Thread: How does the WAL work? Need to recovery the database to a specific point of time

How does the WAL work? Need to recovery the database to a specific point of time

From

"Wang, Mary Y"

Date:

22 April 2014, 15:40:45

Hi All,

I’d like to recovery the database to a specific point of time. I see some WAL files in the pg_xlog directories.

I read some blogs, and some have indicated that a recovery.conf needs to be updated. I do not see a recovery.conf file in the $PG_DATA directory ( I see pg_hba.conf and postgresql.conf).

1. Do I need to create a recovery.conf file?

2. What specific settings should I set?

3. How do I ensure that postgres will read the recovery.conf file?

4. What is the start/stop postgres command to use when working with a recovery.conf file?

I am running on Postgres 8.4.11 and RHEL 6.4.

Thanks

Mary Wang

Re: How does the WAL work? Need to recovery the database to a specific point of time

From

Steve Crawford

Date:

22 April 2014, 16:00:49

On 04/22/2014 08:40 AM, Wang, Mary Y wrote:

Hi All,

I’d like to recovery the database to a specific point of time. I see some WAL files in the pg_xlog directories.
I read some blogs, and some have indicated that a recovery.conf needs to be updated. I do not see a recovery.conf file in the $PG_DATA directory ( I see pg_hba.conf and postgresql.conf).

1.      Do I need to create a recovery.conf file?
2.      What specific settings should I set?
3.      How do I ensure that postgres will read the recovery.conf file?
4.      What is the start/stop postgres command to use when working with a recovery.conf file?

I am running on Postgres 8.4.11 and RHEL 6.4.

Point-in-time recovery is not something that can typically be performed arbitrarily "after the fact" - you need to have established the necessary backup and archiving in advance.

This capability is essentially a useful side-effect of the process through which PostgreSQL recovers from an inconsistent state following a power-failure or other disruptive event by replaying the appropriate WAL files.

Point-in-time recovery assumes you have both a *file-system* backup (not a pg_dump) of the PostgreSQL data files that was completed prior to the point in time to which you wish to recover *and* all the write-ahead log (WAL) files created from when the file-system was backed-up and the point to which you wish to recover. This is typically set up through a combination of scheduled file-system backups and archiving of WAL files.

WAL files are not stored indefinitely but rather are rotated and optionally archived based on your configuration settings.

If you have the base file-system backup and all the required WAL files, have a look at http://www.postgresql.org/docs/8.4/static/continuous-archiving.html

Cheers,
Steve

Re: How does the WAL work? Need to recovery the database to a specific point of time

From

Matheus de Oliveira

Date:

22 April 2014, 16:05:40

On Tue, Apr 22, 2014 at 12:40 PM, Wang, Mary Y <mary.y.wang@boeing.com> wrote:

Hi All,

I’d like to recovery the database to a specific point of time. I see some WAL files in the pg_xlog directories.

If you haven't setup incremental backups (basebackup + archiving) you can't use PITR feature now. You really should have done such setup before.

I read some blogs, and some have indicated that a recovery.conf needs to be updated. I do not see a recovery.conf file in the $PG_DATA directory ( I see pg_hba.conf and postgresql.conf).

1. Do I need to create a recovery.conf file?

Yes. It should not exactly be "updated", but "created". And you do that not in the production system, but in the one you want to do the recover.

2. What specific settings should I set?

There are plenty, for PITR you will need at least restore_command and one of recovery_*. See [1].

3. How do I ensure that postgres will read the recovery.conf file?

4. What is the start/stop postgres command to use when working with a recovery.conf file?

If the file exists when you start PostgreSQL, it will read and use it. So it is basically the existence of the file that will make PG use.

I recommend you reading through the documentation for "Continuous Archiving and Point-in-Time Recovery (PITR)" at [2]. After reading all there, you can come back here if you still have doubts.

I am running on Postgres 8.4.11 and RHEL 6.4.

You should upgrade ASAP to 8.4.21 and plan an upgrade to a more recent version (like 9.3), as 8.4 will be out of support soon. See [3].

Thanks
Mary Wang

[1] http://www.postgresql.org/docs/current/static/recovery-target-settings.html
[2] http://www.postgresql.org/docs/current/static/continuous-archiving.html
[3] http://www.postgresql.org/support/versioning/

--
Matheus de Oliveira
Analista de Banco de Dados
Dextra Sistemas - MPS.Br nível F!
www.dextra.com.br/postgres

Re: How does the WAL work? Need to recovery the database to a specific point of time

From

Simon Riggs

Date:

22 April 2014, 16:07:18

On 22 April 2014 16:40, Wang, Mary Y <mary.y.wang@boeing.com> wrote:
> Hi All,
>
>
>
> I’d like to recovery the database to a specific point of time. I see some
> WAL files in the pg_xlog directories.
>
> I read some blogs, and some have indicated that a recovery.conf needs to be
> updated.  I do not see a recovery.conf file in the $PG_DATA directory ( I
> see pg_hba.conf and postgresql.conf).
>
>
>
> 1.       Do I need to create a recovery.conf file?
>
> 2.       What specific settings should I set?
>
> 3.       How do I ensure that postgres will read the recovery.conf file?
>
> 4.       What is the start/stop postgres command to use when working with a
> recovery.conf file?
>
>
>
> I am running on Postgres 8.4.11 and RHEL 6.4.

Point in Time Recovery is supported, though you have to have done some
planning beforehand. The settings and commands you'll use depend upon
the exact nature of your problem and the backups and other file copies
you possess.

http://www.postgresql.org/docs/devel/static/continuous-archiving.html
etc

--
 Simon Riggs                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services