openreplay/ee/api/chalicelib/core/sessions_insights.py
2023-01-09 17:20:13 +01:00

353 lines
17 KiB
Python

import schemas_ee
from chalicelib.core import metrics
from chalicelib.utils import ch_client
def _table_slice(table, index):
col = list()
for row in table:
col.append(row[index])
return col
def _table_where(table, index, value):
new_table = list()
for row in table:
if row[index] == value:
new_table.append(row)
return new_table
def _sum_table_index(table, index):
s = 0
count = 0
for row in table:
v = row[index]
if v is None:
continue
s += v
count += 1
return s
def _mean_table_index(table, index):
s = _sum_table_index(table, index)
c = len(table)
return s / c
def _sort_table_index(table, index, reverse=False):
return sorted(table, key=lambda k: k[index], reverse=reverse)
def _select_rec(l, selector):
print('selector:', selector)
print('list:', l)
if len(selector) == 1:
return l[selector[0]]
else:
s = selector[0]
L = l[s]
type_ = type(s)
if type_ == slice:
return [_select_rec(l_, selector[1:]) for l_ in L]
elif type_ == int:
return [_select_rec(L, selector[1:])]
# TODO Deal with None values
def __get_two_values(response, time_index='hh', name_index='name'):
columns = list(response[0].keys())
name_index_val = columns.index(name_index)
time_index_value = columns.index(time_index)
table = [list(r.values()) for r in response]
table_hh1 = list()
table_hh2 = list()
hh_vals = list()
names_hh1 = list()
names_hh2 = list()
for e in table:
if e[time_index_value] not in hh_vals and len(hh_vals) == 2:
break
elif e[time_index_value] not in hh_vals:
hh_vals.append(e[time_index_value])
if len(hh_vals) == 1:
table_hh1.append(e)
if e[name_index_val] not in names_hh1:
names_hh1.append(e[name_index_val])
elif len(hh_vals) == 2:
table_hh2.append(e)
if e[name_index_val] not in names_hh2:
names_hh2.append(e[name_index_val])
return table_hh1, table_hh2, columns, names_hh1, names_hh2
def __handle_timestep(time_step):
base = "{0}"
if time_step == 'hour':
return f"toStartOfHour({base})", 3600
elif time_step == 'day':
return f"toStartOfDay({base})", 24 * 3600
elif time_step == 'week':
return f"toStartOfWeek({base})", 7 * 24 * 3600
else:
assert type(
time_step) == int, "time_step must be {'hour', 'day', 'week'} or an integer representing the time step in minutes"
return f"toStartOfInterval({base}, INTERVAL {time_step} minute)", int(time_step) * 60
def query_requests_by_period(project_id, start_time, end_time, conn=None):
params = {
"project_id": project_id, "startTimestamp": start_time, "endTimestamp": end_time,
"step_size": metrics.__get_step_size(endTimestamp=end_time, startTimestamp=start_time, density=3)
}
conditions = ["event_type = 'REQUEST'"]
query = f"""WITH toUInt32(toStartOfInterval(toDateTime(%(startTimestamp)s/1000), INTERVAL %(step_size)s second)) AS start,
toUInt32(toStartOfInterval(toDateTime(%(endTimestamp)s/1000), INTERVAL %(step_size)s second)) AS end
SELECT T1.hh, count(T2.session_id) as sessions, avg(T2.success) as success_rate, T2.url_host as names,
T2.url_path as source, avg(T2.duration) as avg_duration
FROM (SELECT arrayJoin(arrayMap(x -> toDateTime(x), range(start, end, %(step_size)s))) as hh) AS T1
