On Mon, Mar 31, 2008 at 2:10 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> "Morris Goldstein" <morris(dot)x(dot)goldstein(at)gmail(dot)com> writes:
> > Suppose I have a database with $PGDATA on /dev/sda, and a tablespace
> > directory on /dev/sdb. Will Postgres start successfully if /dev/sda is
> > mounted and /dev/sdb is not? If not, why not?
> It will start, but you will have unpleasant failures when you try to use
> tables in the secondary tablespace
Understood. I can guarantee that there will be no such accesses until
/dev/sdb is mounted.
> ... note that if autovacuum is on,
> that is likely to happen even without any explicit action on your part.
That's a good point. We don't use autovacuum, and I guess we'd have to
forego it. (Or can it be turned on and off dynamically?)
But that makes me wonder: what about this sequence of events:
- Postgres running normally on /dev/sda and /dev/sdb.
- Update to table in /dev/sdb tablespace is committed but still exists in WAL.
- Postgres crashes (e.g. power failure).
- Postgres starts with /dev/sda only.
- Recovery needs to update table in /dev/sdb tablespace.
I assume bad things will happen in this case.
In response to
pgsql-hackers by date
|Next:||From: Michael Paesold||Date: 2008-04-01 14:21:34|
|Subject: Re: [HACKERS] How embarrassing: optimization of a one-shot query doesn't work|
|Previous:||From: Tom Lane||Date: 2008-04-01 14:06:01|
|Subject: Re: [HACKERS] How embarrassing: optimization of a one-shot query doesn't work |
pgsql-general by date
|Next:||From: Tom Lane||Date: 2008-04-01 14:22:03|
|Subject: Re: [HACKERS] ANALYZE getting dead tuple count hopelessly wrong |
|Previous:||From: korry||Date: 2008-04-01 14:11:30|
|Subject: Re: Role attribute for user email?|