data for workload %s.", workload_name)
num_valid_results = knob_data['data'].shape[0] # pylint: disable=unsubscriptable-object
if num_valid_results < MIN_WORKLOAD_RESULTS_COUNT:
# Check that there are enough valid results in the workload
LOG.info(
"Not enough valid results in workload %s (# valid results: "
"%s, # required: %s).", workload_name, num_valid_results,
MIN_WORKLOAD_RESULTS_COUNT)
workload.status = WorkloadStatusType.PROCESSED
workload.save()
continue
# Knob_data and metric_data are 2D numpy arrays. Convert them into a
# JSON-friendly (nested) lists and then save them as new PipelineData
# objects.
knob_data_copy = copy.deepcopy(knob_data)
knob_data_copy['data'] = knob_data_copy['data'].tolist()
knob_data_copy = JSONUtil.dumps(knob_data_copy)
knob_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.KNOB_DATA,
workload=workload,
data=knob_data_copy,
creation_time=now())
knob_entry.save()
metric_data_copy = copy.deepcopy(metric_data)
metric_data_copy['data'] = metric_data_copy['data'].tolist()
metric_data_copy = JSONUtil.dumps(metric_data_copy)
metric_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.METRIC_DATA,
workload=workload,
data=metric_data_copy,
creation_time=now())
metric_entry.save()
# Execute the Workload Characterization task to compute the list of
# pruned metrics for this workload and save them in a new PipelineData
# object.
LOG.info("Pruning metrics for workload %s...", workload_name)
pruned_metrics = run_workload_characterization(metric_data=metric_data,
dbms=workload.dbms)
LOG.info(
"Done pruning metrics for workload %s (# pruned metrics: %s).\n\n"
"Pruned metrics: %s\n", workload_name, len(pruned_metrics),
pruned_metrics)
pruned_metrics_entry = PipelineData(
pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.PRUNED_METRICS,
workload=workload,
data=JSONUtil.dumps(pruned_metrics),
creation_time=now())
pruned_metrics_entry.save()
# Workload target objective data
ranked_knob_metrics = sorted(
wkld_results.distinct('session').values_list(
'session__target_objective', flat=True).distinct())
LOG.debug("Target objectives for workload %s: %s", workload_name,
', '.join(ranked_knob_metrics))
if KNOB_IDENT_USE_PRUNED_METRICS:
ranked_knob_metrics = sorted(
set(ranked_knob_metrics) + set(pruned_metrics))
# Use the set of metrics to filter the metric_data
metric_idxs = [
i for i, metric_name in enumerate(metric_data['columnlabels'])
if metric_name in ranked_knob_metrics
]
ranked_metric_data = {
'data': metric_data['data'][:, metric_idxs],
'rowlabels': copy.deepcopy(metric_data['rowlabels']),
'columnlabels':
[metric_data['columnlabels'][i] for i in metric_idxs]
}
# Execute the Knob Identification task to compute an ordered list of knobs
# ranked by their impact on the DBMS's performance. Save them in a new
# PipelineData object.
LOG.info(
"Ranking knobs for workload %s (use pruned metric data: %s)...",
workload_name, KNOB_IDENT_USE_PRUNED_METRICS)
sessions = []
for result in wkld_results:
if result.session not in sessions:
sessions.append(result.session)
rank_knob_data = copy.deepcopy(knob_data)
rank_knob_data['data'], rank_knob_data['columnlabels'] =\
DataUtil.clean_knob_data(knob_data['data'], knob_data['columnlabels'], sessions)
ranked_knobs = run_knob_identification(knob_data=rank_knob_data,
metric_data=ranked_metric_data,
dbms=workload.dbms)
LOG.info(
"Done ranking knobs for workload %s (# ranked knobs: %s).\n\n"
"Ranked knobs: %s\n", workload_name, len(ranked_knobs),
ranked_knobs)
ranked_knobs_entry = PipelineData(
pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.RANKED_KNOBS,
workload=workload,
data=JSONUtil.dumps(ranked_knobs),
creation_time=now())
ranked_knobs_entry.save()
workload.status = WorkloadStatusType.PROCESSED
workload.save()
LOG.info("Done processing workload %s (%s/%s).", workload_name, i + 1,
num_modified)
LOG.info("Finished processing %s modified workloads.", num_modified)
non_modified_workloads = Workload.objects.filter(
pk__in=non_modified_workloads)
# Update the latest pipeline data for the non modified workloads to have this pipeline run
PipelineData.objects.filter(workload__in=non_modified_workloads,
pipeline_run=last_pipeline_run)\
.update(pipeline_run=pipeline_run_obj)
# Set the end_timestamp to the current time to indicate that we are done running
# the background tasks
pipeline_run_obj.end_time = now()
pipeline_run_obj.save()
save_execution_time(start_ts, "run_background_tasks")
LOG.info("Finished background tasks (%.0f seconds).",
time.time() - start_ts)
def train_ddpg(result_id):
LOG.info('Add training data to ddpg and train ddpg')
result = Result.objects.get(pk=result_id)
session = Result.objects.get(pk=result_id).session
params = JSONUtil.loads(session.hyperparameters)
session_results = Result.objects.filter(
session=session, creation_time__lt=result.creation_time)
result_info = {}
result_info['newest_result_id'] = result_id
# Extract data from result and previous results
result = Result.objects.filter(pk=result_id)
if len(session_results) == 0:
base_result_id = result_id
prev_result_id = result_id
else:
base_result_id = session_results[0].pk
prev_result_id = session_results[len(session_results) - 1].pk
base_result = Result.objects.filter(pk=base_result_id)
prev_result = Result.objects.filter(pk=prev_result_id)
agg_data = DataUtil.aggregate_data(result)
base_metric_data = (
DataUtil.aggregate_data(base_result))['y_matrix'].flatten()
prev_metric_data = (
DataUtil.aggregate_data(prev_result))['y_matrix'].flatten()
result = Result.objects.get(pk=result_id)
target_objective = result.session.target_objective
prev_obj_idx = [
i for i, n in enumerate(agg_data['y_columnlabels'])
if n == target_objective
]
# Clean metric data
metric_data, metric_labels = clean_metric_data(agg_data['y_matrix'],
agg_data['y_columnlabels'],
session)
metric_data = metric_data.flatten()
metric_scalar = MinMaxScaler().fit(metric_data.reshape(1, -1))
normalized_metric_data = metric_scalar.transform(metric_data.reshape(
1, -1))[0]
# Clean knob data
cleaned_knob_data = clean_knob_data(agg_data['X_matrix'],
agg_data['X_columnlabels'], session)
knob_data = np.array(cleaned_knob_data[0])
knob_labels = np.array(cleaned_knob_data[1])
knob_bounds = np.vstack(
DataUtil.get_knob_bounds(knob_labels.flatten(), session))
knob_data = MinMaxScaler().fit(knob_bounds).transform(knob_data)[0]
knob_num = len(knob_data)
metric_num = len(metric_data)
LOG.info('knob_num: %d, metric_num: %d', knob_num, metric_num)
# Filter ys by current target objective metric
target_obj_idx = [
