openreplay/api/chalicelib/core/insights.py
Taha Yassine Kraiem d4ffd10af6 feat(api): changed insights feature retention query
feat(db): optimized indexes for feature retention
2021-09-22 15:04:35 +02:00

296 lines
17 KiB
Python

from chalicelib.core import sessions_metas
from chalicelib.utils import args_transformer
from chalicelib.utils import helper, dev
from chalicelib.utils import pg_client
from chalicelib.utils.TimeUTC import TimeUTC
from chalicelib.utils.metrics_helper import __get_step_size
import math
from chalicelib.core.dashboard import __get_constraints, __get_constraint_values
def __transform_journey(rows):
nodes = []
links = []
for r in rows:
source = r["source_event"][r["source_event"].index("_"):]
target = r["target_event"][r["target_event"].index("_"):]
if source not in nodes:
nodes.append(source)
if target not in nodes:
nodes.append(target)
links.append({"source": nodes.index(source), "target": nodes.index(target), "value": r["value"]})
return {"nodes": nodes, "links": sorted(links, key=lambda x: x["value"], reverse=True)}
JOURNEY_DEPTH = 5
JOURNEY_TYPES = {
"PAGES": {"table": "events.pages", "column": "base_path", "table_id": "message_id"},
"CLICK": {"table": "events.clicks", "column": "label", "table_id": "message_id"},
"VIEW": {"table": "events_ios.views", "column": "name", "table_id": "seq_index"},
"EVENT": {"table": "events_common.customs", "column": "name", "table_id": "seq_index"}
}
@dev.timed
def get_journey(project_id, startTimestamp=TimeUTC.now(delta_days=-1), endTimestamp=TimeUTC.now(), filters=[], **args):
pg_sub_query_subset = __get_constraints(project_id=project_id, data=args, duration=True, main_table="sessions",
time_constraint=True)
event_start = None
event_table = JOURNEY_TYPES["PAGES"]["table"]
event_column = JOURNEY_TYPES["PAGES"]["column"]
event_table_id = JOURNEY_TYPES["PAGES"]["table_id"]
extra_values = {}
for f in filters:
if f["type"] == "START_POINT":
event_start = f["value"]
elif f["type"] == "EVENT_TYPE" and JOURNEY_TYPES.get(f["value"]):
event_table = JOURNEY_TYPES[f["value"]]["table"]
event_column = JOURNEY_TYPES[f["value"]]["column"]
elif f["type"] in [sessions_metas.meta_type.USERID, sessions_metas.meta_type.USERID_IOS]:
pg_sub_query_subset.append(f"sessions.user_id = %(user_id)s")
extra_values["user_id"] = f["value"]
with pg_client.PostgresClient() as cur:
pg_query = f"""SELECT source_event,
target_event,
MAX(target_id) max_target_id,
MAX(source_id) max_source_id,
count(*) AS value
FROM (SELECT event_number || '_' || value as target_event,
message_id AS target_id,
LAG(event_number || '_' || value, 1) OVER ( PARTITION BY session_rank ) AS source_event,
LAG(message_id, 1) OVER ( PARTITION BY session_rank ) AS source_id
FROM (SELECT value,
session_rank,
message_id,
ROW_NUMBER() OVER ( PARTITION BY session_rank ORDER BY timestamp ) AS event_number
{f"FROM (SELECT * FROM (SELECT *, MIN(mark) OVER ( PARTITION BY session_id , session_rank ORDER BY timestamp ) AS max FROM (SELECT *, CASE WHEN value = %(event_start)s THEN timestamp ELSE NULL END as mark"
if event_start else ""}
FROM (SELECT session_id,
message_id,
timestamp,
value,
SUM(new_session) OVER (ORDER BY session_id, timestamp) AS session_rank
FROM (SELECT *,
CASE
WHEN source_timestamp IS NULL THEN 1
ELSE 0 END AS new_session
FROM (SELECT session_id,
{event_table_id} AS message_id,
timestamp,
{event_column} AS value,
LAG(timestamp)
OVER (PARTITION BY session_id ORDER BY timestamp) AS source_timestamp
FROM {event_table} INNER JOIN public.sessions USING (session_id)
WHERE {" AND ".join(pg_sub_query_subset)}
) AS related_events) AS ranked_events) AS processed
{") AS marked) AS maxed WHERE timestamp >= max) AS filtered" if event_start else ""}
) AS sorted_events
WHERE event_number <= %(JOURNEY_DEPTH)s) AS final
WHERE source_event IS NOT NULL
and target_event IS NOT NULL
