Re: Recovery Features

From: Jan Wieck <JanWieck(at)Yahoo(dot)com>
To: Simon Riggs <simon(at)2ndquadrant(dot)com>
Cc: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>, pgsql-hackers(at)postgresql(dot)org
Subject: Re: Recovery Features
Date: 2004-07-10 14:04:48
Message-ID: 40EFF780.30009@Yahoo.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 7/5/2004 6:16 PM, Simon Riggs wrote:

> On Mon, 2004-07-05 at 22:30, Tom Lane wrote:
>> Simon Riggs <simon(at)2ndquadrant(dot)com> writes:
>> > ...While recovering, it is very straightforward to simply ignore every
>> > record associated with one (or more) transactions. That gives us the
>> > ability to recover "all apart from txnid X".
>>
>> Don't even *think* of going there.
>
> Hmmm... thinking is important, as are differing viewpoints. I value
> yours and those of everyone else on this list, hence the post.
>
>> What will happen when transaction Y comes along and wants to modify or
>> delete a row that was inserted by X? There's no chance of staying
>> consistent.
>
> I did point out this downside...a few sentences down.
> **This is awful because: transactions are isolated from each other, but
> they also provide changes of state that rely on previous committed
> transactions. If you change the past, you could well invalidate the
> future. If you blow away a transaction and a later one depends upon it,
> then you will have broken the recovery chain and will not be able to
> recover to present time.**
>
> Theoretically, this is a disaster area.
>
> Practically, Oracle10g provides similar-ish features...

IF ... the recovery process would be primary key based, and IF the
database definitions would allow for balance type field handling (the
log contains value deltas for balance fields instead of overwriting
them), THEN this would be a direction I would be looking into.

But as things are, the whole recovery is ctid and binary block based. So
you would now leave out the ctid based changes to several tuples because
of belonging to said transaction. Later on, an original whole block
appears in the WAL and overwrites ... so you get what ... partial
transactions into the recoverd DB?

>
> ...Nobody is shouting YES, so its a dodo...

No way!

>
> Best regards, Simon Riggs
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 8: explain analyze is your friend

--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#================================================== JanWieck(at)Yahoo(dot)com #

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Jan Wieck 2004-07-10 14:17:40 Re: Point in Time Recovery
Previous Message Andrew Dunstan 2004-07-10 13:05:39 Re: plperl vs. plperlu