i for i, n in enumerate(metric_labels) if n == target_objective
]
if len(target_obj_idx) == 0:
raise Exception(('Could not find target objective in metrics '
'(target_obj={})').format(target_objective))
elif len(target_obj_idx) > 1:
raise Exception(
('Found {} instances of target objective in '
'metrics (target_obj={})').format(len(target_obj_idx),
target_objective))
objective = metric_data[target_obj_idx]
base_objective = base_metric_data[prev_obj_idx]
prev_objective = prev_metric_data[prev_obj_idx]
metric_meta = db.target_objectives.get_metric_metadata(
result.session.dbms.pk, result.session.target_objective)
# Calculate the reward
if params['DDPG_SIMPLE_REWARD']:
objective = objective / base_objective
if metric_meta[target_objective].improvement == '(less is better)':
reward = -objective
else:
reward = objective
else:
if metric_meta[target_objective].improvement == '(less is better)':
if objective - base_objective <= 0: # positive reward
reward = (np.square((2 * base_objective - objective) / base_objective) - 1)\
* abs(2 * prev_objective - objective) / prev_objective
else: # negative reward
reward = -(np.square(objective / base_objective) -
1) * objective / prev_objective
else:
if objective - base_objective > 0: # positive reward
reward = (np.square(objective / base_objective) -
1) * objective / prev_objective
else: # negative reward
reward = -(np.square((2 * base_objective - objective) / base_objective) - 1)\
* abs(2 * prev_objective - objective) / prev_objective
LOG.info('reward: %f', reward)
# Update ddpg
ddpg = DDPG(n_actions=knob_num,
n_states=metric_num,
alr=params['DDPG_ACTOR_LEARNING_RATE'],
clr=params['DDPG_CRITIC_LEARNING_RATE'],
gamma=params['DDPG_GAMMA'],
batch_size=params['DDPG_BATCH_SIZE'],
a_hidden_sizes=params['DDPG_ACTOR_HIDDEN_SIZES'],
c_hidden_sizes=params['DDPG_CRITIC_HIDDEN_SIZES'],
use_default=params['DDPG_USE_DEFAULT'])
if session.ddpg_actor_model and session.ddpg_critic_model:
ddpg.set_model(session.ddpg_actor_model, session.ddpg_critic_model)
if session.ddpg_reply_memory:
ddpg.replay_memory.set(session.ddpg_reply_memory)
ddpg.add_sample(normalized_metric_data, knob_data, reward,
normalized_metric_data)
for _ in range(params['DDPG_UPDATE_EPOCHS']):
ddpg.update()
session.ddpg_actor_model, session.ddpg_critic_model = ddpg.get_model()
session.ddpg_reply_memory = ddpg.replay_memory.get()
session.save()
return result_info
def run_background_tasks():
# Find all unique workloads that we have data for
unique_workloads = Workload.objects.all()
if len(unique_workloads) == 0:
# No previous workload data yet. Try again later.
return
## 1. Create a new entry in the PipelineRun table for this
## background task
# Create new PipelineRun object
pipeline_run_obj = PipelineRun(start_time=now(), end_time=None)
# Call save() to commit the transaction and so the new entry is
# actually updated in the PipelineRun table
pipeline_run_obj.save()
## 2. Iterate over all unique workloads.
for workload in unique_workloads:
## 2a. Call first subtask to aggregates the knob & metric data for
## this workload. Save the knob & metric data as (separate)
## entries in the PipelineData table
knob_data, metric_data = aggregate_data(workload)
# Make a copy, convert the 2D numpy array into a JSON-friendly
# (nested) list. Then convert to a JSON string and save as a new
# PipelineData object.
knob_data_copy = copy.deepcopy(knob_data)
knob_data_copy['data'] = knob_data_copy['data'].tolist()
knob_data_copy = JSONUtil.dumps(knob_data_copy)
knob_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.KNOB_DATA,
workload=workload,
data=knob_data_copy,
creation_time=now())
knob_entry.save()
# Do the same thing for the metric data (except change the task_type
# to PipelineTaskType.METRIC_DATA.
metric_data_copy = copy.deepcopy(metric_data)
metric_data_copy['data'] = metric_data_copy['data'].tolist()
metric_data_copy = JSONUtil.dumps(metric_data_copy)
metric_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.METRIC_DATA,
workload=workload,
data=metric_data_copy,
creation_time=now())
metric_entry.save()
## 2b. Call the Workload Characterization subtask to compute
## the list of pruned metrics for this workload
pruned_metrics = run_workload_characterization(metric_data=metric_data)
# Create entry in PipelineData for pruned metrics and save
pruned_metrics_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.PRUNED_METRICS,
workload=workload,
data=JSONUtil.dumps(pruned_metrics),
creation_time=now())
pruned_metrics_entry.save()
## 2c. Call the Knob Identification subtask to compute a ranked list
## of the most impactful knobs
# First, used the pruned metrics to filter the metric_data
pruned_metric_idxs = [i for i, metric_name in enumerate(metric_data['columnlabels'])
if metric_name in pruned_metrics]
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]
}
# Now run the knob identification subtask
ranked_knobs = run_knob_identification(knob_data=knob_data,
metric_data=pruned_metric_data)
# Save ranked knob data
ranked_knobs_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.RANKED_KNOBS,
workload=workload,
data=JSONUtil.dumps(ranked_knobs),
creation_time=now())
ranked_knobs_entry.save()
## 3. Once we are finished computing & storing the pipeline data for each subtask,
## finally set the end_timestamp to the current time to indicate that we are done
## running this background task
pipeline_run_obj.end_time = now()
pipeline_run_obj.save()
def load_data_helper(filtered_pipeline_data, workload, task_type):
pipeline_data = filtered_pipeline_data.get(workload=workload,
task_type=task_type)
LOG.debug("PIPELINE DATA: %s", str(pipeline_data.data))
return JSONUtil.loads(pipeline_data.data)
def configuration_recommendation(target_data):
LOG.info('configuration_recommendation called')
latest_pipeline_run = PipelineRun.objects.get_latest()
if target_data['bad'] is True:
target_data_res = {}
target_data_res['status'] = 'bad'
target_data_res[
'info'] = 'WARNING: no training data, the config is generated randomly'
target_data_res['recommendation'] = target_data['config_recommend']
return target_data_res
# Load mapped workload data
mapped_workload_id = target_data['mapped_workload'][0]
mapped_workload = Workload.objects.get(pk=mapped_workload_id)
workload_knob_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.KNOB_DATA)
workload_knob_data = JSONUtil.loads(workload_knob_data.data)
workload_metric_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.METRIC_DATA)
workload_metric_data = JSONUtil.loads(workload_metric_data.data)
