Re: changes "during checkpointing" - Mailing list pgsql-general

From Jeff Janes
Subject Re: changes "during checkpointing"
Date
Msg-id CAMkU=1xpO9HV=vhCqAsVQi3boh3E_4=ESj7qNW7_7phws4kH6A@mail.gmail.com
Whole thread Raw
In response to changes "during checkpointing"  ("Sahagian, David" <david.sahagian@emc.com>)
List pgsql-general
On Fri, Jan 11, 2013 at 1:04 PM, Sahagian, David <david.sahagian@emc.com> wrote:
> In regards to 9.1.x, I would like to learn some details of the nature of
> “checkpointing”
>
> === Question 1 ===
>
> - page 123 is dirty
>
> - "checkpointing" starts
>
> - page 123 gets written to disk, as part of this checkpoint
>
> - page 123 gets modified again
>
>   ? Does it get written to disk again, as part of this checkpoint?

No, the checkpointer doesn't write it again.

>
> === Question 2 ===
>
> - page 123 is dirty
>
> - "checkpointing" starts
>
> - page 123 gets modified again
>
> - page 123 gets written to disk, as part of this checkpoint
>
>   ? So does the most recent mod get written to disk, even if that mod is not
> committed yet ?

Yes, it the most recent version that gets written.  The notion of
"committed' does not really apply to pages.

> === Question 3 ===
>
>
>
> When does the full-page-writing to WAL happen ?
>
> Is it after the start of the "checkpointing" or after the finish of the
> "checkpointing" ?
>

The start.

Cheers,

Jeff


pgsql-general by date:

Previous
From: Alban Hertroys
Date:
Subject: Re: reducing number of ANDs speeds up query
Next
From: "Rhys A.D. Stewart"
Date:
Subject: syntax error in a trigger. can't find it.