Thread: File system level copy

Hao Wang wrote:
> I installed PostgresSQL-8.3 on my linux machine.
>
> The cluster directory is /usr/local/data and I created three databases
named db1, db2, and db3. db1 is
> in the default tablespace 'pg_default'. db2 is in
'/home/tablespace/space1/' and db3 is in
> '/home/tablespace/space2/'.  I want to copy the cluster directory and
the db3  tablespace
> folder('/home/tablespace/space2/') without stopping the database
server. Then I want to use the
> cluster directory and db3's tablespace in another linux machine to
recover 'db3' database. Does this
> way work? If not, why?

First, you need a correct backup for recovery.
Before copying, run pg_start_backup, and pg_stop_backup afterwards.

Then you need to have recovery.conf and WAL archives
(or be lucky and all WALs are still in pg_xlog).

WAL contains changes to all databases in the cluster, so
you cannot recover only one database, you'll have to
recover them all.

Read
http://www.postgresql.org/docs/current/static/continuous-archiving.html
for background and details.

Yours,
Laurenz Albe

This is PITR, right?
I don't want to use this way because I'm not allowed to change the configuration parameter of database server. I just
wantto use some whole DB copy to restore db3 in another machine. And I don't want to use pg_dump because I think db3 is
solarge that pg_dump will probably have bad performance. 

-----Original Message-----
From: Albe Laurenz [mailto:laurenz.albe@wien.gv.at]
Sent: Wednesday, November 14, 2012 6:49 PM
To: Wang, Hao; pgsql-general@postgresql.org
Subject: RE: [GENERAL] File system level copy

Hao Wang wrote:
> I installed PostgresSQL-8.3 on my linux machine.
>
> The cluster directory is /usr/local/data and I created three databases
named db1, db2, and db3. db1 is
> in the default tablespace 'pg_default'. db2 is in
'/home/tablespace/space1/' and db3 is in
> '/home/tablespace/space2/'.  I want to copy the cluster directory and
the db3  tablespace
> folder('/home/tablespace/space2/') without stopping the database
server. Then I want to use the
> cluster directory and db3's tablespace in another linux machine to
recover 'db3' database. Does this
> way work? If not, why?

First, you need a correct backup for recovery.
Before copying, run pg_start_backup, and pg_stop_backup afterwards.

Then you need to have recovery.conf and WAL archives (or be lucky and all WALs are still in pg_xlog).

WAL contains changes to all databases in the cluster, so you cannot recover only one database, you'll have to recover
themall. 

Read
http://www.postgresql.org/docs/current/static/continuous-archiving.html
for background and details.

Yours,
Laurenz Albe

Hao Wang wrote:
>>> I installed PostgresSQL-8.3 on my linux machine.
>>>
>>> The cluster directory is /usr/local/data and I created three
databases
>>> named db1, db2, and db3. db1 is
>>> in the default tablespace 'pg_default'. db2 is in
>>> '/home/tablespace/space1/' and db3 is in
>>> '/home/tablespace/space2/'.  I want to copy the cluster directory
and
>>> the db3  tablespace
>>> folder('/home/tablespace/space2/') without stopping the database
>>> server. Then I want to use the
>>> cluster directory and db3's tablespace in another linux machine to
>>> recover 'db3' database. Does this
>>> way work? If not, why?
>>
>> First, you need a correct backup for recovery.
>> Before copying, run pg_start_backup, and pg_stop_backup afterwards.
>>
>> Then you need to have recovery.conf and WAL archives (or be lucky and
all WALs are still in pg_xlog).
>>
>> WAL contains changes to all databases in the cluster, so you cannot
recover only one database, you'll
>> have to recover them all.
>>
>> Read
>>
http://www.postgresql.org/docs/current/static/continuous-archiving.html
>> for background and details.

> This is PITR, right?
> I don't want to use this way because I'm not allowed to change the
configuration
> parameter of database server. I just want to use some whole DB copy to
restore
> db3 in another machine. And I don't want to use pg_dump because I
think db3
> is so large that pg_dump will probably have bad performance.

That's a whole lot of arbitrary restrictions.

If all you want is a copy of the database, pg_dump is what
you should use.  Besides, it is the only way to get a copy
of just one database.  What's the problem if pg_dump takes
a few hours or days (I don't know how big you DB is)?

A side thought: if the DB is not configured for PITR and
pg_dump takes too long, how do you perform your backups?

Yours,
Laurenz Albe

My purpose is not to do backup for my database. I just want to copy the whole db3 database to another machine and
restoreit. That database could be very large so I think directly copy is more efficient than pg_dump.  So I'd like to
dosome test to see if this way works. If it doesn't work, I will consider to use pg_dump.   
Thank you for your feedback.

-----Original Message-----
From: pgsql-general-owner@postgresql.org [mailto:pgsql-general-owner@postgresql.org] On Behalf Of Albe Laurenz
Sent: Thursday, November 15, 2012 4:52 PM
To: Wang, Hao; pgsql-general@postgresql.org
Subject: Re: [GENERAL] File system level copy

Hao Wang wrote:
>>> I installed PostgresSQL-8.3 on my linux machine.
>>>
>>> The cluster directory is /usr/local/data and I created three
databases
>>> named db1, db2, and db3. db1 is
>>> in the default tablespace 'pg_default'. db2 is in
>>> '/home/tablespace/space1/' and db3 is in '/home/tablespace/space2/'.
>>> I want to copy the cluster directory
and
>>> the db3  tablespace
>>> folder('/home/tablespace/space2/') without stopping the database
>>> server. Then I want to use the cluster directory and db3's
>>> tablespace in another linux machine to recover 'db3' database. Does
>>> this way work? If not, why?
>>
>> First, you need a correct backup for recovery.
>> Before copying, run pg_start_backup, and pg_stop_backup afterwards.
>>
>> Then you need to have recovery.conf and WAL archives (or be lucky and
all WALs are still in pg_xlog).
>>
>> WAL contains changes to all databases in the cluster, so you cannot
recover only one database, you'll
>> have to recover them all.
>>
>> Read
>>
http://www.postgresql.org/docs/current/static/continuous-archiving.html
>> for background and details.

> This is PITR, right?
> I don't want to use this way because I'm not allowed to change the
configuration
> parameter of database server. I just want to use some whole DB copy to
restore
> db3 in another machine. And I don't want to use pg_dump because I
think db3
> is so large that pg_dump will probably have bad performance.

That's a whole lot of arbitrary restrictions.

If all you want is a copy of the database, pg_dump is what you should use.  Besides, it is the only way to get a copy
ofjust one database.  What's the problem if pg_dump takes a few hours or days (I don't know how big you DB is)? 

A side thought: if the DB is not configured for PITR and pg_dump takes too long, how do you perform your backups?

Yours,
Laurenz Albe


--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

Hao Wang wrote:

> My purpose is not to do backup for my database.

I understood that.  It was just a side comment.

> I just want to copy the whole db3 database to another
> machine and restore it. That database could be very large so I think
directly copy is more efficient
> than pg_dump.  So I'd like to do some test to see if this way works.
If it doesn't work, I will
> consider to use pg_dump.

Any attempt to make a file system copy of a live system without
doing PITR will very likely cause data corruption.

You'll have to use pg_dump.

Yours,
Laurenz Albe