Re: Bug: Unreferenced temp tables disables vacuum to update xid

"Gregory Stark" <stark(at)enterprisedb(dot)com> writes:

> "Joshua D. Drake" <jd(at)commandprompt(dot)com> writes:
>
>> On Mon, 7 Jan 2008 11:58:29 -0500 (EST)
>> Bruce Momjian <bruce(at)momjian(dot)us> wrote:
>>
>>> I assume it is this TODO item:
>>>
>>> o Prevent long-lived temporary tables from causing frozen-xid
>>> advancement starvation
>>>
>>> The problem is that autovacuum cannot vacuum them to set
>>> frozen xids; only the session that created them can do that.
>>> http://archives.postgresql.org/pgsql-general/2007-06/msg01645.php
>>>
>>> but am confused how the fix worked. Have all of these backends been
>>> active for 1 billion transactions?
>>
>> Well it certainly appears that the TODO item is related. However there
>> are a couple of differences.
>
> Yeah, it seems there are two bugs here. 1) temporary tables prevent frozen-xid
> advancement and 2) if a process dies at the wrong moment it's possible to
> temporary tables. Either one alone is pretty minor but I guess the combination
> is lethal.

oops, "2) if a process dies at the wrong moment it's possible to *leak*
temporary tables"

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
Ask me about EnterpriseDB's On-Demand Production Tuning