Re: Fix of fake unlogged LSN initialization

On Mon, 21 Oct 2019 at 06:03, Michael Paquier <michael@paquier.xyz> wrote:

> On Sat, Oct 19, 2019 at 05:03:00AM +0000, tsunakawa.takay@fujitsu.com
> wrote:
> > The attached trivial patch fixes the initialization of the fake
> > unlogged LSN.  Currently, BootstrapXLOG() in initdb sets the initial
> > fake unlogged LSN to FirstNormalUnloggedLSN (=1000), but the
> > recovery and pg_resetwal sets it to 1.  The patch modifies the
> > latter two cases to match initdb.
> >
> > I don't know if this do actual harm, because the description of
> > FirstNormalUnloggedLSN doesn't give me any idea.
>
> From xlogdefs.h added by 9155580:
> /*
>  * First LSN to use for "fake" LSNs.
>  *
>  * Values smaller than this can be used for special per-AM purposes.
>  */
> #define FirstNormalUnloggedLSN  ((XLogRecPtr) 1000)
>
> So it seems to me that you have caught a bug here, and that we had
> better back-patch to v12 so as recovery and pg_resetwal don't mess up
> with AMs using lower values than that.
>

I wonder why is that value 1000, rather than an aligned value or a whole
WAL page?

-- 
Simon Riggs                http://www.2ndQuadrant.com/
<http://www.2ndquadrant.com/>
PostgreSQL Solutions for the Enterprise