Simplify failure rate by building separate query

This commit is contained in:
Pea Tyczynska
2019-12-03 16:18:07 +00:00
parent 53efd87e28
commit cfbb080f57
9 changed files with 77 additions and 37 deletions

View File

@@ -266,11 +266,7 @@ def check_for_services_with_high_failure_rates_or_sending_to_tv_numbers():
message = "" message = ""
services_with_failures = get_services_with_high_failure_rates(start_date=start_date, end_date=end_date) services_with_failures = get_services_with_high_failure_rates(start_date=start_date, end_date=end_date)
services_sending_to_tv_numbers = dao_find_services_sending_to_tv_numbers( services_sending_to_tv_numbers = dao_find_services_sending_to_tv_numbers(start_date=start_date, end_date=end_date)
threshold=100,
start_date=start_date,
end_date=end_date
)
if services_with_failures: if services_with_failures:
message += "{} service(s) have had high permanent-failure rates for sms messages in last 24 hours:\n".format( message += "{} service(s) have had high permanent-failure rates for sms messages in last 24 hours:\n".format(
@@ -291,6 +287,7 @@ def check_for_services_with_high_failure_rates_or_sending_to_tv_numbers():
current_app.logger.exception(message) current_app.logger.exception(message)
if current_app.config['NOTIFY_ENVIRONMENT'] in ['live', 'production', 'test']: if current_app.config['NOTIFY_ENVIRONMENT'] in ['live', 'production', 'test']:
message += "\nThings to do: contact service? revoke their key?"
zendesk_client.create_ticket( zendesk_client.create_ticket(
subject="[{}] High failure rates for sms spotted for services".format( subject="[{}] High failure rates for sms spotted for services".format(
current_app.config['NOTIFY_ENVIRONMENT'] current_app.config['NOTIFY_ENVIRONMENT']

View File

@@ -524,9 +524,7 @@ def dao_fetch_active_users_for_service(service_id):
def dao_find_services_sending_to_tv_numbers(start_date, end_date, threshold=100): def dao_find_services_sending_to_tv_numbers(start_date, end_date, threshold=100):
return db.session.query( return db.session.query(
Service.name.label('service_name'),
Notification.service_id.label('service_id'), Notification.service_id.label('service_id'),
func.count(Notification.id).label('notification_count') func.count(Notification.id).label('notification_count')
).filter( ).filter(
@@ -541,7 +539,27 @@ def dao_find_services_sending_to_tv_numbers(start_date, end_date, threshold=100)
Service.active == True, Service.active == True,
).group_by( ).group_by(
Notification.service_id, Notification.service_id,
Service.name
).having( ).having(
func.count(Notification.id) > threshold func.count(Notification.id) > threshold
).all() ).all()
def dao_find_real_sms_notification_count_by_status_for_live_services(start_date, end_date):
# only works within services' retention period
return db.session.query(
Notification.service_id.label('service_id'),
Notification.status.label('status'),
func.count(Notification.id).label('count')
).filter(
Notification.service_id == Service.id,
Notification.created_at >= start_date,
Notification.created_at <= end_date,
Notification.key_type != KEY_TYPE_TEST,
Notification.notification_type == SMS_TYPE,
Service.restricted == False, # noqa
Service.research_mode == False,
Service.active == True,
).group_by(
Notification.service_id,
Notification.status
).all()

View File

@@ -24,8 +24,7 @@ def format_statistics(statistics):
def get_rate_of_permanent_failures_for_service(statistics, threshold=100): def get_rate_of_permanent_failures_for_service(statistics, threshold=100):
counts = {"permanent_failure": 0, "all_other_statuses": 0} counts = {"permanent_failure": 0, "all_other_statuses": 0}
for row in statistics: for row in statistics:
if row.notification_type == 'sms': _count_if_status_is_permanent_failure_from_row(counts, row)
_count_if_status_is_permanent_failure_from_row(counts, row)
if counts['permanent_failure'] + counts['all_other_statuses'] >= threshold: if counts['permanent_failure'] + counts['all_other_statuses'] >= threshold:
rate = counts['permanent_failure'] / (counts['permanent_failure'] + counts['all_other_statuses']) rate = counts['permanent_failure'] / (counts['permanent_failure'] + counts['all_other_statuses'])

