blob: 3bc565d92b56a9a0aac78054fc11714c4a99881c [file] [log] [blame]
import os, pickle, datetime, itertools, operator
from django.db import models as dbmodels
from autotest_lib.frontend import thread_local
from autotest_lib.frontend.afe import rpc_utils, model_logic
from autotest_lib.frontend.afe import readonly_connection
from autotest_lib.new_tko.tko import models, tko_rpc_utils, graphing_utils
from autotest_lib.new_tko.tko import preconfigs
# table/spreadsheet view support
def get_test_views(**filter_data):
return rpc_utils.prepare_for_serialization(
models.TestView.list_objects(filter_data))
def get_num_test_views(**filter_data):
return models.TestView.query_count(filter_data)
def get_group_counts(group_by, header_groups=[], fixed_headers={},
machine_label_headers={}, extra_select_fields={},
**filter_data):
"""
Queries against TestView grouping by the specified fields and computings
counts for each group.
* group_by should be a list of field names.
* extra_select_fields can be used to specify additional fields to select
(usually for aggregate functions).
* header_groups can be used to get lists of unique combinations of group
fields. It should be a list of tuples of fields from group_by. It's
primarily for use by the spreadsheet view.
* fixed_headers can map header fields to lists of values. the header will
guaranteed to return exactly those value. this does not work together
with header_groups.
* machine_label_headers can specify special headers to be constructed from
machine labels. It should map arbitrary names to lists of machine labels.
a field will be created with the given name containing a comma-separated
list indicating which of the given machine labels are on each test. this
field can then be grouped on.
Returns a dictionary with two keys:
* header_values contains a list of lists, one for each header group in
header_groups. Each list contains all the values for the corresponding
header group as tuples.
* groups contains a list of dicts, one for each row. Each dict contains
keys for each of the group_by fields, plus a 'group_count' key for the
total count in the group, plus keys for each of the extra_select_fields.
The keys for the extra_select_fields are determined by the "AS" alias of
the field.
"""
extra_select_fields = dict(extra_select_fields)
query = models.TestView.objects.get_query_set_with_joins(
filter_data, include_host_labels=bool(machine_label_headers))
query = models.TestView.query_objects(filter_data, initial_query=query)
count_alias, count_sql = models.TestView.objects.get_count_sql(query)
extra_select_fields[count_alias] = count_sql
if 'test_idx' not in group_by:
extra_select_fields['test_idx'] = 'test_idx'
tko_rpc_utils.add_machine_label_headers(machine_label_headers,
extra_select_fields)
group_processor = tko_rpc_utils.GroupDataProcessor(query, group_by,
header_groups,
fixed_headers,
extra_select_fields)
group_processor.process_group_dicts()
return rpc_utils.prepare_for_serialization(group_processor.get_info_dict())
def get_num_groups(group_by, **filter_data):
"""
Gets the count of unique groups with the given grouping fields.
"""
query = models.TestView.query_objects(filter_data)
return models.TestView.objects.get_num_groups(query, group_by)
def get_status_counts(group_by, header_groups=[], fixed_headers={},
machine_label_headers={}, **filter_data):
"""
Like get_group_counts, but also computes counts of passed, complete (and
valid), and incomplete tests, stored in keys "pass_count', 'complete_count',
and 'incomplete_count', respectively.
"""
return get_group_counts(group_by, header_groups=header_groups,
fixed_headers=fixed_headers,
machine_label_headers=machine_label_headers,
extra_select_fields=tko_rpc_utils.STATUS_FIELDS,
**filter_data)
def get_latest_tests(group_by, header_groups=[], fixed_headers={},
machine_label_headers={}, extra_info=[], **filter_data):
"""
Similar to get_status_counts, but return only the latest test result per
group. It still returns the same information (i.e. with pass count etc.)
for compatibility.
@param extra_info a list containing the field names that should be returned
with each cell. The fields are returned in the extra_info
field of the return dictionary.
