Обсуждение: copy / backup question - copying a data directory while the db is shutdown

Поиск
Список
Период
Сортировка

copy / backup question - copying a data directory while the db is shutdown

От
CS_DBA
Дата:
Hi All;

If I shutdown a HOT standby slave (via the -m fast command) can I then
copy the data directory to a "backup" location and be guaranteed that
the backup copy will start?

For some reason I thought there were issues (or potential issues) with
this method and a PITR based recovery approach was a better guarantee
that the copy will start


Thoughts?



Re: copy / backup question - copying a data directory while the db is shutdown

От
Jürgen Fuchsberger
Дата:
Hi,

On 03/24/2014 11:46 PM, CS_DBA wrote:
> Hi All;
>
> If I shutdown a HOT standby slave (via the -m fast command) can I then
> copy the data directory to a "backup" location and be guaranteed that
> the backup copy will start?

Yes, but you must also backup the WAL archive from which you are
recovering (in case you are using file based log shipping and not
streaming replication). For streaming replication I don't know if it works.

Also you should check if postgres really stopped, since on high load
even the fast option can fail to stop it. Look here on how to do this:
http://www.postgresql.org/message-id/53040588.9020803@coxinc.com
>
> For some reason I thought there were issues (or potential issues) with
> this method and a PITR based recovery approach was a better guarantee
> that the copy will start
>
>
> Thoughts?
>
Info about other options appreciated.
>
>

Best,
Juergen

--
| Jürgen Fuchsberger, M.Sc.
| Wegener Center for Climate and Global Change
| University of Graz



Вложения

Re: copy / backup question - copying a data directory while the db is shutdown

От
desmodemone
Дата:



2014-03-25 8:40 GMT+01:00 Jürgen Fuchsberger <juergen.fuchsberger@uni-graz.at>:
Hi,

On 03/24/2014 11:46 PM, CS_DBA wrote:
> Hi All;
>
> If I shutdown a HOT standby slave (via the -m fast command) can I then
> copy the data directory to a "backup" location and be guaranteed that
> the backup copy will start?

Yes, but you must also backup the WAL archive from which you are
recovering (in case you are using file based log shipping and not
streaming replication). For streaming replication I don't know if it works.

Also you should check if postgres really stopped, since on high load
even the fast option can fail to stop it. Look here on how to do this:
http://www.postgresql.org/message-id/53040588.9020803@coxinc.com
>
> For some reason I thought there were issues (or potential issues) with
> this method and a PITR based recovery approach was a better guarantee
> that the copy will start
>
>
> Thoughts?
>
Info about other options appreciated.
>
>

Best,
Juergen

--
| Jürgen Fuchsberger, M.Sc.
| Wegener Center for Climate and Global Change
| University of Graz



If you pause the recovery ( pg_xlog_pause )  , the process try to take a spin lock , and then pause the recovery , so it's consistent, at that point if you close the database, the consistent is reached, so I think you have no problem to have a "snapshoot" of the database.

Obviously you need also the archived checkpoint segments if you need to make a PITR.

Have a nice day

Mat Dba