X_workload = np.array(workload_knob_data['data'])
X_columnlabels = np.array(workload_knob_data['columnlabels'])
y_workload = np.array(workload_metric_data['data'])
y_columnlabels = np.array(workload_metric_data['columnlabels'])
rowlabels_workload = np.array(workload_metric_data['rowlabels'])
# Target workload data
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
X_target = target_data['X_matrix']
y_target = target_data['y_matrix']
rowlabels_target = np.array(target_data['rowlabels'])
if not np.array_equal(X_columnlabels, target_data['X_columnlabels']):
raise Exception(('The workload and target data should have '
'identical X columnlabels (sorted knob names)'))
if not np.array_equal(y_columnlabels, target_data['y_columnlabels']):
raise Exception(('The workload and target data should have '
'identical y columnlabels (sorted metric names)'))
# Filter Xs by top 10 ranked knobs
ranked_knobs = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.RANKED_KNOBS)
ranked_knobs = JSONUtil.loads(ranked_knobs.data)[:IMPORTANT_KNOB_NUMBER]
ranked_knob_idxs = [
i for i, cl in enumerate(X_columnlabels) if cl in ranked_knobs
]
X_workload = X_workload[:, ranked_knob_idxs]
X_target = X_target[:, ranked_knob_idxs]
X_columnlabels = X_columnlabels[ranked_knob_idxs]
# Filter ys by current target objective metric
target_objective = newest_result.session.target_objective
target_obj_idx = [
i for i, cl in enumerate(y_columnlabels) if cl == target_objective
]
if len(target_obj_idx) == 0:
raise Exception(('Could not find target objective in metrics '
'(target_obj={})').format(target_objective))
elif len(target_obj_idx) > 1:
raise Exception(
('Found {} instances of target objective in '
'metrics (target_obj={})').format(len(target_obj_idx),
target_objective))
metric_meta = MetricCatalog.objects.get_metric_meta(
newest_result.session.dbms, newest_result.session.target_objective)
if metric_meta[target_objective].improvement == '(less is better)':
lessisbetter = True
else:
lessisbetter = False
y_workload = y_workload[:, target_obj_idx]
y_target = y_target[:, target_obj_idx]
y_columnlabels = y_columnlabels[target_obj_idx]
# Combine duplicate rows in the target/workload data (separately)
X_workload, y_workload, rowlabels_workload = DataUtil.combine_duplicate_rows(
X_workload, y_workload, rowlabels_workload)
X_target, y_target, rowlabels_target = DataUtil.combine_duplicate_rows(
X_target, y_target, rowlabels_target)
# Delete any rows that appear in both the workload data and the target
# data from the workload data
dups_filter = np.ones(X_workload.shape[0], dtype=bool)
target_row_tups = [tuple(row) for row in X_target]
for i, row in enumerate(X_workload):
if tuple(row) in target_row_tups:
dups_filter[i] = False
X_workload = X_workload[dups_filter, :]
y_workload = y_workload[dups_filter, :]
rowlabels_workload = rowlabels_workload[dups_filter]
# Combine target & workload Xs for preprocessing
X_matrix = np.vstack([X_target, X_workload])
# Dummy encode categorial variables
categorical_info = DataUtil.dummy_encoder_helper(X_columnlabels,
mapped_workload.dbms)
dummy_encoder = DummyEncoder(categorical_info['n_values'],
categorical_info['categorical_features'],
categorical_info['cat_columnlabels'],
categorical_info['noncat_columnlabels'])
X_matrix = dummy_encoder.fit_transform(X_matrix)
# below two variables are needed for correctly determing max/min on dummies
binary_index_set = set(categorical_info['binary_vars'])
total_dummies = dummy_encoder.total_dummies()
# Scale to N(0, 1)
X_scaler = StandardScaler()
X_scaled = X_scaler.fit_transform(X_matrix)
if y_target.shape[0] < 5: # FIXME
# FIXME (dva): if there are fewer than 5 target results so far
# then scale the y values (metrics) using the workload's
# y_scaler. I'm not sure if 5 is the right cutoff.
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_matrix = np.vstack([y_target, y_workload])
y_scaled = y_workload_scaler.fit_transform(y_matrix)
else:
# FIXME (dva): otherwise try to compute a separate y_scaler for
# the target and scale them separately.
try:
y_target_scaler = StandardScaler()
y_workload_scaler = StandardScaler()
y_target_scaled = y_target_scaler.fit_transform(y_target)
y_workload_scaled = y_workload_scaler.fit_transform(y_workload)
y_scaled = np.vstack([y_target_scaled, y_workload_scaled])
except ValueError:
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_scaled = y_workload_scaler.fit_transform(y_target)
# Set up constraint helper
constraint_helper = ParamConstraintHelper(
scaler=X_scaler,
encoder=dummy_encoder,
binary_vars=categorical_info['binary_vars'],
init_flip_prob=INIT_FLIP_PROB,
flip_prob_decay=FLIP_PROB_DECAY)
# FIXME (dva): check if these are good values for the ridge
# ridge = np.empty(X_scaled.shape[0])
# ridge[:X_target.shape[0]] = 0.01
# ridge[X_target.shape[0]:] = 0.1
# FIXME: we should generate more samples and use a smarter sampling
# technique
num_samples = NUM_SAMPLES
X_samples = np.empty((num_samples, X_scaled.shape[1]))
X_min = np.empty(X_scaled.shape[1])
X_max = np.empty(X_scaled.shape[1])
knobs_mem = KnobCatalog.objects.filter(dbms=newest_result.session.dbms,
tunable=True,
resource=1)
knobs_mem_catalog = {k.name: k for k in knobs_mem}
mem_max = newest_result.workload.hardware.memory
X_mem = np.zeros([1, X_scaled.shape[1]])
X_default = np.empty(X_scaled.shape[1])
# Get default knob values
for i, k_name in enumerate(X_columnlabels):
k = KnobCatalog.objects.filter(dbms=newest_result.session.dbms,
name=k_name)[0]
X_default[i] = k.default
X_default_scaled = X_scaler.transform(
X_default.reshape(1, X_default.shape[0]))[0]
# Determine min/max for knob values
for i in range(X_scaled.shape[1]):
if i < total_dummies or i in binary_index_set:
col_min = 0
col_max = 1
else:
col_min = X_scaled[:, i].min()
col_max = X_scaled[:, i].max()
if X_columnlabels[i] in knobs_mem_catalog:
X_mem[0][i] = mem_max * 1024 * 1024 * 1024 # mem_max GB
col_max = min(col_max, X_scaler.transform(X_mem)[0][i])
# Set min value to the default value
# FIXME: support multiple methods can be selected by users
col_min = X_default_scaled[i]
X_min[i] = col_min
X_max[i] = col_max
X_samples[:, i] = np.random.rand(num_samples) * (col_max -
col_min) + col_min
# Maximize the throughput, moreisbetter
# Use gradient descent to minimize -throughput
if not lessisbetter:
y_scaled = -y_scaled
q = queue.PriorityQueue()
for x in range(0, y_scaled.shape[0]):
q.put((y_scaled[x][0], x))
i = 0
while i < TOP_NUM_CONFIG:
try:
item = q.get_nowait()
