[Postgres-xl-developers] Backup and recovery

Joakim Lundgren lundgren700 at gmail.com
Fri Aug 14 01:02:15 PDT 2015

Thanks Pavan!

If I understand correctly, creating a barrier is only needed if you want to
recover to a specific PITR, set by the given *barrier_name, *but it secures
that any cluster wide transactions will be consistent when recovering. If a
barrier is not created, recovery is done until the time when the database
crashed, which is what I want to achive, but in that case cluster wide
transactions may be inconsistent. Do I get that right?

If my understanding is right then, there is no secure way to recover until
"crash time" knowing that no transaction will be inconsistent? How should
one do to achive recovery "almost to the time" when the database crashes?
Is that to set barriers regularly and live with the facy that some
transactions might be lost?

/Regards Joakim

2015-08-13 15:53 GMT+02:00 Pavan Deolasee <pavan.deolasee at gmail.com>:

> On Tue, Aug 11, 2015 at 2:17 PM, Joakim Lundgren <lundgren700 at gmail.com>
> wrote:
>> Hi!
>> I'm struggling some with backup and recovery of a postgres-XL cluster
>> over several nodes. I have read the manual and are using pg_basebackup to
>> do a hot backup.
>> My question is if there are any best practices how to do a full backup
>> and recovery of a cluster. Do I need to pg_basesetup on each node storing
>> data or is there any script that do the full backup on each node?
> Yes, you need to backup each node individually. There had been some
> discussion about either extending pgxc_ctl or extending one of the other
> backup/recovery utility to do cluster wide backup. But AFAIK there is
> nothing in place yet.
>> If you run pg_basesetup on each node, do they have to be synchronized in
>> some way.
> There is a concept of BARRIER which can be used to create cluster-wide
> point to which each component can recovery and guarantee global consistency.
> Barriers can then be used in recovery.conf.  Each node is supposed to
> recover to the same barrier. Also please keep in mind that the cluster does
> not provide any mechanism to guarantee uniqueness of barrier names. Hope
> this helps.
> Thanks,
> Pavan
> --
>  Pavan Deolasee                   http://www.2ndQuadrant.com/
>  PostgreSQL Development, 24x7 Support, Training & Services
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.postgres-xl.org/private.cgi/postgres-xl-developers-postgres-xl.org/attachments/20150814/4a112825/attachment.htm>

More information about the Postgres-xl-developers mailing list