Обсуждение: Re: Log-Shipping Standby Server: USE_FLOAT8_BYVAL compatibility error

On 09/04/2012 06:04 PM, Mathias Breuninger wrote:
> Hi,
>
> I configure PostgreSQL 9.1.5 on two servers: one with FreeBSD-9.0/powerpc64 and the other with FreeBSD-9.0/powerpc.
>
> The 64 bits server is the master and I tried to configure the 32 bits one as a standby server.

That isn't a supported configuration. The standby needs to be the same
architecture as the master and needs to use the same compilation options.

I'm surprised the error message wasn't more useful in this case.

> I can recompile PostgreSQL on the 64bit architecture without this configure option, but I wonder if there will be a
performancedrawback. 

It still won't work. If you want to replicate from a 64-bit machine to a
32-bit machine you will need to build a 32-bit version of PostgreSQL on
the 64-bit machine.

--
Craig Ringer

On 09/04/2012 11:45 PM CEST, Craig Ringer wrote:
> > The 64 bits server is the master and I tried to configure the 32 bits
> one as a standby server.
>
> That isn't a supported configuration. The standby needs to be the same
> architecture as the master and needs to use the same compilation
> options.
>
> I'm surprised the error message wasn't more useful in this case.
>
> > I can recompile PostgreSQL on the 64bit architecture without this
> configure option, but I wonder if there will be a performance drawback.
>
> It still won't work. If you want to replicate from a 64-bit machine to
> a 32-bit machine you will need to build a 32-bit version of PostgreSQL
> on the 64-bit machine.
>
> --
> Craig Ringer

Thank you for your answers.

I wasn't aware of the binary problem with replication.
Maybe the PostgreSQL docs should emphasize the architecture restriction.

I'll explore the other solutions.

--
Mathias

On Wed, Sep 5, 2012 at 2:21 AM, Mathias Breuninger
> I wasn't aware of the binary problem with replication.
> Maybe the PostgreSQL docs should emphasize the architecture restriction.

That's sort of what "binary file" means -- a raw,
architecture-specific representation of data that's optimized for
performance and/or space, not portability.

Craig

On 09/05/2012 4:59 PM, Craig James wrote:
> On Wed, Sep 5, 2012 at 2:21 AM, Mathias Breuninger
> > I wasn't aware of the binary problem with replication.
> > Maybe the PostgreSQL docs should emphasize the architecture
> restriction.
>
> That's sort of what "binary file" means -- a raw, architecture-specific
> representation of data that's optimized for performance and/or space,
> not portability.
>
> Craig

In fact, it is written clearly in the docs (25.2.1 Planning: " In any case the hardware architecture must be the same —
shippingfrom, say, a 32-bit to a 64-bit system will not work."), but I missed it. 
So I apologize for the annoyance.

Regards.

--
Mathias