# Tensorflow get broken if we use the training data points as
# starting points for GPRGD. We add a small bias for the
# starting points. GPR_EPS default value is 0.001
# if the starting point is X_max, we minus a small bias to
# make sure it is within the range.
dist = sum(np.square(X_max - X_scaled[item[1]]))
if dist < 0.001:
X_samples = np.vstack(
(X_samples, X_scaled[item[1]] - abs(GPR_EPS)))
else:
X_samples = np.vstack(
(X_samples, X_scaled[item[1]] + abs(GPR_EPS)))
i = i + 1
except queue.Empty:
break
model = GPRGD(length_scale=DEFAULT_LENGTH_SCALE,
magnitude=DEFAULT_MAGNITUDE,
max_train_size=MAX_TRAIN_SIZE,
batch_size=BATCH_SIZE,
num_threads=NUM_THREADS,
learning_rate=DEFAULT_LEARNING_RATE,
epsilon=DEFAULT_EPSILON,
max_iter=MAX_ITER,
sigma_multiplier=DEFAULT_SIGMA_MULTIPLIER,
mu_multiplier=DEFAULT_MU_MULTIPLIER)
model.fit(X_scaled, y_scaled, X_min, X_max, ridge=DEFAULT_RIDGE)
res = model.predict(X_samples, constraint_helper=constraint_helper)
best_config_idx = np.argmin(res.minl.ravel())
best_config = res.minl_conf[best_config_idx, :]
best_config = X_scaler.inverse_transform(best_config)
# Decode one-hot encoding into categorical knobs
best_config = dummy_encoder.inverse_transform(best_config)
# Although we have max/min limits in the GPRGD training session, it may
# lose some precisions. e.g. 0.99..99 >= 1.0 may be True on the scaled data,
# when we inversely transform the scaled data, the different becomes much larger
# and cannot be ignored. Here we check the range on the original data
# directly, and make sure the recommended config lies within the range
X_min_inv = X_scaler.inverse_transform(X_min)
X_max_inv = X_scaler.inverse_transform(X_max)
best_config = np.minimum(best_config, X_max_inv)
best_config = np.maximum(best_config, X_min_inv)
conf_map = {k: best_config[i] for i, k in enumerate(X_columnlabels)}
conf_map_res = {}
conf_map_res['status'] = 'good'
conf_map_res['recommendation'] = conf_map
conf_map_res['info'] = 'INFO: training data size is {}'.format(
X_scaled.shape[0])
return conf_map_res
def configuration_recommendation(target_data):
LOG.info('configuration_recommendation called')
latest_pipeline_run = PipelineRun.objects.get_latest()
if target_data['bad'] is True:
target_data_res = {}
target_data_res['status'] = 'bad'
target_data_res['info'] = 'WARNING: no training data, the config is generated randomly'
target_data_res['recommendation'] = target_data['config_recommend']
return target_data_res
# Load mapped workload data
mapped_workload_id = target_data['mapped_workload'][0]
mapped_workload = Workload.objects.get(pk=mapped_workload_id)
workload_knob_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.KNOB_DATA)
workload_knob_data = JSONUtil.loads(workload_knob_data.data)
workload_metric_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.METRIC_DATA)
workload_metric_data = JSONUtil.loads(workload_metric_data.data)
X_workload = np.array(workload_knob_data['data'])
X_columnlabels = np.array(workload_knob_data['columnlabels'])
y_workload = np.array(workload_metric_data['data'])
y_columnlabels = np.array(workload_metric_data['columnlabels'])
rowlabels_workload = np.array(workload_metric_data['rowlabels'])
# Target workload data
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
X_target = target_data['X_matrix']
y_target = target_data['y_matrix']
rowlabels_target = np.array(target_data['rowlabels'])
if not np.array_equal(X_columnlabels, target_data['X_columnlabels']):
raise Exception(('The workload and target data should have '
'identical X columnlabels (sorted knob names)'))
if not np.array_equal(y_columnlabels, target_data['y_columnlabels']):
raise Exception(('The workload and target data should have '
'identical y columnlabels (sorted metric names)'))
# Filter Xs by top 10 ranked knobs
ranked_knobs = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.RANKED_KNOBS)
ranked_knobs = JSONUtil.loads(ranked_knobs.data)[:10] # FIXME
ranked_knob_idxs = [i for i, cl in enumerate(X_columnlabels) if cl in ranked_knobs]
X_workload = X_workload[:, ranked_knob_idxs]
X_target = X_target[:, ranked_knob_idxs]
X_columnlabels = X_columnlabels[ranked_knob_idxs]
# Filter ys by current target objective metric
target_objective = newest_result.session.target_objective
target_obj_idx = [i for i, cl in enumerate(y_columnlabels) if cl == target_objective]
if len(target_obj_idx) == 0:
raise Exception(('Could not find target objective in metrics '
'(target_obj={})').format(target_objective))
elif len(target_obj_idx) > 1:
raise Exception(('Found {} instances of target objective in '
'metrics (target_obj={})').format(len(target_obj_idx),
target_objective))
y_workload = y_workload[:, target_obj_idx]
y_target = y_target[:, target_obj_idx]
y_columnlabels = y_columnlabels[target_obj_idx]
# Combine duplicate rows in the target/workload data (separately)
X_workload, y_workload, rowlabels_workload = DataUtil.combine_duplicate_rows(
X_workload, y_workload, rowlabels_workload)
X_target, y_target, rowlabels_target = DataUtil.combine_duplicate_rows(
X_target, y_target, rowlabels_target)
# Delete any rows that appear in both the workload data and the target
# data from the workload data
dups_filter = np.ones(X_workload.shape[0], dtype=bool)
target_row_tups = [tuple(row) for row in X_target]
for i, row in enumerate(X_workload):
if tuple(row) in target_row_tups:
dups_filter[i] = False
X_workload = X_workload[dups_filter, :]
y_workload = y_workload[dups_filter, :]
rowlabels_workload = rowlabels_workload[dups_filter]
# Combine target & workload Xs then scale
X_matrix = np.vstack([X_target, X_workload])
X_scaler = StandardScaler()
X_scaled = X_scaler.fit_transform(X_matrix)
if y_target.shape[0] < 5: # FIXME
# FIXME (dva): if there are fewer than 5 target results so far
# then scale the y values (metrics) using the workload's
# y_scaler. I'm not sure if 5 is the right cutoff.
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_matrix = np.vstack([y_target, y_workload])
y_scaled = y_workload_scaler.fit_transform(y_matrix)
else:
# FIXME (dva): otherwise try to compute a separate y_scaler for
# the target and scale them separately.
try:
y_target_scaler = StandardScaler()
y_workload_scaler = StandardScaler()
y_target_scaled = y_target_scaler.fit_transform(y_target)
y_workload_scaled = y_workload_scaler.fit_transform(y_workload)
y_scaled = np.vstack([y_target_scaled, y_workload_scaled])
except ValueError:
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_scaled = y_workload_scaler.fit_transform(y_target)
# FIXME (dva): check if these are good values for the ridge
ridge = np.empty(X_scaled.shape[0])
ridge[:X_target.shape[0]] = 0.01
ridge[X_target.shape[0]:] = 0.1
# FIXME: we should generate more samples and use a smarter sampling
# technique
num_samples = 20
X_samples = np.empty((num_samples, X_scaled.shape[1]))
X_min = np.empty(X_scaled.shape[1])
X_max = np.empty(X_scaled.shape[1])
for i in range(X_scaled.shape[1]):
col_min = X_scaled[:, i].min()
col_max = X_scaled[:, i].max()
X_min[i] = col_min
X_max[i] = col_max
X_samples[:, i] = np.random.rand(
num_samples) * (col_max - col_min) + col_min
# FIXME: Maximize the throughput, hardcode
# Use gradient descent to minimize -throughput
y_scaled = -y_scaled
model = GPRGD()
model.fit(X_scaled, y_scaled, X_min, X_max, ridge)
res = model.predict(X_samples)
# FIXME: whether we select the min/max for the best config depends
# on the target objective
best_config_idx = np.argmin(res.minl.ravel())
best_config = res.minl_conf[best_config_idx, :]
best_config = X_scaler.inverse_transform(best_config)
conf_map = {k: best_config[i] for i, k in enumerate(X_columnlabels)}
conf_map_res = {}
conf_map_res['status'] = 'good'
conf_map_res['recommendation'] = conf_map
conf_map_res['info'] = 'INFO: training data size is {}'.format(X_scaled.shape[0])
return conf_map_res
def load_data_helper(filtered_pipeline_data, workload, task_type):
pipeline_data = filtered_pipeline_data.get(workload=workload,
task_type=task_type)
print "PIPELINE DATA: {}".format(pipeline_data.data)
return JSONUtil.loads(pipeline_data.data)
def configuration_recommendation(target_data):
if target_data['scores'] is None:
raise NotImplementedError('Implement me!')
