Re: Streaming Replication limitations

Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
> 2011/4/13 Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>:
>> Short answer is to test the case you have in mind and see.

> That's the long answer, not least because the absence of a failure in
> a test is not conclusive proof that it won't fail at some point in the
> future while in production.

Not really. Every known source of incompatibility (endianness,
alignment, float format, etc) is checked at postmaster startup via
entries in pg_control. If you get the slave postmaster to start at all,
it will probably work, though certainly more extensive testing than that
would be advisable.

> The short answer is "don't do it".

DBAs are paid to be incredibly paranoid, and from that mindset this
answer makes sense. But there's a big difference between "it won't work"
and "I'm afraid to risk my paycheck on this because there might possibly
be some problem that no one knows about yet". Let's be perfectly clear
that this is a question of the second case not the first.

regards, tom lane