Re: TOAST Fields serialisation/deserialization performance

From: Laurenz Albe <laurenz(dot)albe(at)cybertec(dot)at>
To: Piyush Katariya <corporate(dot)piyush(at)gmail(dot)com>, pgsql-performance(at)lists(dot)postgresql(dot)org
Subject: Re: TOAST Fields serialisation/deserialization performance
Date: 2023-07-26 19:39:25
Message-ID: 5325acd55ae46230a3f77a05649f9d66e4c36c39.camel@cybertec.at
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-performance

On Wed, 2023-07-26 at 18:15 +0530, Piyush Katariya wrote:
> I have a few queries regarding the TOAST Fields serialisation/deserialization performance.
>
> The use case i am trying to solve here is to have millions of partitions and aggregate the data in array field.
>
> I wish to know if i declare certain column in table as "array of UDT/JSONB" and enable
> either lz4 or zstd compression on it, does appending or prepending to that array or even
> changing the intermediate fields of UDT/JSONB objects. in that array has a runtime cost
> of full array data de-serialization every single time. If i perform any UPDATE operation
> on its elements or add/remove new elements from any position, does PG rewrites the new
> version of the column value regardless of its size.

Updating even a small part of a large JSONB value requires that the entire thing is
read and written, causing a lot of data churn.

This is inefficient, and you shouldn't use large JSONB values if you plan to do that.

If the data have a regular structure, use a regular relational data model.
Otherwise, one idea might be to split the JSONB in several parts and store each
of those parts in a different table row. That would reduce the impact.

Yours,
Laurenz Albe

In response to

Responses

Browse pgsql-performance by date

  From Date Subject
Next Message Piyush Katariya 2023-07-26 19:48:17 Re: TOAST Fields serialisation/deserialization performance
Previous Message Piyush Katariya 2023-07-26 12:45:38 TOAST Fields serialisation/deserialization performance