"""
# find latest test per group
query = models.TestView.objects.get_query_set_with_joins(
filter_data, include_host_labels=bool(machine_label_headers))
query = models.TestView.query_objects(filter_data, initial_query=query)
query.exclude(status__in=tko_rpc_utils._INVALID_STATUSES)
extra_fields = {'latest_test_idx' : 'MAX(%s)' %
models.TestView.objects.get_key_on_this_table('test_idx')}
tko_rpc_utils.add_machine_label_headers(machine_label_headers,
extra_fields)
group_processor = tko_rpc_utils.GroupDataProcessor(query, group_by,
header_groups,
fixed_headers,
extra_fields)
group_processor.process_group_dicts()
info = group_processor.get_info_dict()
# fetch full info for these tests so we can access their statuses
all_test_ids = [group['latest_test_idx'] for group in info['groups']]
test_views = models.TestView.objects.in_bulk(all_test_ids)
for group_dict in info['groups']:
test_idx = group_dict.pop('latest_test_idx')
group_dict['test_idx'] = test_idx
test_view = test_views[test_idx]
tko_rpc_utils.add_status_counts(group_dict, test_view.status)
group_dict['extra_info'] = []
for field in extra_info:
group_dict['extra_info'].append(getattr(test_view, field))
return rpc_utils.prepare_for_serialization(info)
def get_job_ids(**filter_data):
"""
Returns AFE job IDs for all tests matching the filters.
"""
query = models.TestView.query_objects(filter_data)
job_ids = set()
for test_view in query.values('job_tag').distinct():
# extract job ID from tag
first_tag_component = test_view['job_tag'].split('-')[0]
try:
job_id = int(first_tag_component)
job_ids.add(job_id)
except ValueError:
# a nonstandard job tag, i.e. from contributed results
pass
return list(job_ids)
# iteration support
def get_iteration_views(result_keys, **test_filter_data):
"""
Similar to get_test_views, but returns a dict for each iteration rather
than for each test. Accepts the same filter data as get_test_views.
@param result_keys: list of iteration result keys to include. Only
iterations contains all these keys will be included.
@returns a list of dicts, one for each iteration. Each dict contains:
* all the same information as get_test_views()
* all the keys specified in result_keys
* an additional key 'iteration_index'
"""
iteration_views = tko_rpc_utils.get_iteration_view_query(result_keys,
test_filter_data)
final_filter_data = tko_rpc_utils.extract_presentation_params(
test_filter_data)
final_filter_data['no_distinct'] = True
fields = (models.TestView.get_field_dict().keys() + result_keys +
['iteration_index'])
iteration_dicts = models.TestView.list_objects(
final_filter_data, initial_query=iteration_views, fields=fields)
return rpc_utils.prepare_for_serialization(iteration_dicts)
def get_num_iteration_views(result_keys, **test_filter_data):
iteration_views = tko_rpc_utils.get_iteration_view_query(result_keys,
test_filter_data)
return iteration_views.count()
# test detail view
def _attributes_to_dict(attribute_list):
return dict((attribute.attribute, attribute.value)
for attribute in attribute_list)
def _iteration_attributes_to_dict(attribute_list):
iter_keyfunc = operator.attrgetter('iteration')
attribute_list.sort(key=iter_keyfunc)
iterations = {}
for key, group in itertools.groupby(attribute_list, iter_keyfunc):
iterations[key] = _attributes_to_dict(group)
return iterations
def _format_iteration_keyvals(test):
iteration_attr = _iteration_attributes_to_dict(test.iteration_attributes)
iteration_perf = _iteration_attributes_to_dict(test.iteration_results)
all_iterations = iteration_attr.keys() + iteration_perf.keys()
max_iterations = max(all_iterations + [0])
# merge the iterations into a single list of attr & perf dicts
return [{'attr': iteration_attr.get(index, {}),
'perf': iteration_perf.get(index, {})}
for index in xrange(1, max_iterations + 1)]
def get_detailed_test_views(**filter_data):
test_views = models.TestView.list_objects(filter_data)
tests_by_id = models.Test.objects.in_bulk([test_view['test_idx']
for test_view in test_views])
tests = tests_by_id.values()
models.Test.objects.populate_relationships(tests, models.TestAttribute,
'attributes')
models.Test.objects.populate_relationships(tests, models.IterationAttribute,
'iteration_attributes')
models.Test.objects.populate_relationships(tests, models.IterationResult,
'iteration_results')
models.Test.objects.populate_relationships(tests, models.TestLabel,
'labels')
for test_view in test_views:
test = tests_by_id[test_view['test_idx']]
test_view['attributes'] = _attributes_to_dict(test.attributes)
test_view['iterations'] = _format_iteration_keyvals(test)
test_view['labels'] = [label.name for label in test.labels]
return rpc_utils.prepare_for_serialization(test_views)
# graphing view support
def get_hosts_and_tests():
"""\
Gets every host that has had a benchmark run on it. Additionally, also
gets a dictionary mapping the host names to the benchmarks.
