RE: reindex option for tuning load large data

"James Pang (chaolpan)" <chaolpan@xxxxxxxxx> · Sat, 18 Jun 2022 04:00:14 +0000

We have more than 8500 indexes , > 1000 tables, many partition tables too ; it’s safe to update pg_index set indisready=false and indisvisilbe=false , then reindexdb with parallel ?   reindex parallel got done by multiple sessions , each
 session reindex one index at the same time ,  reindex the one index done in serial instead of parallel ?

 Compared with  “set max_maintain_parallel_workers, and  run CREATE INDEX …” , which is faster ?

Thanks,

From: Vitalii Tymchyshyn <vit@xxxxxx> 

Sent: Saturday, June 18, 2022 11:49 AM

To: James Pang (chaolpan) <chaolpan@xxxxxxxxx>

Cc: pgsql-performance@xxxxxxxxxxxxxxxxxxxx

Subject: Re: reindex option for tuning load large data

I believe you should be able to use reindexdb with parallel jobs:

https://www.postgresql.org/docs/13/app-reindexdb.html

It will still create multiple connections, but you won't need to run multiple commands.

чт, 16 черв. 2022 р. о 22:34 James Pang (chaolpan) <chaolpan@xxxxxxxxx> пише:

Hi ,
  We plan to migrate large database from Oracle to Postgres(version 13.6, OS Redhat8 Enterprise), we are checking options to make data load in Postgres fast. Data volume is about several TB,  thousands of indexes,  many large table with
 partitions.  We want to make data load running fast and avoid miss any indexes when reindexing. There are 2 options about reindex. Could you give some suggestions about the 2 options, which option is better.

Create tables and indexes( empty database) ,   update pg_index set indisready=false and inisvalid=false,  then  load data use COPY from csv ,  then reindex table …

Reindex on Postgres 13.6 not support parallel ,right?  So we need to start multiple session to reindex  multiple tables/indexes in parallel.

2).  Use pg_dump to dump meta data only , then copy “CREATE INDEX … sql “

        Drop indexes before data load

       After data load, increase
max_parallel_maintenance_workers, maintenance_work_mem
       Run CREATE INDEX … sql  to leverage parallel create index feature.

Thanks,

James 