Comments on: Understanding pg_repack: What Can Go Wrong – and How to Avoid It

By: jimis

jimis — Fri, 10 Nov 2023 14:17:22 +0000

Thank you for the insight into pg_repack, this I haven’t found documented elsewhere.

There is still, one mystery to me: why does pg_repack require a primary key or a UNIQUE key?

According to your article, it must be because of the CDC table:
CREATE TABLE repack.log_16423 (id bigserial PRIMARY KEY, pk repack.pk_16423, row public.t1)

We see it contains the public key of our table as a separate column. But where is it used and why? It baffles me that I can’t run pg_repack on a table without a UNIQUE column.

By: Amaan

Amaan — Wed, 06 Sep 2023 16:58:11 +0000

In reply to sarvesh. Man, this seems to be a bigger issue. AFAIK, You should have at least twice the amount of table size on the DB disk before running pg_repack. That is one of the basic validations of this extension. This seems to be an outage-like situation. Can you provide some insights into what happened after this issue?

By: sarvesh

sarvesh — Mon, 07 Aug 2023 09:45:51 +0000

Seeing below issue while pg_repack, db is hosted ve aws postgres aurora 12.x.
Table size is 650GB plus including index

ERROR: query failed: ERROR: could not write to file “base/pgsql_tmp/pgsql_tmp17644.58”: No space left on device

By: Jobin Augustine

Jobin Augustine — Thu, 01 Dec 2022 03:37:54 +0000

In reply to sushant kamboj. "facing issues" issue is a very broad statement. Without more details, others won't be able to understand.

By: sushant kamboj

sushant kamboj — Fri, 05 Aug 2022 07:24:47 +0000

We are facing issues while doing the repacking of tables for which we have kept the replica identity : FULL , to full fill our debezium use case.

By: Jobin Augustine

Jobin Augustine — Tue, 06 Jul 2021 14:43:51 +0000

In reply to Michael Banck. Thank you @Michael, Thank you for your feedback. Yes it was a formatting mistake and it is corrected as per your feedback. Hope you liked the article.

By: Michael Banck

Michael Banck — Sun, 27 Jun 2021 07:55:06 +0000

The ‘followed by the data copy:” sentence is malformatted and looks like part of the SQL session, so I read over it on first pass and then wondered where the actual copying happens.