"""
host_info = {}
q = (dbmodels.Q(test_name__startswith='kernbench') |
dbmodels.Q(test_name__startswith='dbench') |
dbmodels.Q(test_name__startswith='tbench') |
dbmodels.Q(test_name__startswith='unixbench') |
dbmodels.Q(test_name__startswith='iozone'))
test_query = models.TestView.objects.filter(q).values(
'test_name', 'hostname', 'machine_idx').distinct()
for result_dict in test_query:
hostname = result_dict['hostname']
test = result_dict['test_name']
machine_idx = result_dict['machine_idx']
host_info.setdefault(hostname, {})
host_info[hostname].setdefault('tests', [])
host_info[hostname]['tests'].append(test)
host_info[hostname]['id'] = machine_idx
return rpc_utils.prepare_for_serialization(host_info)
def create_metrics_plot(queries, plot, invert, drilldown_callback,
normalize=None):
return graphing_utils.create_metrics_plot(
queries, plot, invert, normalize, drilldown_callback=drilldown_callback)
def create_qual_histogram(query, filter_string, interval, drilldown_callback):
return graphing_utils.create_qual_histogram(
query, filter_string, interval, drilldown_callback=drilldown_callback)
# TODO(showard) - this extremely generic RPC is used only by one place in the
# client. We should come up with a more opaque RPC for that place to call and
# get rid of this.
def execute_query_with_param(query, param):
cursor = readonly_connection.connection().cursor()
cursor.execute(query, param)
return cursor.fetchall()
def get_preconfig(name, type):
return preconfigs.manager.get_preconfig(name, type)
def get_embedding_id(url_token, graph_type, params):
try:
model = models.EmbeddedGraphingQuery.objects.get(url_token=url_token)
except models.EmbeddedGraphingQuery.DoesNotExist:
params_str = pickle.dumps(params)
now = datetime.datetime.now()
model = models.EmbeddedGraphingQuery(url_token=url_token,
graph_type=graph_type,
params=params_str,
last_updated=now)
model.cached_png = graphing_utils.create_embedded_plot(model,
now.ctime())
model.save()
return model.id
def get_embedded_query_url_token(id):
model = models.EmbeddedGraphingQuery.objects.get(id=id)
return model.url_token
# test label management
def add_test_label(name, description=None):
return models.TestLabel.add_object(name=name, description=description).id
def modify_test_label(label_id, **data):
models.TestLabel.smart_get(label_id).update_object(data)
def delete_test_label(label_id):
models.TestLabel.smart_get(label_id).delete()
def get_test_labels(**filter_data):
return rpc_utils.prepare_for_serialization(
models.TestLabel.list_objects(filter_data))
def get_test_labels_for_tests(**test_filter_data):
label_ids = models.TestView.objects.query_test_label_ids(test_filter_data)
labels = models.TestLabel.list_objects({'id__in' : label_ids})
return rpc_utils.prepare_for_serialization(labels)
def test_label_add_tests(label_id, **test_filter_data):
test_ids = models.TestView.objects.query_test_ids(test_filter_data)
models.TestLabel.smart_get(label_id).tests.add(*test_ids)
def test_label_remove_tests(label_id, **test_filter_data):
label = models.TestLabel.smart_get(label_id)
# only include tests that actually have this label
extra_where = test_filter_data.get('extra_where', '')
if extra_where:
extra_where = '(' + extra_where + ') AND '
extra_where += 'test_labels.id = %s' % label.id
test_filter_data['extra_where'] = extra_where
test_ids = models.TestView.objects.query_test_ids(test_filter_data)