View File

@@ -2,6 +2,8 @@ import itertools
from notifications_utils.recipients import allowed_to_send_to from notifications_utils.recipients import allowed_to_send_to
from app.dao.services_dao import dao_find_real_sms_notification_count_by_status_for_live_services
from app.models import ( from app.models import (
ServiceWhitelist, ServiceWhitelist,
MOBILE_TYPE, EMAIL_TYPE, MOBILE_TYPE, EMAIL_TYPE,
@@ -9,8 +11,6 @@ from app.models import (
from app.service import statistics from app.service import statistics
from app.dao.fact_notification_status_dao import fetch_stats_for_all_services_by_date_range
def get_recipients_from_request(request_json, key, type): def get_recipients_from_request(request_json, key, type):
return [(type, recipient) for recipient in request_json.get(key)] return [(type, recipient) for recipient in request_json.get(key)]
@@ -59,20 +59,18 @@ def service_allowed_to_send_to(recipient, service, key_type, allow_whitelisted_r
def get_services_with_high_failure_rates(start_date, end_date, rate=0.25, threshold=100): def get_services_with_high_failure_rates(start_date, end_date, rate=0.25, threshold=100):
stats = fetch_stats_for_all_services_by_date_range( stats = dao_find_real_sms_notification_count_by_status_for_live_services(
start_date=start_date.date(), start_date=start_date,
end_date=end_date.date(), end_date=end_date,
include_from_test_key=False,
) )
results = [] results = []
for service_id, rows in itertools.groupby(stats, lambda x: x.service_id): for service_id, rows in itertools.groupby(stats, lambda x: x.service_id):
rows = list(rows) rows = list(rows)
if not rows[0].restricted and not rows[0].research_mode and rows[0].active:
permanent_failure_rate = statistics.get_rate_of_permanent_failures_for_service(rows, threshold=threshold) permanent_failure_rate = statistics.get_rate_of_permanent_failures_for_service(rows, threshold=threshold)
if permanent_failure_rate >= rate: if permanent_failure_rate >= rate:
results.append({ results.append({
'id': str(rows[0].service_id), 'id': str(rows[0].service_id),
'name': rows[0].name, 'permanent_failure_rate': permanent_failure_rate
'permanent_failure_rate': permanent_failure_rate })
})
return results return results

View File

@@ -510,7 +510,7 @@ MockServicesSendingToTVNumbers = namedtuple(
@pytest.mark.parametrize("failure_rates, sms_to_tv_numbers, expected_message", [ @pytest.mark.parametrize("failure_rates, sms_to_tv_numbers, expected_message", [
[ [
[{"id": "123", "name": "Service 1", "permanent_failure_rate": 0.3}], [{"id": "123", "permanent_failure_rate": 0.3}],
[], [],
"1 service(s) have had high permanent-failure rates for sms messages in last " "1 service(s) have had high permanent-failure rates for sms messages in last "
"24 hours:\nservice id: 123 failure rate: 0.3,\n" "24 hours:\nservice id: 123 failure rate: 0.3,\n"
@@ -534,13 +534,15 @@ def test_check_for_services_with_high_failure_rates_or_sending_to_tv_numbers(
'app.celery.scheduled_tasks.dao_find_services_sending_to_tv_numbers', return_value=sms_to_tv_numbers 'app.celery.scheduled_tasks.dao_find_services_sending_to_tv_numbers', return_value=sms_to_tv_numbers
) )
zendesk_actions = "\nThings to do: contact service? revoke their key?"
check_for_services_with_high_failure_rates_or_sending_to_tv_numbers() check_for_services_with_high_failure_rates_or_sending_to_tv_numbers()
assert mock_failure_rates.called assert mock_failure_rates.called
assert mock_sms_to_tv_numbers.called assert mock_sms_to_tv_numbers.called
mock_logger.assert_called_once_with(expected_message) mock_logger.assert_called_once_with(expected_message)
mock_create_ticket.assert_called_with( mock_create_ticket.assert_called_with(
message=expected_message, message=expected_message + zendesk_actions,
subject="[test] High failure rates for sms spotted for services", subject="[test] High failure rates for sms spotted for services",
ticket_type='incident' ticket_type='incident'
) )

View File

@@ -14,7 +14,7 @@ from app.dao.fact_notification_status_dao import (
fetch_notification_status_totals_for_all_services, fetch_notification_status_totals_for_all_services,
fetch_notification_statuses_for_job, fetch_notification_statuses_for_job,
fetch_stats_for_all_services_by_date_range, fetch_monthly_template_usage_for_service, fetch_stats_for_all_services_by_date_range, fetch_monthly_template_usage_for_service,
get_total_sent_notifications_for_day_and_type, get_total_sent_notifications_for_day_and_type
) )
from app.models import ( from app.models import (
FactNotificationStatus, FactNotificationStatus,

