Обсуждение: RAID0 and pg_xlog

"Carlos Benkendorf" <carlosbenkendorf@yahoo.com.br> wrote in message news:20050903204532.70403.qmail@web35514.mail.mud.yahoo.com...

Hi,

 

Does it make any sense to change the pg_xlog position if all my disks are configured in RAID level 0?

 

Thanks in advance!

 

Benkendorf

---

Yahoo! Messenger com voz: PROMOÇÃO VOCÊ PODE LEVAR UMA VIAGEM NA CONVERSA. Participe!

On Wed, Sep 07, 2005 at 12:47:43PM -0700, Qingqing Zhuo wrote:
> Xlog will be the only believable data if your system crashed. So it is a dangerous practice to put xlog stuff in
RAID0.

No more or less so than putting your main database on RAID0. If any
drive fails, you lose everything.
--
Jim C. Nasby, Sr. Engineering Consultant      jnasby@pervasive.com
Pervasive Software      http://pervasive.com    work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf       cell: 512-569-9461

Jim C. Nasby wrote:
>No more or less so than putting your main database on RAID0. If any
>drive fails, you lose everything.

perhaps it's time to start writing it [^r]A[^i]D 0 to try and make
the point.

richard

On Wed, 2005-09-07 at 12:40, Jim C. Nasby wrote:
> On Wed, Sep 07, 2005 at 12:47:43PM -0700, Qingqing Zhuo wrote:
> > Xlog will be the only believable data if your system crashed. So it is a dangerous practice to put xlog stuff in
RAID0.
>
> No more or less so than putting your main database on RAID0. If any
> drive fails, you lose everything.

Sounds like a good place to have replication.

On Wed, Sep 07, 2005 at 01:02:18PM -0500, Scott Marlowe wrote:
> On Wed, 2005-09-07 at 12:40, Jim C. Nasby wrote:
> > On Wed, Sep 07, 2005 at 12:47:43PM -0700, Qingqing Zhuo wrote:
> > > Xlog will be the only believable data if your system crashed. So it is a dangerous practice to put xlog stuff in
RAID0.
> >
> > No more or less so than putting your main database on RAID0. If any
> > drive fails, you lose everything.
>
> Sounds like a good place to have replication.

If you used syncronous replication, maybe. Otherwise failure of any
drive means you just lost data. And remember that the more drives you
have in your array the more likely you'll have a failure in a given
time period.

Basically, if you can afford to setup replication on 2 machines with
RAID0 you can afford to setup RAID10 on one machine, which will usually
be a better bet.
--
Jim C. Nasby, Sr. Engineering Consultant      jnasby@pervasive.com
Pervasive Software      http://pervasive.com    work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf       cell: 512-569-9461

On Wed, 2005-09-07 at 16:15, Jim C. Nasby wrote:
> On Wed, Sep 07, 2005 at 01:02:18PM -0500, Scott Marlowe wrote:
> > On Wed, 2005-09-07 at 12:40, Jim C. Nasby wrote:
> > > On Wed, Sep 07, 2005 at 12:47:43PM -0700, Qingqing Zhuo wrote:
> > > > Xlog will be the only believable data if your system crashed. So it is a dangerous practice to put xlog stuff
inRAID0. 
> > >
> > > No more or less so than putting your main database on RAID0. If any
> > > drive fails, you lose everything.
> >
> > Sounds like a good place to have replication.
>
> If you used syncronous replication, maybe. Otherwise failure of any
> drive means you just lost data. And remember that the more drives you
> have in your array the more likely you'll have a failure in a given
> time period.
>
> Basically, if you can afford to setup replication on 2 machines with
> RAID0 you can afford to setup RAID10 on one machine, which will usually
> be a better bet.

Yeah, I was thinking pgpool here.

On Fri, Sep 09, 2005 at 09:43:56AM -0500, Scott Marlowe wrote:
> On Wed, 2005-09-07 at 16:15, Jim C. Nasby wrote:
> > On Wed, Sep 07, 2005 at 01:02:18PM -0500, Scott Marlowe wrote:
> > > On Wed, 2005-09-07 at 12:40, Jim C. Nasby wrote:
> > > > On Wed, Sep 07, 2005 at 12:47:43PM -0700, Qingqing Zhuo wrote:
> > > > > Xlog will be the only believable data if your system crashed. So it is a dangerous practice to put xlog stuff
inRAID0. 
> > > >
> > > > No more or less so than putting your main database on RAID0. If any
> > > > drive fails, you lose everything.
> > >
> > > Sounds like a good place to have replication.
> >
> > If you used syncronous replication, maybe. Otherwise failure of any
> > drive means you just lost data. And remember that the more drives you
> > have in your array the more likely you'll have a failure in a given
> > time period.
> >
> > Basically, if you can afford to setup replication on 2 machines with
> > RAID0 you can afford to setup RAID10 on one machine, which will usually
> > be a better bet.
>
> Yeah, I was thinking pgpool here.

pgpool is a connection pool; it has (almost) nothing to do with
replication. It certainly doesn't work to provide any kind of data
security on a RAID0 setup.

