Reduce use of window function in sessions

dbt_project.yml

@@ -1,5 +1,6 @@
+profile: dbt-clickhouse
 name: 'jitsu'
-version: '0.1.1'
+version: '0.1.2'
 require-dbt-version: ">=0.20.0"
 config-version: 2
 
@@ -22,7 +23,7 @@ clean-targets:         # directories to be removed by `dbt clean`
 # Configuring models
 # Full documentation: https://docs.getdbt.com/docs/configuring-models
 models:
-  materialized: "{{var('jitsu_model_materialized', 'incremental')}}"
+  materialized: "{{var('jitsu_model_materialized', 'table')}}"
   pre-hook:
     - "{{ onrunstart() }}"   # On clickhouse enables allow_experimental_window_functions flag
 
@@ -33,7 +34,7 @@ dispatch:
 
 vars:
   # location of source data table
-  jitsu_events_table:
+  jitsu_events_table: "{{ source('jitsu', 'pageviews') }}" 
 
   # length of trailing window in minutes for session attribution.
   # We recommend enabling builtin Retrospective User Recognition feature for user stitching
@@ -49,4 +50,6 @@ vars:
   # define them here. Columns will be included in the `jitsu_sessions`
   # model as `first_<column>` and `last_<column>`.
   jitsu_pass_through_columns: []
-
+
+  jitsu_events_table_filter: "event_type='pageview'"
+  jitsu_model_materialized: 'table'

models/sessions/jitsu_events.sql

@@ -1,5 +1,9 @@
 {{ config(
     materialized = 'view'
     )}}
+{% if var('jitsu_events_table_filter') is defined %}
+    select * from {{var('jitsu_events_table')}} where user_anonymous_id is not null and {{var('jitsu_events_table_filter')}}
+{% else %}
+    select * from {{var('jitsu_events_table')}} where user_anonymous_id is not null
+{% endif %}
 
-select * from {{var('jitsu_events_table')}} where user_anonymous_id is not null

models/sessions/jitsu_sessions.sql

@@ -54,24 +54,26 @@ referrer_mapping as (
 
 agg as (
 
-    select distinct
+    select
 
         session_id,
-        user_anonymous_id,
-        last_value(coalesce(user_anonymous_id, user_internal_id, user_email)) over ({{window_clause}}) as blended_user_id,
-        min(_timestamp) over ( {{partition_by}} ) as session_start_timestamp,
-        max(_timestamp) over ( {{partition_by}} ) as session_end_timestamp,
-        count(*) over ( {{partition_by}} ) as pageviews,
+        coalesce(user_anonymous_id, user_id, user_email) as blended_user_id,
+        max(user_anonymous_id) as user_anonymous_id,
+        max(user_id) as user_id,
+        max(user_email) as user_email,
+        min(_timestamp) as session_start_timestamp,
+        max(_timestamp) session_end_timestamp,
+        count() as pageviews,
 
         {% for (key, value) in first_values.items() %}
-        first_value({{key}}) over ({{window_clause}}) as {{value}},
+        min({{key}}) as {{value}},
         {% endfor %}
 
         {% for (key, value) in last_values.items() %}
-        last_value({{key}}) over ({{window_clause}}) as {{value}}{% if not loop.last %},{% endif %}
+        max({{key}}) as {{value}}{% if not loop.last %},{% endif %}
         {% endfor %}
 
-    from pageviews_sessionized
+    from pageviews_sessionized GROUP BY session_id
 
 ),
 
@@ -104,7 +106,6 @@ tiers as (
     from diffs
 
 ),
-
 mapped as (
 
     select
@@ -119,3 +120,4 @@ mapped as (
 )
 
 select * from mapped
+