On Tue, Feb 1, 2011 at 12:41 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>> So I'm back to proposing that we just apply FPI-free WAL records
>> unconditionally, without regard to the LSN. This could potentially
>> corrupt the page, of course.
> Yes. So you're still assuming that there will be a later FPI-containing
> WAL record to fix up the mess you created. What if there isn't?
In that case, the page shouldn't be corrupted. The possibility of
corruption comes from the fact that a future WAL record might
rearrange the page contents so that the current WAL record is no
longer applying to the set of tuples it expects to be seeing. But any
such action would necessarily induce an FPI. If there is no such
action, then how can the page get into a state where replaying a heap
delete will corrupt it? For that to happen, the item pointer list has
to have changed.
The Enterprise PostgreSQL Company
In response to
- Re: FPI at 2011-02-01 17:41:38 from Tom Lane
pgsql-hackers by date
|Next:||From: Jeff Davis||Date: 2011-02-01 17:51:04|
|Subject: Re: SSI patch version 14|
|Previous:||From: Robert Haas||Date: 2011-02-01 17:44:05|
|Subject: Re: Spread checkpoint sync|