best_wkld_id = target_data['mapped_workload'][0]
# Load specific workload data
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
target_obj = newest_result.application.target_objective
dbms_id = newest_result.dbms.pk
hw_id = newest_result.application.hardware.pk
agg_data = PipelineResult.get_latest(dbms_id, hw_id,
PipelineTaskType.AGGREGATED_DATA)
if agg_data is None:
return None
data_map = JSONUtil.loads(agg_data.value)
if best_wkld_id not in data_map['data']:
raise Exception(('Cannot find mapped workload'
'(id={}) in aggregated data').format(best_wkld_id))
workload_data = np.load(data_map['data'][best_wkld_id])
# Mapped workload data
X_wkld_matrix = workload_data['X_matrix']
y_wkld_matrix = workload_data['y_matrix']
wkld_rowlabels = workload_data['rowlabels']
X_columnlabels = workload_data['X_columnlabels']
y_columnlabels = workload_data['y_columnlabels']
# Target workload data
X_target_matrix = target_data['X_matrix']
y_target_matrix = target_data['y_matrix']
target_rowlabels = target_data['rowlabels']
if not np.array_equal(X_columnlabels, target_data['X_columnlabels']):
raise Exception(('The workload and target data should have '
'identical X columnlabels (sorted knob names)'))
if not np.array_equal(y_columnlabels, target_data['y_columnlabels']):
raise Exception(('The workload and target data should have '
'identical y columnlabels (sorted metric names)'))
# Filter knobs
ranked_knobs = JSONUtil.loads(
PipelineResult.get_latest(
dbms_id, hw_id, PipelineTaskType.RANKED_KNOBS).value)[:10] # FIXME
X_idxs = [
i for i in range(X_columnlabels.shape[0])
if X_columnlabels[i] in ranked_knobs
]
X_wkld_matrix = X_wkld_matrix[:, X_idxs]
X_target_matrix = X_target_matrix[:, X_idxs]
X_columnlabels = X_columnlabels[X_idxs]
# Filter metrics by current target objective metric
y_idx = [
i for i in range(y_columnlabels.shape[0])
if y_columnlabels[i] == target_obj
]
if len(y_idx) == 0:
raise Exception(('Could not find target objective in metrics '
'(target_obj={})').format(target_obj))
elif len(y_idx) > 1:
raise Exception(
('Found {} instances of target objective in '
'metrics (target_obj={})').format(len(y_idx), target_obj))
y_wkld_matrix = y_wkld_matrix[:, y_idx]
y_target_matrix = y_target_matrix[:, y_idx]
y_columnlabels = y_columnlabels[y_idx]
# Combine duplicate rows in the target/workload data (separately)
X_wkld_matrix, y_wkld_matrix, wkld_rowlabels = DataUtil.combine_duplicate_rows(
X_wkld_matrix, y_wkld_matrix, wkld_rowlabels)
X_target_matrix, y_target_matrix, target_rowlabels = DataUtil.combine_duplicate_rows(
X_target_matrix, y_target_matrix, target_rowlabels)
# Delete any rows that appear in both the workload data and the target
# data from the workload data
dups_filter = np.ones(X_wkld_matrix.shape[0], dtype=bool)
target_row_tups = [tuple(row) for row in X_target_matrix]
for i, row in enumerate(X_wkld_matrix):
if tuple(row) in target_row_tups:
dups_filter[i] = False
X_wkld_matrix = X_wkld_matrix[dups_filter, :]
y_wkld_matrix = y_wkld_matrix[dups_filter, :]
wkld_rowlabels = wkld_rowlabels[dups_filter]
# Combine Xs and scale
X_matrix = np.vstack([X_target_matrix, X_wkld_matrix])
X_scaler = StandardScaler()
X_scaled = X_scaler.fit_transform(X_matrix)
if y_target_matrix.shape[0] < 5: # FIXME
y_target_scaler = None
y_wkld_scaler = StandardScaler()
y_matrix = np.vstack([y_target_matrix, y_wkld_matrix])
y_scaled = y_wkld_scaler.fit_transform(y_matrix)
else:
try:
y_target_scaler = StandardScaler()
y_wkld_scaler = StandardScaler()
y_target_scaled = y_target_scaler.fit_transform(y_target_matrix)
y_wkld_scaled = y_wkld_scaler.fit_transform(y_wkld_matrix)
y_scaled = np.vstack([y_target_scaled, y_wkld_scaled])
except ValueError:
y_target_scaler = None
y_wkld_scaler = StandardScaler()
y_matrix = np.vstack([y_target_matrix, y_wkld_matrix])
y_scaled = y_wkld_scaler.fit_transform(y_matrix)
ridge = np.empty(X_scaled.shape[0])
ridge[:X_target_matrix.shape[0]] = 0.01
ridge[X_target_matrix.shape[0]:] = 0.1
# FIXME
num_samples = 5
X_samples = np.empty((num_samples, X_scaled.shape[1]))
for i in range(X_scaled.shape[1]):
col_min = X_scaled[:, i].min()
col_max = X_scaled[:, i].max()
X_samples[:, i] = np.random.rand(num_samples) * (col_max -
col_min) + col_min
model = GPR_GD()
model.fit(X_scaled, y_scaled, ridge)
res = model.predict(X_samples)
best_idx = np.argmin(res.minL.ravel())
best_conf = res.minL_conf[best_idx, :]
best_conf = X_scaler.inverse_transform(best_conf)
conf_map = {k: best_conf[i] for i, k in enumerate(X_columnlabels)}
return conf_map
def run_background_tasks():
start_ts = time.time()
LOG.debug("Starting background tasks")
# Find modified and not modified workloads, we only have to calculate for the
# modified workloads.
modified_workloads = Workload.objects.filter(
status=WorkloadStatusType.MODIFIED)
non_modified_workloads = Workload.objects.filter(
status=WorkloadStatusType.PROCESSED)
non_modified_workloads = list(
non_modified_workloads.values_list('pk', flat=True))
last_pipeline_run = PipelineRun.objects.get_latest()
if len(modified_workloads) == 0:
