I'd like to apply the patch to HEAD and previous releases because
the issue seems to be a bug in the core. Any comments or objections?
Some users actually use STOP WAL LOCATION in their backup script,
and they've countered the bug with 1/256 probability in recent days.
Fujii Masao <masao.fujii@gmail.com> wrote:
> On Fri, Feb 5, 2010 at 9:08 AM, Takahiro Itagaki
> <itagaki.takahiro@oss.ntt.co.jp> wrote:
> >
> > Fujii Masao <masao.fujii@gmail.com> wrote:
> >
> >> On Fri, Dec 5, 2008 at 11:41 PM, Randy Isbell <jisbell@cisco.com> wrote:
> >> > An inconsistency exists between the segment name reported by
> >> > pg_stop_backup() and the actual WAL file name.
> >> >
> >> > START WAL LOCATION: 10/FE1E2BAC (file 0000000200000010000000FE)
> >> > STOP WAL LOCATION: 10/FF000000 (file 0000000200000010000000FF)
> >
> >> But it was rejected because its change might break the existing app.
> >
> > It might break existing applications if it returns "FE" instead of "FF",
> > but never-used filename surprises users. (IMO, the existing apps probably
> > crash if "FF" returned, i.e, 1/256 of the time.)
> >
> > Should it return the *next* reasonable log filename instead of "FF"?
> > For example, 000000020000002000000000 for the above case.
>
> Here is the patch that avoids a nonexistent file name, according to
> Itagaki-san's suggestion. If we are crossing a logid boundary, the
> next reasonable file name is used instead of a nonexistent one.
Regards,
---
Takahiro Itagaki
NTT Open Source Software Center