GROUP BY source_event, target_event
ORDER BY value DESC
LIMIT 20;"""
params = {"project_id": project_id, "startTimestamp": startTimestamp,
"endTimestamp": endTimestamp, "event_start": event_start, "JOURNEY_DEPTH": JOURNEY_DEPTH,
**__get_constraint_values(args), **extra_values}
# print(cur.mogrify(pg_query, params))
cur.execute(cur.mogrify(pg_query, params))
rows = cur.fetchall()
return __transform_journey(rows)
def __compute_retention_percentage(rows):
if rows is None or len(rows) == 0:
return rows
t = -1
for r in rows:
if r["week"] == 0:
t = r["usersCount"]
r["percentage"] = r["usersCount"] / t
return rows
def __complete_retention(rows, start_date, end_date=None):
if rows is None or len(rows) == 0:
return rows
max_week = 10
week = 0
delta_date = 0
while max_week > 0:
start_date += TimeUTC.MS_WEEK
if end_date is not None and start_date >= end_date:
break
delta = 0
if delta_date + week >= len(rows) \
or delta_date + week < len(rows) and rows[delta_date + week]["firstConnexionWeek"] > start_date:
for i in range(max_week):
if end_date is not None and start_date + i * TimeUTC.MS_WEEK >= end_date:
break
neutral = {
"firstConnexionWeek": start_date,
"week": i,
"usersCount": 0,
"connectedUsers": [],
"percentage": 0
}
rows.insert(delta_date + week + i, neutral)
delta = i
else:
for i in range(max_week):
if end_date is not None and start_date + i * TimeUTC.MS_WEEK >= end_date:
break
neutral = {
"firstConnexionWeek": start_date,
"week": i,
"usersCount": 0,
"connectedUsers": [],
"percentage": 0
}
if delta_date + week + i < len(rows) \
and i != rows[delta_date + week + i]["week"]:
rows.insert(delta_date + week + i, neutral)
elif delta_date + week + i >= len(rows):
rows.append(neutral)
delta = i
week += delta
max_week -= 1
delta_date += 1
return rows
@dev.timed
def get_users_retention(project_id, startTimestamp=TimeUTC.now(delta_days=-70), endTimestamp=TimeUTC.now(), filters=[],
**args):
startTimestamp = TimeUTC.trunc_week(startTimestamp)
endTimestamp = startTimestamp + 10 * TimeUTC.MS_WEEK
pg_sub_query = __get_constraints(project_id=project_id, data=args, duration=True, main_table="sessions",
time_constraint=True)
pg_sub_query.append("user_id IS NOT NULL")
with pg_client.PostgresClient() as cur:
pg_query = f"""SELECT EXTRACT(EPOCH FROM first_connexion_week::date)::bigint*1000 AS first_connexion_week,
FLOOR(DATE_PART('day', connexion_week - first_connexion_week) / 7)::integer AS week,
COUNT(DISTINCT connexions_list.user_id) AS users_count,
ARRAY_AGG(DISTINCT connexions_list.user_id) AS connected_users
FROM (SELECT DISTINCT user_id, MIN(DATE_TRUNC('week', to_timestamp(start_ts / 1000))) AS first_connexion_week
FROM sessions
WHERE {" AND ".join(pg_sub_query)}
AND NOT EXISTS((SELECT 1
FROM sessions AS bsess
WHERE bsess.start_ts<EXTRACT('EPOCH' FROM DATE_TRUNC('week', to_timestamp(%(startTimestamp)s / 1000))) * 1000
AND project_id = %(project_id)s
AND bsess.user_id = sessions.user_id
LIMIT 1))
GROUP BY user_id) AS users_list
LEFT JOIN LATERAL (SELECT DATE_TRUNC('week', to_timestamp(start_ts / 1000)::timestamp) AS connexion_week,
user_id
FROM sessions
WHERE users_list.user_id = sessions.user_id
AND first_connexion_week <=
DATE_TRUNC('week', to_timestamp(sessions.start_ts / 1000)::timestamp)
AND sessions.project_id = 1
AND sessions.start_ts < (%(endTimestamp)s - 1)
GROUP BY connexion_week, user_id) AS connexions_list ON (TRUE)
GROUP BY first_connexion_week, week
ORDER BY first_connexion_week, week;"""
params = {"project_id": project_id, "startTimestamp": startTimestamp,
"endTimestamp": endTimestamp, **__get_constraint_values(args)}
# print(cur.mogrify(pg_query, params))
cur.execute(cur.mogrify(pg_query, params))
rows = cur.fetchall()
rows = __compute_retention_percentage(helper.list_to_camel_case(rows))
return __complete_retention(rows=rows, start_date=startTimestamp, end_date=TimeUTC.now())
@dev.timed
