Message-ID: <505557CD.8090304@gmail.com>
Date: Sun, 16 Sep 2012 12:38:37 +0800
From: Rural Hunter <ruralhunter@gmail.com>
User-Agent: Mozilla/5.0 (Windows NT 6.1;
	rv:15.0) Gecko/20120907 Thunderbird/15.0.1
MIME-Version: 1.0
To: Bruce Momjian <bruce@momjian.us>
CC: pgsql-admin@postgresql.org
Subject: Re: pg_upgrade from 9.1.3 to 9.2 failed
References: <5052C402.8020003@gmail.com> <20120914142616.GA11265@momjian.us>
	<50535302.8080000@gmail.com> <20120914183915.GA5933@momjian.us>
	<5053F896.60005@gmail.com> <20120915180603.GA20907@momjian.us>
In-Reply-To: <20120915180603.GA20907@momjian.us>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit

于 2012/9/16 2:06, Bruce Momjian 写道:
> On Sat, Sep 15, 2012 at 11:40:06AM +0800, Rural Hunter wrote:
>>>>> The check is to make sure that once we have created all the user schema
>>>>> details in the new cluster, that there are the same number of objects in
>>>>> the new and old databases.
>>>>>
>>>>> Obviously there are a different number in your case here, but I don't
>>>>> know why those would be different, and in fact, because we have never
>>>>> hit this, there isn't even any debug output that shows the source of the
>>>>> difference.
>>>>>
>>>>> If I send you a patch can you compile it and send back the debug output
>>>>> it produces?
>>>>>
>>>> Yes sure, I will try to compile and retest with it.
>>> Actually, I have a simpler idea.  At the point where it fails, you can
>>> run pg_dump --schema-only on the testdb database in the old and new
>>> cluster and then diff those output files and email the result to us;  it
>>> should show the mismatch.  I am not sure if the dumps will output the
>>> objects in the same order, it might.
>>>
>> diff attached.
> OK, I see many new ALTER TABLE commands, but nothing that would cause a
> difference in relation count.
>
> Attached is a patch that will return the OID of the old/new mismatched
> entries.  Please research the pg_class objects on the old/new clusters
> that have the mismatch and let me know.  It might be something that
> isn't in the old cluster, or not in the new cluster.
>
I ran the pg_upgrade with the patch and found the problematic object is 
a toast object.
Copying user relation files
/raid/pgsql/base/6087920/6088238
Mismatch of relation OID in database "forummon": old OID 16439148, new 
OID 16439322

In old cluster:
# select * from pg_class WHERE oid=16439148;
relname | relnamespace | reltype | reloftype | relowner | relam | 
relfilenode | reltablespace | relpages | reltuples | reltoastrelid | 
reltoastidxid | relhasindex | relisshared | relpersistence | relkind | 
relnatts | relchecks | relhasoids | relhaspkey | relhasrules | 
relhastriggers | relhassubclass | relfrozenxid | relacl | reloptions
-------------------+--------------+----------+-----------+----------+-------+-------------+---------------+----------+-----------+---------------+---------------+-------------+-------------+----------------+---------+----------+-----------+------------+------------+-------------+----------------+----------------+--------------+--------+------------
pg_toast_16439145 | 99 | 16439149 | 0 | 10 | 0 | 16439148 | 0 | 0 | 0 | 
0 | 16439150 | t | f | p | t | 3 | 0 | f | t | f | f | f | 630449585 | |
(1 row)

But it doesn't exist in new cluster:
select * from pg_class WHERE oid=16439148;
relname | relnamespace | reltype | reloftype | relowner | relam | 
relfilenode | reltablespace | relpages | reltuples | relallvisible | 
reltoastrelid | reltoastidxid | relhasindex | relisshared | 
relpersistence | relkind | relnatts | relchecks | relhasoids | 
relhaspkey | relhasrules | relhastriggers | relhassubclass | 
relfrozenxid | relacl | reloptions
---------+--------------+---------+-----------+----------+-------+-------------+---------------+----------+-----------+---------------+---------------+---------------+-------------+-------------+----------------+---------+----------+-----------+------------+------------+-------------+----------------+----------------+--------------+--------+------------
(0 rows)