View File

@@ -23,6 +23,7 @@ from app.dao.services_dao import (
dao_fetch_live_services_data, dao_fetch_live_services_data,
dao_fetch_service_by_id, dao_fetch_service_by_id,
dao_fetch_all_services_by_user, dao_fetch_all_services_by_user,
dao_find_real_sms_notification_count_by_status_for_live_services,
dao_find_services_sending_to_tv_numbers, dao_find_services_sending_to_tv_numbers,
dao_update_service, dao_update_service,
delete_service_and_all_associated_db_objects, delete_service_and_all_associated_db_objects,
@@ -1103,8 +1104,8 @@ def create_email_sms_letter_template():
@freeze_time("2019-12-02 12:00:00.000000") @freeze_time("2019-12-02 12:00:00.000000")
def test_dao_find_services_sending_to_tv_numbers(notify_db_session): def test_dao_find_services_sending_to_tv_numbers(notify_db_session, fake_uuid):
service_1 = create_service(service_name="Service 1") service_1 = create_service(service_name="Service 1", service_id=fake_uuid)
service_3 = create_service(service_name="Service 3", restricted=True) # restricted service_3 = create_service(service_name="Service 3", restricted=True) # restricted
service_4 = create_service(service_name="Service 4", research_mode=True) # research mode service_4 = create_service(service_name="Service 4", research_mode=True) # research mode
service_5 = create_service(service_name="Service 5", active=False) # not active service_5 = create_service(service_name="Service 5", active=False) # not active
@@ -1137,4 +1138,36 @@ def test_dao_find_services_sending_to_tv_numbers(notify_db_session):
result = dao_find_services_sending_to_tv_numbers(start_date, end_date, threshold=4) result = dao_find_services_sending_to_tv_numbers(start_date, end_date, threshold=4)
assert len(result) == 1 assert len(result) == 1
assert result[0].service_name == "Service 1" assert str(result[0].service_id) == fake_uuid
def test_dao_find_real_sms_notification_count_by_status_for_live_services(notify_db_session, fake_uuid):
service_1 = create_service(service_name="Service 1", service_id=fake_uuid)
service_3 = create_service(service_name="Service 3", restricted=True) # restricted
service_4 = create_service(service_name="Service 4", research_mode=True) # research mode
service_5 = create_service(service_name="Service 5", active=False) # not active
services = [service_1, service_3, service_4, service_5]
for service in services:
template = create_template(service)
for x in range(0, 3):
create_notification(template, status="permanent-failure")
create_notification(template, status="delivered")
create_notification(template, status="temporary-failure")
service_6 = create_service(service_name="Service 6") # notifications too old
with freeze_time("2019-11-30 15:00:00.000000"):
template_6 = create_template(service_6)
for x in range(0, 2):
create_notification(template_6, status="permanent-failure")
service_2 = create_service(service_name="Service 2") # below threshold
template_2 = create_template(service_2)
create_notification(template_2, status="permanent-failure", key_type='test') # test key type
start_date = (datetime.utcnow() - timedelta(days=1))
end_date = datetime.utcnow()
result = dao_find_real_sms_notification_count_by_status_for_live_services(start_date, end_date)
assert len(result) == 3
assert str(result[0].service_id) == fake_uuid

View File

@@ -79,12 +79,6 @@ def test_format_statistics(stats, email_counts, sms_counts, letter_counts):
StatsRow('sms', 'permanent-failure', 100), StatsRow('sms', 'permanent-failure', 100),
StatsRow('sms', 'delivered', 300), StatsRow('sms', 'delivered', 300),
], 0.25), ], 0.25),
'ignores_other_notification_types': ([
StatsRow('sms', 'permanent-failure', 100),
StatsRow('sms', 'delivered', 300),
StatsRow('email', 'delivered', 300),
StatsRow('letter', 'created', 300),
], 0.25),
'only_counts_permanent_failure_as_failed': ([ 'only_counts_permanent_failure_as_failed': ([
StatsRow('sms', 'permanent-failure', 100), StatsRow('sms', 'permanent-failure', 100),
StatsRow('sms', 'temporary-failure', 100), StatsRow('sms', 'temporary-failure', 100),

View File

@@ -45,6 +45,5 @@ def test_get_services_with_high_failure_rates(notify_db_session):
assert get_services_with_high_failure_rates(start_date, end_date, threshold=3) == [{ assert get_services_with_high_failure_rates(start_date, end_date, threshold=3) == [{
'id': str(service_1.id), 'id': str(service_1.id),
'name': service_1.name,
'permanent_failure_rate': 0.25 'permanent_failure_rate': 0.25
}] }]