Tom Lane escribió:
> I wrote:
> > Hmm ... this doesn't really feel cleaner to me, although I'm not sure
> > why not.
>
> Oh, I thought of a more concrete point: InRecovery is inherently a
> system-wide state, but XLogInsertAllowed is *not*. While we write
> the EOR checkpoint, we really want only the bgwriter to be authorized
> to write WAL, but the scheme you propose would effectively authorize
> all processes during that window.
BTW one of the problems of the block-CRC patch was that we needed to
emit WAL records for hint bit states set during shutdown, as buffers
were flushed out, which obviously caused that error message to show up.
Due to another more serious problem I didn't investigate the solution to
this one, but it looks like it would be easy to set XLogInsertAllowed to
true while flushing buffers.
--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.