Re: Bug in pg_upgrade standby rsync doc - Mailing list pgsql-docs

On Thu, Dec  7, 2017 at 11:28:12AM -0600, Don Seiler wrote:
> In step 10 of the pg_upgrade doc at https://www.postgresql.org/docs/9.6/static/
> pgupgrade.html, it uses this example for rsyncing the main $PGDATA dir to the
> standby:
> 
> 
> rsync --archive --delete --hard-links --size-only --no-inc-recursive /opt/PostgreSQL/9.5/data \
>       /opt/PostgreSQL/9.6/data standby.example.com:/opt/PostgreSQL
> 
> 
> However when I ran this (substituting /var/lib/pgsql for /opt/PostgreSQL), I
> found that it put a directory under /var/lib/pgsql/data, and my 9.6/data dir
> was still empty. Furthermore, what was in /var/lib/pgsql/data appeared to be
> the old 9.2 contents.
> 
> I suspect this is because the "data" directories are two levels below the
> parent directory specified at the end of the command. When I used the similar
> command for separate tablespaces, it worked as expected.
> 
> When I run the command with the "data" dir, so that I'm just specifying 9.2 and
> 9.6, it works as desired. eg:
> 
> 
> rsync --archive --delete --hard-links --size-only --no-inc-recursive /opt/PostgreSQL/9.5 \
>       /opt/PostgreSQL/9.6 standby.example.com:/opt/PostgreSQL
> 
> 
> FWIW I was upgrading from 9.2.22 to 9.6.6 on CentOS 6.

Sorry I am just getting to this.  I was able to reproduce your results
with rsync version 3.1.1.  With /opt/PostgreSQL containing:

    /opt/PostgreSQL
    /opt/PostgreSQL/9.5
    /opt/PostgreSQL/9.5/data
    /opt/PostgreSQL/9.5/data/x
    /opt/PostgreSQL/9.6
    /opt/PostgreSQL/9.6/data
    /opt/PostgreSQL/9.6/data/y

This script:

    TMP="/tmp"
    
    rm -rf $TMP/PostgreSQL
    mkdir -p $TMP/PostgreSQL/9.5/data
    mkdir -p $TMP/PostgreSQL/9.6/data
    
    rsync --archive --delete --hard-links --size-only \
        --no-inc-recursive /opt/PostgreSQL/9.5 \
        /opt/PostgreSQL/9.6 $TMP/PostgreSQL
    
    find $TMP/PostgreSQL -print

Yields:

    /tmp/PostgreSQL
    /tmp/PostgreSQL/9.5
    /tmp/PostgreSQL/9.5/data
-->    /tmp/PostgreSQL/9.5/data/x
    /tmp/PostgreSQL/9.6
    /tmp/PostgreSQL/9.6/data
-->    /tmp/PostgreSQL/9.6/data/y

which is correct, but if I change rsync to match our docs:

    rsync --archive --delete --hard-links --size-only \
        --no-inc-recursive /opt/PostgreSQL/9.5/data \
        /opt/PostgreSQL/9.6/data $TMP/PostgreSQL

I get:

    /rtmp/PostgreSQL
    /rtmp/PostgreSQL/9.5
    /rtmp/PostgreSQL/9.5/data
    /rtmp/PostgreSQL/9.6
    /rtmp/PostgreSQL/9.6/data
-->    /rtmp/PostgreSQL/data
-->    /rtmp/PostgreSQL/data/x
-->    /rtmp/PostgreSQL/data/y

which is incorrect and the behavior you reported.

The incorrect example was added a few months ago:

    commit 9521ce4a7a1125385fb4de9689f345db594c516a
    Author: Bruce Momjian <bruce@momjian.us>
    Date:   Wed Sep 13 09:11:28 2017 -0400
    
        docs:  improve pg_upgrade standby instructions
    
        This makes it clear that pg_upgrade standby upgrade instructions should
        only be used in link mode, adds examples, and explains how rsync works
        with links.
    
        Reported-by: Andreas Joseph Krogh
    
        Discussion: https://postgr.es/m/VisenaEmail.6c.c0e592c5af4ef0a2.15e785dcb61@tc7-visena
    
        Backpatch-through: 9.5

but the generic syntax mentioning the data directory has been there for
a while.  I am wondering if people had to test this to get it working
and didn't report that saying datadir was inaccurate.

I propse the attached patch to fix the generic syntax and the example. 
I will backpatch it through 9.5.  As you stated, the tablespace example
is fine.

-- 
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

+ As you are, so once was I.  As I am, so you will be. +
+                      Ancient Roman grave inscription +