# No previous workload data yet. Try again later.
LOG.debug("No workload data yet. Ending background tasks")
return
# Create new entry in PipelineRun table to store the output of each of
# the background tasks
pipeline_run_obj = PipelineRun(start_time=now(), end_time=None)
pipeline_run_obj.save()
for workload in modified_workloads:
wkld_results = Result.objects.filter(workload=workload)
if wkld_results.exists() is False:
# delete the workload
LOG.debug("Deleting workload %d because it has no results.",
workload.id)
workload.delete()
continue
# Check that there are enough results in the workload
if wkld_results.count() < MIN_WORKLOAD_RESULTS_COUNT:
LOG.debug("Not enough results in workload %d (only %d results).",
workload.id, wkld_results.count())
continue
workload.status = WorkloadStatusType.PROCESSING
workload.save()
LOG.debug("Aggregating data for workload %d", workload.id)
# Aggregate the knob & metric data for this workload
knob_data, metric_data = aggregate_data(wkld_results)
LOG.debug("knob_data: %s", str(knob_data))
LOG.debug("metric_data: %s", str(metric_data))
# Knob_data and metric_data are 2D numpy arrays. Convert them into a
# JSON-friendly (nested) lists and then save them as new PipelineData
# objects.
knob_data_copy = copy.deepcopy(knob_data)
knob_data_copy['data'] = knob_data_copy['data'].tolist()
knob_data_copy = JSONUtil.dumps(knob_data_copy)
knob_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.KNOB_DATA,
workload=workload,
data=knob_data_copy,
creation_time=now())
knob_entry.save()
metric_data_copy = copy.deepcopy(metric_data)
metric_data_copy['data'] = metric_data_copy['data'].tolist()
metric_data_copy = JSONUtil.dumps(metric_data_copy)
metric_entry = PipelineData(pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.METRIC_DATA,
workload=workload,
data=metric_data_copy,
creation_time=now())
metric_entry.save()
# Execute the Workload Characterization task to compute the list of
# pruned metrics for this workload and save them in a new PipelineData
# object.
LOG.debug("Pruning metrics for workload %d.", workload.id)
pruned_metrics = run_workload_characterization(metric_data=metric_data)
LOG.debug("pruned_metrics: %s", str(pruned_metrics))
pruned_metrics_entry = PipelineData(
pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.PRUNED_METRICS,
workload=workload,
data=JSONUtil.dumps(pruned_metrics),
creation_time=now())
pruned_metrics_entry.save()
# Use the pruned metrics to filter the metric_data
pruned_metric_idxs = [
i for i, metric_name in enumerate(metric_data['columnlabels'])
if metric_name in pruned_metrics
]
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]
}
# Execute the Knob Identification task to compute an ordered list of knobs
# ranked by their impact on the DBMS's performance. Save them in a new
# PipelineData object.
LOG.debug("Ranking knobs for workload %d.", workload.id)
ranked_knobs = run_knob_identification(knob_data=knob_data,
metric_data=pruned_metric_data,
dbms=workload.dbms)
LOG.debug("ranked_knobs: %s", str(ranked_knobs))
ranked_knobs_entry = PipelineData(
pipeline_run=pipeline_run_obj,
task_type=PipelineTaskType.RANKED_KNOBS,
workload=workload,
data=JSONUtil.dumps(ranked_knobs),
creation_time=now())
ranked_knobs_entry.save()
workload.status = WorkloadStatusType.PROCESSED
workload.save()
LOG.debug("Finished processing modified workloads")
non_modified_workloads = Workload.objects.filter(
pk__in=non_modified_workloads)
# Update the latest pipeline data for the non modified workloads to have this pipeline run
PipelineData.objects.filter(workload__in=non_modified_workloads,
pipeline_run=last_pipeline_run)\
.update(pipeline_run=pipeline_run_obj)
# Set the end_timestamp to the current time to indicate that we are done running
# the background tasks
pipeline_run_obj.end_time = now()
pipeline_run_obj.save()
LOG.debug("Finished background tasks")
save_execution_time(start_ts, "run_background_tasks")
def combine_workload(target_data):
# Load mapped workload data
mapped_workload_id = target_data['mapped_workload'][0]
latest_pipeline_run = PipelineRun.objects.get(
pk=target_data['pipeline_run'])
mapped_workload = Workload.objects.get(pk=mapped_workload_id)
workload_knob_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.KNOB_DATA)
workload_knob_data = JSONUtil.loads(workload_knob_data.data)
workload_metric_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.METRIC_DATA)
workload_metric_data = JSONUtil.loads(workload_metric_data.data)
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
session = newest_result.session
params = JSONUtil.loads(session.hyperparameters)
cleaned_workload_knob_data = clean_knob_data(
workload_knob_data["data"], workload_knob_data["columnlabels"],
newest_result.session)
X_workload = np.array(cleaned_workload_knob_data[0])
X_columnlabels = np.array(cleaned_workload_knob_data[1])
y_workload = np.array(workload_metric_data['data'])
y_columnlabels = np.array(workload_metric_data['columnlabels'])
rowlabels_workload = np.array(workload_metric_data['rowlabels'])
# Target workload data
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
X_target = target_data['X_matrix']
y_target = target_data['y_matrix']
rowlabels_target = np.array(target_data['rowlabels'])
if not np.array_equal(X_columnlabels, target_data['X_columnlabels']):
raise Exception(('The workload and target data should have '
'identical X columnlabels (sorted knob names)'),
X_columnlabels, target_data['X_columnlabels'])
if not np.array_equal(y_columnlabels, target_data['y_columnlabels']):
raise Exception(('The workload and target data should have '
'identical y columnlabels (sorted metric names)'),
y_columnlabels, target_data['y_columnlabels'])
# Filter Xs by top 10 ranked knobs
ranked_knobs = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.RANKED_KNOBS)
ranked_knobs = JSONUtil.loads(
ranked_knobs.data)[:params['IMPORTANT_KNOB_NUMBER']]
ranked_knob_idxs = [
i for i, cl in enumerate(X_columnlabels) if cl in ranked_knobs
]
X_workload = X_workload[:, ranked_knob_idxs]
X_target = X_target[:, ranked_knob_idxs]
X_columnlabels = X_columnlabels[ranked_knob_idxs]
# Filter ys by current target objective metric
target_objective = newest_result.session.target_objective
target_obj_idx = [
i for i, cl in enumerate(y_columnlabels) if cl == target_objective
]
if len(target_obj_idx) == 0:
raise Exception(('Could not find target objective in metrics '
'(target_obj={})').format(target_objective))
elif len(target_obj_idx) > 1:
raise Exception(
('Found {} instances of target objective in '
'metrics (target_obj={})').format(len(target_obj_idx),
target_objective))
y_workload = y_workload[:, target_obj_idx]
y_target = y_target[:, target_obj_idx]
y_columnlabels = y_columnlabels[target_obj_idx]
# Combine duplicate rows in the target/workload data (separately)
X_workload, y_workload, rowlabels_workload = DataUtil.combine_duplicate_rows(
X_workload, y_workload, rowlabels_workload)
X_target, y_target, rowlabels_target = DataUtil.combine_duplicate_rows(
X_target, y_target, rowlabels_target)
# Delete any rows that appear in both the workload data and the target
# data from the workload data
dups_filter = np.ones(X_workload.shape[0], dtype=bool)
target_row_tups = [tuple(row) for row in X_target]
for i, row in enumerate(X_workload):
if tuple(row) in target_row_tups:
dups_filter[i] = False
X_workload = X_workload[dups_filter, :]
y_workload = y_workload[dups_filter, :]
rowlabels_workload = rowlabels_workload[dups_filter]
# Combine target & workload Xs for preprocessing
X_matrix = np.vstack([X_target, X_workload])
# Dummy encode categorial variables
if ENABLE_DUMMY_ENCODER:
categorical_info = DataUtil.dummy_encoder_helper(
X_columnlabels, mapped_workload.dbms)
dummy_encoder = DummyEncoder(categorical_info['n_values'],
categorical_info['categorical_features'],
categorical_info['cat_columnlabels'],
categorical_info['noncat_columnlabels'])
X_matrix = dummy_encoder.fit_transform(X_matrix)
binary_encoder = categorical_info['binary_vars']
# below two variables are needed for correctly determing max/min on dummies
binary_index_set = set(categorical_info['binary_vars'])
total_dummies = dummy_encoder.total_dummies()
else:
dummy_encoder = None
binary_encoder = None
binary_index_set = []
total_dummies = 0
# Scale to N(0, 1)
X_scaler = StandardScaler()
X_scaled = X_scaler.fit_transform(X_matrix)