LEFT JOIN (SELECT session_id, url_host, url_path, success, message, duration, toStartOfInterval(datetime, INTERVAL %(step_size)s second) as dtime
FROM experimental.events
WHERE project_id = {project_id}
AND {" AND ".join(conditions)}) AS T2 ON T2.dtime = T1.hh
GROUP BY T1.hh, T2.url_host, T2.url_path
ORDER BY T1.hh DESC;"""
if conn is None:
with ch_client.ClickHouseClient() as conn:
query = conn.format(query=query, params=params)
res = conn.execute(query=query)
else:
query = conn.format(query=query, params=params)
res = conn.execute(query=query)
table_hh1, table_hh2, columns, this_period_hosts, last_period_hosts = __get_two_values(res, time_index='hh',
name_index='source')
del res
new_hosts = [x for x in this_period_hosts if x not in last_period_hosts]
common_names = [x for x in this_period_hosts if x not in new_hosts]
source_idx = columns.index('source')
duration_idx = columns.index('avg_duration')
success_idx = columns.index('success_rate')
delta_duration = dict()
delta_success = dict()
for n in common_names:
d1_tmp = _table_where(table_hh1, source_idx, n)
# d1_tmp = table_hh1[table_hh1[:, source_idx] == n]
d2_tmp = _table_where(table_hh2, source_idx, n)
# d2_tmp = table_hh2[table_hh2[:, source_idx] == n]
delta_duration[n] = _mean_table_index(d1_tmp, duration_idx) - _mean_table_index(d2_tmp, duration_idx)
# delta_duration[n] = d1_tmp[:, duration_idx].mean() - d2_tmp[:, duration_idx].mean()
delta_success[n] = _mean_table_index(d1_tmp, success_idx) - _mean_table_index(d2_tmp, success_idx)
# delta_success[n] = d1_tmp[:, success_idx].mean() - d2_tmp[:, success_idx].mean()
names_idx = columns.index('names')
d1_tmp = _sort_table_index(table_hh1, success_idx)
# d1_tmp = d1_tmp[d1_tmp[:, success_idx].argsort()]
return {'ratio': list(zip(_table_slice(d1_tmp, source_idx), _table_slice(d1_tmp, success_idx))),
'increase': sorted(delta_success.items(), key=lambda k: k[1], reverse=False),
'newEvents': new_hosts}
def query_most_errors_by_period(project_id, start_time, end_time, conn=None):
params = {
"project_id": project_id, "startTimestamp": start_time, "endTimestamp": end_time,
"step_size": metrics.__get_step_size(endTimestamp=end_time, startTimestamp=start_time, density=3)
}
conditions = ["event_type = 'ERROR'"]
query = f"""WITH toUInt32(toStartOfInterval(toDateTime(%(startTimestamp)s/1000), INTERVAL %(step_size)s second)) AS start,
toUInt32(toStartOfInterval(toDateTime(%(endTimestamp)s/1000), INTERVAL %(step_size)s second)) AS end
SELECT T1.hh, count(T2.session_id) as sessions, T2.name as names,
groupUniqArray(T2.source) as sources
FROM (SELECT arrayJoin(arrayMap(x -> toDateTime(x), range(start, end, %(step_size)s))) as hh) AS T1
LEFT JOIN (SELECT session_id, name, source, message, toStartOfInterval(datetime, INTERVAL %(step_size)s second) as dtime
FROM experimental.events
WHERE project_id = {project_id}
AND {" AND ".join(conditions)}) AS T2 ON T2.dtime = T1.hh
GROUP BY T1.hh, T2.name
ORDER BY T1.hh DESC;"""
# print("----------------------------------")
# print(query)
# print("----------------------------------")
if conn is None:
with ch_client.ClickHouseClient() as conn:
query = conn.format(query=query, params=params)
res = conn.execute(query=query)
else:
query = conn.format(query=query, params=params)
res = conn.execute(query=query)
table_hh1, table_hh2, columns, this_period_errors, last_period_errors = __get_two_values(res, time_index='hh',
name_index='names')
del res
new_errors = [x for x in this_period_errors if x not in last_period_errors]
common_errors = [x for x in this_period_errors if x not in new_errors]
sessions_idx = columns.index('sessions')
names_idx = columns.index('names')
percentage_errors = dict()
total = _sum_table_index(table_hh1, sessions_idx)
# total = table_hh1[:, sessions_idx].sum()
error_increase = dict()
for n in this_period_errors:
percentage_errors[n] = _sum_table_index(_table_where(table_hh1, names_idx, n), sessions_idx) / total
# percentage_errors[n] = (table_hh1[table_hh1[:, names_idx] == n][:, sessions_idx].sum())/total
for n in common_errors:
error_increase[n] = _sum_table_index(_table_where(table_hh1, names_idx, n), names_idx) - _sum_table_index(
_table_where(table_hh2, names_idx, n), names_idx)
# error_increase[n] = table_hh1[table_hh1[:, names_idx] == n][:, names_idx].sum() - table_hh2[table_hh2[:, names_idx] == n][:, names_idx].sum()
return {'ratio': sorted(percentage_errors.items(), key=lambda k: k[1], reverse=True),
'increase': sorted(error_increase.items(), key=lambda k: k[1], reverse=True),
'newEvents': new_errors}
def query_cpu_memory_by_period(project_id, start_time, end_time, conn=None):
params = {
"project_id": project_id, "startTimestamp": start_time, "endTimestamp": end_time,
"step_size": metrics.__get_step_size(endTimestamp=end_time, startTimestamp=start_time, density=3)
}
conditions = ["event_type = 'PERFORMANCE'"]
query = f"""WITH toUInt32(toStartOfInterval(toDateTime(%(startTimestamp)s/1000), INTERVAL %(step_size)s second)) AS start,
toUInt32(toStartOfInterval(toDateTime(%(endTimestamp)s/1000), INTERVAL %(step_size)s second)) AS end
SELECT T1.hh, count(T2.session_id) as sessions, avg(T2.avg_cpu) as cpu_used,
avg(T2.avg_used_js_heap_size) as memory_used, T2.url_host as names, groupUniqArray(T2.url_path) as sources
FROM (SELECT arrayJoin(arrayMap(x -> toDateTime(x), range(start, end, %(step_size)s))) as hh) AS T1
LEFT JOIN (SELECT session_id, url_host, url_path, avg_used_js_heap_size, avg_cpu, toStartOfInterval(datetime, INTERVAL %(step_size)s second) as dtime
FROM experimental.events
WHERE project_id = {project_id}
AND {" AND ".join(conditions)}) AS T2 ON T2.dtime = T1.hh
GROUP BY T1.hh, T2.url_host
ORDER BY T1.hh DESC;"""
if conn is None:
with ch_client.ClickHouseClient() as conn:
query = conn.format(query=query, params=params)
res = conn.execute(query=query)
else:
query = conn.format(query=query, params=params)
res = conn.execute(query=query)
table_hh1, table_hh2, columns, this_period_resources, last_period_resources = __get_two_values(res, time_index='hh',
name_index='names')
del res
memory_idx = columns.index('memory_used')
cpu_idx = columns.index('cpu_used')
_tmp = _mean_table_index(table_hh2, memory_idx)