label.tests.remove(*test_ids)
# user-created test attributes
def set_test_attribute(attribute, value, **test_filter_data):
"""
* attribute - string name of attribute
* value - string, or None to delete an attribute
* test_filter_data - filter data to apply to TestView to choose tests to act
upon
"""
assert test_filter_data # disallow accidental actions on all hosts
test_ids = models.TestView.objects.query_test_ids(test_filter_data)
tests = models.Test.objects.in_bulk(test_ids)
for test in tests.itervalues():
test.set_or_delete_attribute(attribute, value)
# saved queries
def get_saved_queries(**filter_data):
return rpc_utils.prepare_for_serialization(
models.SavedQuery.list_objects(filter_data))
def add_saved_query(name, url_token):
name = name.strip()
owner = thread_local.get_user()
existing_list = list(models.SavedQuery.objects.filter(owner=owner,
name=name))
if existing_list:
query_object = existing_list[0]
query_object.url_token = url_token
query_object.save()
return query_object.id
return models.SavedQuery.add_object(owner=owner, name=name,
url_token=url_token).id
def delete_saved_queries(id_list):
user = thread_local.get_user()
query = models.SavedQuery.objects.filter(id__in=id_list, owner=user)
if query.count() == 0:
raise model_logic.ValidationError('No such queries found for this user')
query.delete()
# other
def get_motd():
return rpc_utils.get_motd()
def get_static_data():
result = {}
group_fields = []
for field in models.TestView.group_fields:
if field in models.TestView.extra_fields:
name = models.TestView.extra_fields[field]
else:
name = models.TestView.get_field_dict()[field].verbose_name
group_fields.append((name.capitalize(), field))
model_fields = [(field.verbose_name.capitalize(), field.column)
for field in models.TestView._meta.fields]
extra_fields = [(field_name.capitalize(), field_sql)
for field_sql, field_name
in models.TestView.extra_fields.iteritems()]
benchmark_key = {
'kernbench' : 'elapsed',
'dbench' : 'throughput',
'tbench' : 'throughput',
'unixbench' : 'score',
'iozone' : '32768-4096-fwrite'
}
perf_view = [
['Test Index', 'test_idx'],
['Job Index', 'job_idx'],
['Test Name', 'test_name'],
['Subdirectory', 'subdir'],
['Kernel Index', 'kernel_idx'],
['Status Index', 'status_idx'],
['Reason', 'reason'],
['Host Index', 'machine_idx'],
['Test Started Time', 'test_started_time'],
['Test Finished Time', 'test_finished_time'],
['Job Tag', 'job_tag'],
['Job Name', 'job_name'],
['Owner', 'job_owner'],
['Job Queued Time', 'job_queued_time'],
['Job Started Time', 'job_started_time'],
['Job Finished Time', 'job_finished_time'],
['Hostname', 'hostname'],
['Platform', 'platform'],
['Machine Owner', 'machine_owner'],
['Kernel Hash', 'kernel_hash'],
['Kernel Base', 'kernel_base'],
['Kernel', 'kernel'],
['Status', 'status'],
['Iteration Number', 'iteration'],
['Performance Keyval (Key)', 'iteration_key'],
['Performance Keyval (Value)', 'iteration_value'],
]
result['group_fields'] = sorted(group_fields)
result['all_fields'] = sorted(model_fields + extra_fields)
result['test_labels'] = get_test_labels(sort_by=['name'])
result['current_user'] = {'login' : thread_local.get_user()}
result['benchmark_key'] = benchmark_key
result['perf_view'] = perf_view
result['test_view'] = model_fields
result['preconfigs'] = preconfigs.manager.all_preconfigs()
result['motd'] = rpc_utils.get_motd()
return result