I'm not arguing against anything people have suggested, only pointing
out that if you're using RAID0 your data is not safe against a drive
failure, except possible using pgcluster (some would argue that
statement-based replication isn't as reliable as log-based).
--
Jim C. Nasby, Sr. Engineering Consultant      jnasby@pervasive.com
Pervasive Software      http://pervasive.com    work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf       cell: 512-569-9461

On Fri, 2005-09-09 at 18:16, Jim C. Nasby wrote:
> On Fri, Sep 09, 2005 at 09:43:56AM -0500, Scott Marlowe wrote:
> > On Wed, 2005-09-07 at 16:15, Jim C. Nasby wrote:
> > > On Wed, Sep 07, 2005 at 01:02:18PM -0500, Scott Marlowe wrote:
> > > > On Wed, 2005-09-07 at 12:40, Jim C. Nasby wrote:
> > > > > On Wed, Sep 07, 2005 at 12:47:43PM -0700, Qingqing Zhuo wrote:
> > > > > > Xlog will be the only believable data if your system crashed. So it is a dangerous practice to put xlog
stuffin RAID0. 
> > > > >
> > > > > No more or less so than putting your main database on RAID0. If any
> > > > > drive fails, you lose everything.
> > > >
> > > > Sounds like a good place to have replication.
> > >
> > > If you used syncronous replication, maybe. Otherwise failure of any
> > > drive means you just lost data. And remember that the more drives you
> > > have in your array the more likely you'll have a failure in a given
> > > time period.
> > >
> > > Basically, if you can afford to setup replication on 2 machines with
> > > RAID0 you can afford to setup RAID10 on one machine, which will usually
> > > be a better bet.
> >
> > Yeah, I was thinking pgpool here.
>
> pgpool is a connection pool; it has (almost) nothing to do with
> replication. It certainly doesn't work to provide any kind of data
> security on a RAID0 setup.
>
> I'm not arguing against anything people have suggested, only pointing
> out that if you're using RAID0 your data is not safe against a drive
> failure, except possible using pgcluster (some would argue that
> statement-based replication isn't as reliable as log-based).

Um.  No.  It has a synchronous replication mode, which I've used, and it
works quite well.

Look it up, it's pretty cool.  Writes to both pg machines synchronously,
reads from them load balanced.  Of course, there are some limits imposed
by this methodology, re: things like random() and such.

Now, if you're arguing against statement based replication, that I can
understand.  but pgpool can definitely do two box sync replication.

On Fri, Sep 09, 2005 at 06:20:21PM -0500, Scott Marlowe wrote:
> > pgpool is a connection pool; it has (almost) nothing to do with
> > replication. It certainly doesn't work to provide any kind of data
> > security on a RAID0 setup.
> >
> > I'm not arguing against anything people have suggested, only pointing
> > out that if you're using RAID0 your data is not safe against a drive
> > failure, except possible using pgcluster (some would argue that
> > statement-based replication isn't as reliable as log-based).
>
> Um.  No.  It has a synchronous replication mode, which I've used, and it
> works quite well.
>
> Look it up, it's pretty cool.  Writes to both pg machines synchronously,
> reads from them load balanced.  Of course, there are some limits imposed
> by this methodology, re: things like random() and such.
>
> Now, if you're arguing against statement based replication, that I can
> understand.  but pgpool can definitely do two box sync replication.