# _tmp = table_hh2[:, memory_idx].mean()
# TODO: what if _tmp=0 ?
_tmp = 1 if _tmp == 0 else _tmp
return {'cpuIncrease': _mean_table_index(table_hh1, cpu_idx) - _mean_table_index(table_hh2, cpu_idx),
'memoryIncrease': (_mean_table_index(table_hh1, memory_idx) - _tmp) / _tmp}
def query_click_rage_by_period(project_id, start_time, end_time, conn=None):
params = {
"project_id": project_id, "startTimestamp": start_time, "endTimestamp": end_time,
"step_size": metrics.__get_step_size(endTimestamp=end_time, startTimestamp=start_time, density=3)}
conditions = ["issue_type = 'click_rage'", "event_type = 'ISSUE'"]
query = f"""WITH toUInt32(toStartOfInterval(toDateTime(%(startTimestamp)s/1000), INTERVAL %(step_size)s second)) AS start,
toUInt32(toStartOfInterval(toDateTime(%(endTimestamp)s/1000), INTERVAL %(step_size)s second)) AS end
SELECT T1.hh, count(T2.session_id) as sessions, T2.url_host as names, groupUniqArray(T2.url_path) as sources
FROM (SELECT arrayJoin(arrayMap(x -> toDateTime(x), range(start, end, %(step_size)s))) as hh) AS T1
LEFT JOIN (SELECT session_id, url_host, url_path, toStartOfInterval(datetime, INTERVAL %(step_size)s second ) as dtime
FROM experimental.events
WHERE project_id = %(project_id)s
AND datetime >= toDateTime(%(startTimestamp)s/1000)
AND datetime < toDateTime(%(endTimestamp)s/1000)
AND {" AND ".join(conditions)}) AS T2 ON T2.dtime = T1.hh
GROUP BY T1.hh, T2.url_host
ORDER BY T1.hh DESC;"""
if conn is None:
with ch_client.ClickHouseClient() as conn:
query = conn.format(query=query, params=params)
print("--------------------")
print(query)
print("--------------------")
res = conn.execute(query=query)
else:
query = conn.format(query=query, params=params)
print("--------------------")
print(query)
print("--------------------")
res = conn.execute(query=query)
table_hh1, table_hh2, columns, this_period_rage, last_period_rage = __get_two_values(res, time_index='hh',
name_index='names')
del res
new_names = [x for x in this_period_rage if x not in last_period_rage]
common_names = [x for x in this_period_rage if x not in new_names]
sessions_idx = columns.index('sessions')
names_idx = columns.index('names')
raged_increment = dict()
# TODO verify line (188) _tmp = table_hh2[:, sessions_idx][n].sum()
for n in common_names:
if n is None:
continue
_tmp = _sum_table_index(_table_where(table_hh2, names_idx, n), sessions_idx)
# _tmp = table_hh2[:, sessions_idx][n].sum()
raged_increment[n] = (_sum_table_index(_table_where(table_hh1, names_idx, n), sessions_idx) - _tmp) / _tmp
# raged_increment[n] = (table_hh1[:, sessions_idx][n].sum()-_tmp)/_tmp
total = _sum_table_index(table_hh1, sessions_idx)
# total = table_hh1[:, sessions_idx].sum()
return {'ratio': list(
zip(_table_slice(table_hh1, names_idx), map(lambda k: k / total, _table_slice(table_hh1, sessions_idx)))),
'increase': sorted(raged_increment.items(), key=lambda k: k[1], reverse=True),
'newEvents': new_names,
}
def fetch_selected(project_id, data: schemas_ee.GetInsightsSchema):
output = {}
with ch_client.ClickHouseClient() as conn:
if schemas_ee.InsightCategories.errors in data.categories:
output[schemas_ee.InsightCategories.errors] = query_most_errors_by_period(project_id=project_id,
start_time=data.startTimestamp,
end_time=data.endTimestamp,
conn=conn)
if schemas_ee.InsightCategories.network in data.categories:
output[schemas_ee.InsightCategories.network] = query_requests_by_period(project_id=project_id,
start_time=data.startTimestamp,
end_time=data.endTimestamp,
conn=conn)
if schemas_ee.InsightCategories.rage in data.categories:
output[schemas_ee.InsightCategories.rage] = query_click_rage_by_period(project_id=project_id,
start_time=data.startTimestamp,
end_time=data.endTimestamp,
conn=conn)
if schemas_ee.InsightCategories.resources in data.categories:
output[schemas_ee.InsightCategories.resources] = query_cpu_memory_by_period(project_id=project_id,
start_time=data.startTimestamp,
end_time=data.endTimestamp,
conn=conn)
return output
# if __name__ == '__main__':
# # configs
# start = '2022-04-19'
# end = '2022-04-21'
# projectId = 1307
# time_step = 'hour'
#
# # Errors widget
# print('Errors example')
# res = query_most_errors_by_period(projectId, start_time=start, end_time=end, time_step=time_step)
# print(res)
#
# # Resources widgets
# print('resources example')
# res = query_cpu_memory_by_period(projectId, start_time=start, end_time=end, time_step=time_step)
#
# # Network widgets
# print('Network example')
# res = query_requests_by_period(projectId, start_time=start, end_time=end, time_step=time_step)
# print(res)