if y_target.shape[0] < 5: # FIXME
# FIXME (dva): if there are fewer than 5 target results so far
# then scale the y values (metrics) using the workload's
# y_scaler. I'm not sure if 5 is the right cutoff.
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_matrix = np.vstack([y_target, y_workload])
y_scaled = y_workload_scaler.fit_transform(y_matrix)
else:
# FIXME (dva): otherwise try to compute a separate y_scaler for
# the target and scale them separately.
try:
y_target_scaler = StandardScaler()
y_workload_scaler = StandardScaler()
y_target_scaled = y_target_scaler.fit_transform(y_target)
y_workload_scaled = y_workload_scaler.fit_transform(y_workload)
y_scaled = np.vstack([y_target_scaled, y_workload_scaled])
except ValueError:
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_scaled = y_workload_scaler.fit_transform(y_target)
metric_meta = db.target_objectives.get_metric_metadata(
newest_result.session.dbms.pk, newest_result.session.target_objective)
lessisbetter = metric_meta[
target_objective].improvement == db.target_objectives.LESS_IS_BETTER
# Maximize the throughput, moreisbetter
# Use gradient descent to minimize -throughput
if not lessisbetter:
y_scaled = -y_scaled
# Set up constraint helper
constraint_helper = ParamConstraintHelper(
scaler=X_scaler,
encoder=dummy_encoder,
binary_vars=binary_encoder,
init_flip_prob=params['INIT_FLIP_PROB'],
flip_prob_decay=params['FLIP_PROB_DECAY'])
# FIXME (dva): check if these are good values for the ridge
# ridge = np.empty(X_scaled.shape[0])
# ridge[:X_target.shape[0]] = 0.01
# ridge[X_target.shape[0]:] = 0.1
X_min = np.empty(X_scaled.shape[1])
X_max = np.empty(X_scaled.shape[1])
X_scaler_matrix = np.zeros([1, X_scaled.shape[1]])
session_knobs = SessionKnob.objects.get_knobs_for_session(
newest_result.session)
# Set min/max for knob values
for i in range(X_scaled.shape[1]):
if i < total_dummies or i in binary_index_set:
col_min = 0
col_max = 1
else:
col_min = X_scaled[:, i].min()
col_max = X_scaled[:, i].max()
for knob in session_knobs:
if X_columnlabels[i] == knob["name"]:
X_scaler_matrix[0][i] = knob["minval"]
col_min = X_scaler.transform(X_scaler_matrix)[0][i]
X_scaler_matrix[0][i] = knob["maxval"]
col_max = X_scaler.transform(X_scaler_matrix)[0][i]
X_min[i] = col_min
X_max[i] = col_max
return X_columnlabels, X_scaler, X_scaled, y_scaled, X_max, X_min,\
dummy_encoder, constraint_helper
def load_data_helper(filtered_pipeline_data, workload, task_type):
pipeline_data = filtered_pipeline_data.get(workload=workload,
task_type=task_type)
LOG.debug("PIPELINE DATA: %s", str(pipeline_data.data))
return JSONUtil.loads(pipeline_data.data)
def configuration_recommendation(target_data):
LOG.info('configuration_recommendation called')
latest_pipeline_run = PipelineRun.objects.get_latest()
if target_data['bad'] is True:
target_data_res = {}
target_data_res['status'] = 'bad'
target_data_res['info'] = 'WARNING: no training data, the config is generated randomly'
target_data_res['recommendation'] = target_data['config_recommend']
return target_data_res
# Load mapped workload data
mapped_workload_id = target_data['mapped_workload'][0]
mapped_workload = Workload.objects.get(pk=mapped_workload_id)
workload_knob_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.KNOB_DATA)
workload_knob_data = JSONUtil.loads(workload_knob_data.data)
workload_metric_data = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.METRIC_DATA)
workload_metric_data = JSONUtil.loads(workload_metric_data.data)
X_workload = np.array(workload_knob_data['data'])
X_columnlabels = np.array(workload_knob_data['columnlabels'])
y_workload = np.array(workload_metric_data['data'])
y_columnlabels = np.array(workload_metric_data['columnlabels'])
rowlabels_workload = np.array(workload_metric_data['rowlabels'])
# Target workload data
newest_result = Result.objects.get(pk=target_data['newest_result_id'])
X_target = target_data['X_matrix']
y_target = target_data['y_matrix']
rowlabels_target = np.array(target_data['rowlabels'])
if not np.array_equal(X_columnlabels, target_data['X_columnlabels']):
raise Exception(('The workload and target data should have '
'identical X columnlabels (sorted knob names)'))
if not np.array_equal(y_columnlabels, target_data['y_columnlabels']):
raise Exception(('The workload and target data should have '
'identical y columnlabels (sorted metric names)'))
# Filter Xs by top 10 ranked knobs
ranked_knobs = PipelineData.objects.get(
pipeline_run=latest_pipeline_run,
workload=mapped_workload,
task_type=PipelineTaskType.RANKED_KNOBS)
ranked_knobs = JSONUtil.loads(ranked_knobs.data)[:IMPORTANT_KNOB_NUMBER]
ranked_knob_idxs = [i for i, cl in enumerate(X_columnlabels) if cl in ranked_knobs]
X_workload = X_workload[:, ranked_knob_idxs]
X_target = X_target[:, ranked_knob_idxs]
X_columnlabels = X_columnlabels[ranked_knob_idxs]
# Filter ys by current target objective metric
target_objective = newest_result.session.target_objective
target_obj_idx = [i for i, cl in enumerate(y_columnlabels) if cl == target_objective]
if len(target_obj_idx) == 0:
raise Exception(('Could not find target objective in metrics '
'(target_obj={})').format(target_objective))
elif len(target_obj_idx) > 1:
raise Exception(('Found {} instances of target objective in '
'metrics (target_obj={})').format(len(target_obj_idx),
target_objective))
metric_meta = MetricCatalog.objects.get_metric_meta(newest_result.session.dbms,
newest_result.session.target_objective)
if metric_meta[target_objective] == '(less is better)':
lessisbetter = True
else:
lessisbetter = False
y_workload = y_workload[:, target_obj_idx]
y_target = y_target[:, target_obj_idx]
y_columnlabels = y_columnlabels[target_obj_idx]
# Combine duplicate rows in the target/workload data (separately)
X_workload, y_workload, rowlabels_workload = DataUtil.combine_duplicate_rows(
X_workload, y_workload, rowlabels_workload)
X_target, y_target, rowlabels_target = DataUtil.combine_duplicate_rows(
X_target, y_target, rowlabels_target)
# Delete any rows that appear in both the workload data and the target
# data from the workload data
dups_filter = np.ones(X_workload.shape[0], dtype=bool)
target_row_tups = [tuple(row) for row in X_target]
for i, row in enumerate(X_workload):
if tuple(row) in target_row_tups:
dups_filter[i] = False
X_workload = X_workload[dups_filter, :]
y_workload = y_workload[dups_filter, :]
rowlabels_workload = rowlabels_workload[dups_filter]
# Combine target & workload Xs for preprocessing
X_matrix = np.vstack([X_target, X_workload])
# Dummy encode categorial variables
categorical_info = DataUtil.dummy_encoder_helper(X_columnlabels,
mapped_workload.dbms)
dummy_encoder = DummyEncoder(categorical_info['n_values'],
categorical_info['categorical_features'],
categorical_info['cat_columnlabels'],
categorical_info['noncat_columnlabels'])
X_matrix = dummy_encoder.fit_transform(X_matrix)
# below two variables are needed for correctly determing max/min on dummies
binary_index_set = set(categorical_info['binary_vars'])
total_dummies = dummy_encoder.total_dummies()
# Scale to N(0, 1)
X_scaler = StandardScaler()
X_scaled = X_scaler.fit_transform(X_matrix)
