Re: pg_dump bug in 9.4beta2 and HEAD

From: Heikki Linnakangas <hlinnakangas(at)vmware(dot)com>
To: Thom Brown <thom(at)linux(dot)com>, Pavel Stehule <pavel(dot)stehule(at)gmail(dot)com>
Cc: Joachim Wieland <joe(at)mcknight(dot)de>, PostgreSQL Hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: pg_dump bug in 9.4beta2 and HEAD
Date: 2014-09-24 10:59:17
Message-ID: 5422A405.3060509@vmware.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 09/24/2014 01:50 PM, Thom Brown wrote:
> On 15 August 2014 16:31, Pavel Stehule <pavel(dot)stehule(at)gmail(dot)com> wrote:
>
>> 2014-08-14 9:03 GMT+02:00 Heikki Linnakangas <hlinnakangas(at)vmware(dot)com>:
>>
>>> On 08/14/2014 06:53 AM, Joachim Wieland wrote:
>>>
>>>> I'm seeing an assertion failure with "pg_dump -c --if-exists" which is
>>>> not ready to handle BLOBs it seems:
>>>>
>>>> pg_dump: pg_backup_archiver.c:472: RestoreArchive: Assertion `mark !=
>>>> ((void *)0)' failed.
>>>>
>>>> To reproduce:
>>>>
>>>> $ createdb test
>>>> $ pg_dump -c --if-exists test (works, dumps empty database)
>>>> $ psql test -c "select lo_create(1);"
>>>> $ pg_dump -c --if-exists test (fails, with the above mentioned
>>>> assertion)
>>>>
>>>
>>> The code tries to inject an "IF EXISTS" into the already-construct DROP
>>> command, but it doesn't work for large objects, because the deletion
>>> command looks like "SELECT pg_catalog.lo_unlink(xxx)". There is no DROP
>>> there.
>>>
>>> I believe we could use "SELECT pg_catalog.lo_unlink(loid) FROM
>>> pg_catalog.pg_largeobject_metadata WHERE loid = xxx".
>>> pg_largeobject_metadata table didn't exist before version 9.0, but we don't
>>> guarantee pg_dump's output to be compatible in that direction anyway, so I
>>> think that's fine.
>>>
>>> The quick fix would be to add an exception for blobs, close to where
>>> Assert is. There are a few exceptions there already. A cleaner solution
>>> would be to add a new argument to ArchiveEntry and make the callers
>>> responsible for providing an "DROP IF EXISTS" query, but that's not too
>>> appetizing because for most callers it would be boring boilerplate code.
>>> Perhaps add an argument, but if it's NULL, ArchiveEntry would form the
>>> if-exists query automatically from the DROP query.
>>
>> I am sending two patches
>>
>> first is fast fix
>>
>> second fix is implementation of Heikki' idea.
>
> I'm guessing this issue is still unresolved? It would be nice to get this
> off the open items list.

Yeah, I had completely forgotten about this. Alvaro, could you finish
this off?

- Heikki

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Heikki Linnakangas 2014-09-24 11:03:41 Re: make pg_controldata accept "-D dirname"
Previous Message Abhijit Menon-Sen 2014-09-24 10:59:07 make pg_controldata accept "-D dirname"