def get_feature_retention(project_id, startTimestamp=TimeUTC.now(delta_days=-70), endTimestamp=TimeUTC.now(),
filters=[],
**args):
startTimestamp = TimeUTC.trunc_week(startTimestamp)
endTimestamp = startTimestamp + 10 * TimeUTC.MS_WEEK
pg_sub_query = __get_constraints(project_id=project_id, data=args, duration=True, main_table="sessions",
time_constraint=True)
pg_sub_query.append("user_id IS NOT NULL")
pg_sub_query.append("feature.timestamp >= %(startTimestamp)s")
pg_sub_query.append("feature.timestamp < %(endTimestamp)s")
event_table = JOURNEY_TYPES["PAGES"]["table"]
event_column = JOURNEY_TYPES["PAGES"]["column"]
extra_values = {"value": "/"}
default = True
for f in filters:
if f["type"] == "EVENT_TYPE" and JOURNEY_TYPES.get(f["value"]):
event_table = JOURNEY_TYPES[f["value"]]["table"]
event_column = JOURNEY_TYPES[f["value"]]["column"]
elif f["type"] in [sessions_metas.meta_type.USERID, sessions_metas.meta_type.USERID_IOS]:
pg_sub_query.append(f"sessions.user_id = %(user_id)s")
extra_values["user_id"] = f["value"]
# TODO: This will change later when the search is clear
default = False
extra_values["value"] = f["value"]
pg_sub_query.append(f"feature.{event_column} = %(value)s")
with pg_client.PostgresClient() as cur:
if default:
# get most used value
pg_query = f"""SELECT {event_column} AS value, COUNT(*) AS count
FROM {event_table} AS feature INNER JOIN public.sessions USING (session_id)
WHERE {" AND ".join(pg_sub_query[:-1])}
AND length({event_column}) > 2
GROUP BY value
ORDER BY count DESC
LIMIT 1;"""
params = {"project_id": project_id, "startTimestamp": startTimestamp,
"endTimestamp": endTimestamp, **__get_constraint_values(args), **extra_values}
cur.execute(cur.mogrify(pg_query, params))
row = cur.fetchone()
if row is not None:
extra_values["value"] = row["value"]
pg_query = f"""SELECT EXTRACT(EPOCH FROM first_connexion_week::date)::bigint*1000 AS first_connexion_week,
FLOOR(DATE_PART('day', connexion_week - first_connexion_week) / 7)::integer AS week,
COUNT(DISTINCT connexions_list.user_id) AS users_count,
ARRAY_AGG(DISTINCT connexions_list.user_id) AS connected_users
FROM (SELECT user_id, DATE_TRUNC('week', to_timestamp(first_connexion_week / 1000)) AS first_connexion_week
FROM(SELECT DISTINCT user_id, MIN(start_ts) AS first_connexion_week
FROM sessions INNER JOIN {event_table} AS feature USING (session_id)
WHERE {" AND ".join(pg_sub_query)}
AND user_id IS NOT NULL
AND NOT EXISTS((SELECT 1
FROM sessions AS bsess INNER JOIN {event_table} AS bfeature USING (session_id)
WHERE bsess.start_ts<%(startTimestamp)s
AND project_id = %(project_id)s
AND bsess.user_id = sessions.user_id
AND bfeature.timestamp<%(startTimestamp)s
AND bfeature.{event_column}=%(value)s
LIMIT 1))
GROUP BY user_id) AS raw_users_list) AS users_list
LEFT JOIN LATERAL (SELECT DATE_TRUNC('week', to_timestamp(start_ts / 1000)::timestamp) AS connexion_week,
user_id
FROM sessions INNER JOIN {event_table} AS feature USING(session_id)
WHERE users_list.user_id = sessions.user_id
AND first_connexion_week <=
DATE_TRUNC('week', to_timestamp(sessions.start_ts / 1000)::timestamp)
AND sessions.project_id = 1
AND sessions.start_ts < (%(endTimestamp)s - 1)
AND feature.timestamp >= %(startTimestamp)s
AND feature.timestamp < %(endTimestamp)s
AND feature.{event_column} = %(value)s
GROUP BY connexion_week, user_id) AS connexions_list ON (TRUE)
GROUP BY first_connexion_week, week
ORDER BY first_connexion_week, week;"""
params = {"project_id": project_id, "startTimestamp": startTimestamp,
"endTimestamp": endTimestamp, **__get_constraint_values(args), **extra_values}
# print(cur.mogrify(pg_query, params))
cur.execute(cur.mogrify(pg_query, params))
rows = cur.fetchall()
rows = __compute_retention_percentage(helper.list_to_camel_case(rows))
return __complete_retention(rows=rows, start_date=startTimestamp, end_date=TimeUTC.now())