if y_target.shape[0] < 5: # FIXME
# FIXME (dva): if there are fewer than 5 target results so far
# then scale the y values (metrics) using the workload's
# y_scaler. I'm not sure if 5 is the right cutoff.
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_matrix = np.vstack([y_target, y_workload])
y_scaled = y_workload_scaler.fit_transform(y_matrix)
else:
# FIXME (dva): otherwise try to compute a separate y_scaler for
# the target and scale them separately.
try:
y_target_scaler = StandardScaler()
y_workload_scaler = StandardScaler()
y_target_scaled = y_target_scaler.fit_transform(y_target)
y_workload_scaled = y_workload_scaler.fit_transform(y_workload)
y_scaled = np.vstack([y_target_scaled, y_workload_scaled])
except ValueError:
y_target_scaler = None
y_workload_scaler = StandardScaler()
y_scaled = y_workload_scaler.fit_transform(y_target)
# Set up constraint helper
constraint_helper = ParamConstraintHelper(scaler=X_scaler,
encoder=dummy_encoder,
binary_vars=categorical_info['binary_vars'],
init_flip_prob=INIT_FLIP_PROB,
flip_prob_decay=FLIP_PROB_DECAY)
# FIXME (dva): check if these are good values for the ridge
# ridge = np.empty(X_scaled.shape[0])
# ridge[:X_target.shape[0]] = 0.01
# ridge[X_target.shape[0]:] = 0.1
# FIXME: we should generate more samples and use a smarter sampling
# technique
num_samples = NUM_SAMPLES
X_samples = np.empty((num_samples, X_scaled.shape[1]))
X_min = np.empty(X_scaled.shape[1])
X_max = np.empty(X_scaled.shape[1])
knobs_mem = KnobCatalog.objects.filter(
dbms=newest_result.session.dbms, tunable=True, resource=1)
knobs_mem_catalog = {k.name: k for k in knobs_mem}
mem_max = newest_result.workload.hardware.memory
X_mem = np.zeros([1, X_scaled.shape[1]])
X_default = np.empty(X_scaled.shape[1])
# Get default knob values
for i, k_name in enumerate(X_columnlabels):
k = KnobCatalog.objects.filter(dbms=newest_result.session.dbms, name=k_name)[0]
X_default[i] = k.default
X_default_scaled = X_scaler.transform(X_default.reshape(1, X_default.shape[0]))[0]
# Determine min/max for knob values
for i in range(X_scaled.shape[1]):
if i < total_dummies or i in binary_index_set:
col_min = 0
col_max = 1
else:
col_min = X_scaled[:, i].min()
col_max = X_scaled[:, i].max()
if X_columnlabels[i] in knobs_mem_catalog:
X_mem[0][i] = mem_max * 1024 * 1024 * 1024 # mem_max GB
col_max = X_scaler.transform(X_mem)[0][i]
# Set min value to the default value
# FIXME: support multiple methods can be selected by users
col_min = X_default_scaled[i]
X_min[i] = col_min
X_max[i] = col_max
X_samples[:, i] = np.random.rand(num_samples) * (col_max - col_min) + col_min
# Maximize the throughput, moreisbetter
# Use gradient descent to minimize -throughput
if not lessisbetter:
y_scaled = -y_scaled
q = queue.PriorityQueue()
for x in range(0, y_scaled.shape[0]):
q.put((y_scaled[x][0], x))
i = 0
while i < TOP_NUM_CONFIG:
try:
item = q.get_nowait()
# Tensorflow get broken if we use the training data points as
# starting points for GPRGD. We add a small bias for the
# starting points. GPR_EPS default value is 0.001
X_samples = np.vstack((X_samples, X_scaled[item[1]] + GPR_EPS))
i = i + 1
except queue.Empty:
break
model = GPRGD(length_scale=DEFAULT_LENGTH_SCALE,
magnitude=DEFAULT_MAGNITUDE,
max_train_size=MAX_TRAIN_SIZE,
batch_size=BATCH_SIZE,
num_threads=NUM_THREADS,
learning_rate=DEFAULT_LEARNING_RATE,
epsilon=DEFAULT_EPSILON,
max_iter=MAX_ITER,
sigma_multiplier=DEFAULT_SIGMA_MULTIPLIER,
mu_multiplier=DEFAULT_MU_MULTIPLIER)
model.fit(X_scaled, y_scaled, X_min, X_max, ridge=DEFAULT_RIDGE)
res = model.predict(X_samples, constraint_helper=constraint_helper)
best_config_idx = np.argmin(res.minl.ravel())
best_config = res.minl_conf[best_config_idx, :]
best_config = X_scaler.inverse_transform(best_config)
# Decode one-hot encoding into categorical knobs
best_config = dummy_encoder.inverse_transform(best_config)
# Although we have max/min limits in the GPRGD training session, it may
# lose some precisions. e.g. 0.99..99 >= 1.0 may be True on the scaled data,
# when we inversely transform the scaled data, the different becomes much larger
# and cannot be ignored. Here we check the range on the original data
# directly, and make sure the recommended config lies within the range
X_min_inv = X_scaler.inverse_transform(X_min)
X_max_inv = X_scaler.inverse_transform(X_max)
best_config = np.minimum(best_config, X_max_inv)
best_config = np.maximum(best_config, X_min_inv)
conf_map = {k: best_config[i] for i, k in enumerate(X_columnlabels)}
conf_map_res = {}
conf_map_res['status'] = 'good'
conf_map_res['recommendation'] = conf_map
conf_map_res['info'] = 'INFO: training data size is {}'.format(X_scaled.shape[0])
return conf_map_res
