Hello Alexander,

Do you have any date in mind, for releasing the same to have parallel execution?
Also we tried upgrading datasketches version from latest documentation, we are 
getting lot of C++ version issues.
Its very tough to install the new version. Any thoughts?

Thanks,
Rima Bhowmick.

From: Alexander Saydakov <[email protected]>
Reply-To: "[email protected]" <[email protected]>
Date: Friday, 14 April 2023 at 10:58 PM
To: "[email protected]" <[email protected]>
Subject: Re: [E] Postgres HLL is very slow

Hi Rima,
I am working on the datasketches extension to support parallel queries 
(distributed aggregation).
I expect to get this done in a matter of days.
Also we have just made some improvements to HLL merge speed in the core 
library. These changes were not released yet, but available in the master 
branch.
We have another HLL performance improvement in mind. I will work on it once I 
finish the parallel query support.


On Fri, Apr 14, 2023 at 3:33 AM Bhowmick, Rima <[email protected]> 
wrote:
Hello Team,

Here is the snapshot of the existing application:

TechStack: Postgres DB, Hive, Tableau UI
Postgres Plugin: DataSketches

Flow in brief:

  *   Hadoop Data pipeline job pushes pre-aggregated(using hive datasketches 
algo) active card data, along with other details to Hive.
  *   Another job populates that data to Postgres DB, finally having 3 years 
data of 4 regions for multiple countries.
  *   Tableau dashboard having live connection to Postgres DB.
  *   Tableau Query calling Postgres DB, to aggregate the binary/pre-aggregated 
data to get distinct card count (using DataSketches algorithm) and fetch data 
based on multiple filter conditions.
  *   Usually data would be of 3yrs for the span of 2 months, means total 6 
months of data to aggregate for a country on multiple conditions.

Usually this aggregation query response is quite slow. We have tried lot of 
different ways to resolve this,

Mainly datasketches part is making most of the time in execution.

Thanks & Regards,
Rima Bhowmick
Marketing Brand Analytics
[Logo    Description automatically generated]

Reply via email to