Re: Barman disaster recovery solution

On 2/27/19 4:48 PM, Achilleas Mantzios wrote:
> On 27/2/19 4:16 μ.μ., David Steele wrote:
>> On 2/27/19 2:31 PM, Achilleas Mantzios wrote:
>>> On 27/2/19 1:58 μ.μ., richter@simkorp.com.br wrote:
>>>> Just to notice, I d o use backup from standby and WAL archive from 
>>>> standby. It is possible. But you have to configure standby with 
>>>> option of wal archive "always".
>>>
>>> I guess there are issues with it. If this was so easy then pgbarman 
>>> and pgbackrest would support it out of the box.
>>
>> There are a few issues with it:
>>
>> 1) If you allow the primary and standby to archive to the same 
>> repository then there needs to be some conflict resolution if they 
>> write at the same time.  If they write to different repositories then 
>> you need to decided which one to use for a restore, or have some kind 
>> of conflict resolution between them.  It gets complicated.
>>
>> 2) Writing only from the standby reduces load on the primary but if 
>> the connection to the primary is down then you can get behind on 
>> archiving. If something then happens to the primary then your recovery 
>> point will be limited.
> 
> David to quote an older email from you:
> "pgBackRest currently requires some files and all WAL to be sent from 
> the primary even when doing backup from standby.  We may improve this in 
> the future but it's not on the road map right now. "
> So, I had the impression that receiving WALs from the standby was a 
> greater technical problem.

No, it just increases the risk of being behind on archiving.

One of the things pgBackRest does well is move a *lot* of WAL and it is 
orders of magnitude faster than streaming replication, which is 
single-threaded and uncompressed.  So, in spite of the additional load 
it's generally safest to archive from the primary, especially on high 
write volume clusters.

-- 
-David
david@pgmasters.net