Re: logical replication seems broken - Mailing list pgsql-hackers

From er@xs4all.nl
Subject Re: logical replication seems broken
Date
Msg-id 522121056.124036.1613147431818@webmailclassic.xs4all.nl
Whole thread Raw
In response to Re: logical replication seems broken  (Amit Kapila <amit.kapila16@gmail.com>)
Responses Re: logical replication seems broken
List pgsql-hackers
> On 02/12/2021 1:51 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
> 
>  
> On Fri, Feb 12, 2021 at 6:04 PM Erik Rijkers <er@xs4all.nl> wrote:
> >
> > Hello,
> >
> > I am seeing errors in replication in a test program that I've been running for years with very little change (since
2017,really [1]).
 

Hi,

Here is a test program.  Careful, it deletes stuff.  And it will need some changes:

I compile postgres server versions into directories like:
     $HOME/pg_stuff/pg_installations/pgsql.$project    where project is a name

The attached script (logrep_cascade_bug.sh)  assumes that two such compiled versions are present (on my machine they
arecalled HEAD and head0):
 
     $HOME/pg_stuff/pg_installations/pgsql.HEAD   --> git master as of today - friday 12 febr 2021
     $HOME/pg_stuff/pg_installations/pgsql.head0  --> 3063eb17593c  so that's from 11 febr, before the replication
changes

In the test script, bash variables 'project' (and 'BIN') reflect my set up - so should probably be changed.

The instance from today 12 february ('HEAD') has the bug:
  it keeps endlessly waiting/looping with 'NOK' (=Not OK).
  'Not OK' means: primary not identical to all replicas (replica1 seems ok, but replica2 remains empty)

The instance from yesterday 11 february ('head0') is ok:
  it finishes in 20 s after waiting/looping just 2 or 3 times
  'ok' means: all replicas are identical to primary (as proven by the md5s).

That's all I have for now - I have no deeper idea about what exactly goes wrong.

I hope that helps, let me know when you cannot reproduce the problem.

Erik Rijkers
Attachment

pgsql-hackers by date:

Previous
From: Ranier Vilela
Date:
Subject: Re: Possible dereference after null check (src/backend/executor/ExecUtils.c)
Next
From: Anastasia Lubennikova
Date:
Subject: Re: some pointless HeapTupleHeaderIndicatesMovedPartitions calls