Oh, I didn't realize that. Though I have to wonder why they duplicated
what pgcluster provides...
--
Jim C. Nasby, Sr. Engineering Consultant      jnasby@pervasive.com
Pervasive Software      http://pervasive.com    work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf       cell: 512-569-9461

On Fri, 2005-09-09 at 18:54, Jim C. Nasby wrote:
> On Fri, Sep 09, 2005 at 06:20:21PM -0500, Scott Marlowe wrote:
> > > pgpool is a connection pool; it has (almost) nothing to do with
> > > replication. It certainly doesn't work to provide any kind of data
> > > security on a RAID0 setup.
> > >
> > > I'm not arguing against anything people have suggested, only pointing
> > > out that if you're using RAID0 your data is not safe against a drive
> > > failure, except possible using pgcluster (some would argue that
> > > statement-based replication isn't as reliable as log-based).
> >
> > Um.  No.  It has a synchronous replication mode, which I've used, and it
> > works quite well.
> >
> > Look it up, it's pretty cool.  Writes to both pg machines synchronously,
> > reads from them load balanced.  Of course, there are some limits imposed
> > by this methodology, re: things like random() and such.
> >
> > Now, if you're arguing against statement based replication, that I can
> > understand.  but pgpool can definitely do two box sync replication.
>
> Oh, I didn't realize that. Though I have to wonder why they duplicated
> what pgcluster provides...

I doubt it's as good as pgcluster.  It's simple dual machine sync
replication.  I think it was a case of being 95% there when the pooling
part was done, so why not just toss in replication for good measure.

> > On Fri, Sep 09, 2005 at 06:20:21PM -0500, Scott Marlowe wrote:
> > > > pgpool is a connection pool; it has (almost) nothing to do with
> > > > replication. It certainly doesn't work to provide any kind of data
> > > > security on a RAID0 setup.
> > > >
> > > > I'm not arguing against anything people have suggested, only pointing
> > > > out that if you're using RAID0 your data is not safe against a drive
> > > > failure, except possible using pgcluster (some would argue that
> > > > statement-based replication isn't as reliable as log-based).
> > >
> > > Um.  No.  It has a synchronous replication mode, which I've used, and it
> > > works quite well.
> > >
> > > Look it up, it's pretty cool.  Writes to both pg machines synchronously,
> > > reads from them load balanced.  Of course, there are some limits imposed
> > > by this methodology, re: things like random() and such.
> > >
> > > Now, if you're arguing against statement based replication, that I can
> > > understand.  but pgpool can definitely do two box sync replication.
> >
> > Oh, I didn't realize that. Though I have to wonder why they duplicated
> > what pgcluster provides...
>
> I doubt it's as good as pgcluster.  It's simple dual machine sync
> replication.  I think it was a case of being 95% there when the pooling
> part was done, so why not just toss in replication for good measure.

As a developer of pgpool I have to admit above. If it's technically
possible, why not chalenge it? That's a nature of an engineer:-)

BTW, some codes(probably the connection pooling part) in PGCluster
have been copied from pgpool. Instead I have been getting good
feedbacks/bug fixes from the author of PGCluster. It is kind of a
collaboration work, and I think this is one of the greatest thing with
open source softwares.
--
SRA OSS, Inc. Japan
Tatsuo Ishii

> On Fri, 2005-09-09 at 18:54, Jim C. Nasby wrote:
> > On Fri, Sep 09, 2005 at 06:20:21PM -0500, Scott Marlowe wrote:
> > > > pgpool is a connection pool; it has (almost) nothing to do with
> > > > replication. It certainly doesn't work to provide any kind of data
> > > > security on a RAID0 setup.
> > > >
> > > > I'm not arguing against anything people have suggested, only pointing
> > > > out that if you're using RAID0 your data is not safe against a drive
> > > > failure, except possible using pgcluster (some would argue that
> > > > statement-based replication isn't as reliable as log-based).
> > >
> > > Um.  No.  It has a synchronous replication mode, which I've used, and it
> > > works quite well.
> > >
> > > Look it up, it's pretty cool.  Writes to both pg machines synchronously,
> > > reads from them load balanced.  Of course, there are some limits imposed
> > > by this methodology, re: things like random() and such.
> > >
> > > Now, if you're arguing against statement based replication, that I can
> > > understand.  but pgpool can definitely do two box sync replication.
> >
> > Oh, I didn't realize that. Though I have to wonder why they duplicated
> > what pgcluster provides...
>
> I doubt it's as good as pgcluster.  It's simple dual machine sync
> replication.  I think it was a case of being 95% there when the pooling
> part was done, so why not just toss in replication for good measure.

As a developer of pgpool I have to admit above. If it's technically
possible, why not chalenge it? That's a nature of an engineer:-)

BTW, some codes(probably the connection pooling part) in PGCluster
have been copied from pgpool. Instead I have been getting good
feedbacks/bug fixes from the author of PGCluster. It is kind of a
collaboration work, and I think this is one of the greatest thing with
open source softwares.
--
SRA OSS, Inc. Japan
Tatsuo Ishii