Re: Logical Replication - detail message with names of missing columns - Mailing list pgsql-hackers

From Kyotaro Horiguchi
Subject Re: Logical Replication - detail message with names of missing columns
Date
Msg-id 20200908.102034.2079793754955685311.horikyota.ntt@gmail.com
Whole thread Raw
In response to Logical Replication - detail message with names of missing columns  (Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>)
Responses Re: Logical Replication - detail message with names of missing columns  (Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>)
List pgsql-hackers
Thank you for working on this.

At Mon, 7 Sep 2020 16:30:59 +0530, Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> wrote in 
> Hi,
> 
> I observed that, in logical replication when a subscriber is missing
> some columns, it currently emits an error message that says "some"
> columns are missing(see logicalrep_rel_open()), but it doesn't specify
> what the missing column names are. The comment there also says that
> finding the missing column names is a todo item(/* TODO, detail
> message with names of missing columns */).
> 
> I propose a patch to find the missing columns on the subscriber
> relation using the publisher relation columns and show them in the
> error message which can make the error more informative to the user.

+1 for objective. However, that can be done simpler way that doesn't
need additional loops by using bitmapset to hold missing remote
attribute numbers. This also make the variable "found" useless.

> Here's a snapshot how the error looks with the patch:
> 2020-09-04 01:00:36.721 PDT [843128] ERROR:  logical replication
> target relation "public.test_1" is missing "b1, d1" replicated columns
> 2020-09-04 01:00:46.784 PDT [843132] ERROR:  logical replication
> target relation "public.test_1" is missing "b1" replicated columns
> 2020-09-06 21:24:53.645 PDT [902945] ERROR:  logical replication
> target relation "public.test_1" is missing "a1, c1, d1, b1" replicated
> columns
> 
> Thoughts?

FWIW, I would prefer that the message be like

 logical replication target relation "public.test_1" is missing
 replicated columns: "a1", "c1"

I'm not sure we need to have separate messages for singlar and plural.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center



pgsql-hackers by date:

Previous
From: Thomas Munro
Date:
Subject: Re: Division in dynahash.c due to HASH_FFACTOR
Next
From: Amit Kapila
Date:
Subject: Re: Transactions involving multiple postgres foreign servers, take 2