Partition Granularity not working on postgres internal aggregation month wise

[alokthakur142]

Describe the bug
I want to partition my activity data month-wise. However, cubejs is combining data of more than 1 months in this manner:
Oct 2024 -> From 1 Oct 2024 to 17 Dec 2024.
Nov 2024 -> From 1 Oct 2024 to 17 Dec 2024.
Dec 2024 -> From 1 Oct 2024 to 17 Dec 2024

Expected behavior
It should store only one month data in one partition:
Oct 2024 -> From 1 Oct 2024 to 31 Oct 2024.
Nov 2024 -> From 1 Nov 2024 to 30 Nov 2024.
Dec 2024 -> From 1 Dec 2024 to 17 Dec 2024

cube(`Activity`, {
  sql: `SELECT id, user_id, type, activity_time, remarks, modified_date FROM public.activity where company = ${COMPILE_CONTEXT.securityContext.companyId} and ${FILTER_PARAMS.Activity.activityTime.filter('activity_time')}
  `,
  
  pre_aggregations: {
    main: {
      type: `original_sql`,
      external: false,
      time_dimension: CUBE.activityTime,
      partition_granularity: `month`,
      allow_non_strict_date_range_match: true,
      refresh_key: {
        every: `1 hour`,
	      sql: `SELECT MAX(t.modified_date) FROM public.activity where company = ${COMPILE_CONTEXT.securityContext.companyId} and ${FILTER_PARAMS.Activity.activityTime.filter('activity_time')}`,
      },
      build_range_start: {
        sql: `SELECT date_trunc('month', NOW()) - interval '2 year'`
      },
      build_range_end: {
        sql: `SELECT date_trunc('month', NOW()) + interval '1 month'`
      }
    }
  },
  
  measures: {
    count: {
      type: `count`
    }
  },
  
  dimensions: {
    id: {
      sql: `id`,
      type: `number`,
      primaryKey: true
    },

    type: {
      sql: `type`,
      type: `string`
    },
    
    remarks: {
      sql: `remarks`,
      type: `string`
    },
    
    activityTime: {
      sql: `activity_time`,
      type: `time`
    }
  }
});

Request:

{"query":{
"limit":10,
"offset":0,
"order":{},
"timeDimensions":[{"dimension":"Activity.activityTime","granularity":"day","dateRange":"This Quarter"}],
"dimensions":["Activity.id"],
"measures":["Activity.count"],
"filters":[],
"renewQuery":false
},"queryType":"multi"}

Result:

Activity ID  Count
100                 3
101                 3
102                 3

Version:
1.0.0 to 1.0.7

[alokthakur142]

@igorlukanin

[alokthakur142]

Here is log for pre aggregation planning:

"preAggregationsSchema":"pre_x","loadSql":["CREATE TABLE pre_x.activity_main20241001 AS SELECT id, user_id, remarks, activity_time, modified_date FROM public.activity where tenant = x and ((activity_time >= $1::timestamptz AND activity_time <= $2::timestamptz))\n  ",["2024-10-01T00:00:00.000Z","2024-12-31T23:59:59.999Z"],{}]

It is sending wrong parameter in $2. It should be 2024-10-31T23:59:59 instead.

Please help on this.

[igorlukanin]

@alokthakur142 Does anything change if you use the rollup pre-aggregation type?

[alokthakur142]

Yes, enabling rollup will not work with joins.

I have provided a playground to the user with a capability to choose attributes of User, Customer, Activity Type and other related entities.

Queries like: City wise Activity, User wise Activity, Designation wise etc is possible only by using original_sql.

Let me know if you have any question on this.

[igorlukanin]

I guess the issue here is that original_sql pre-aggregations are not supposed to be partitioned by a time dimension. I'll tag @paveltiunov so he can confirm this.

Queries like: City wise Activity, User wise Activity, Designation wise etc is possible only by using original_sql.

I don't think I follow. Once again, what prevents you from using a rollup pre-aggregation here?