On Mon, May 2, 2011 at 7:44 AM, Greg Stark <gsstark@mit.edu> wrote:
> On Sun, May 1, 2011 at 9:31 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
>> I don't think the performance of replication is at issue. This is
>> about resource control.
>>
>
> The unspoken question here is why would replication be affected by i/o
> load anyways? It's reading data file buffers that have only recently
> been written and should be in cache. I wonder if this system has
> chosen O_DIRECT or something like that for writing out wal?
It's not, that is a misunderstanding in the thread.
It appears that the sheer volume of WAL being generated slows down
replication. I would guess it's the same effect as noticing a slow
down on web traffic when somebody is watching streaming video.
The requested solution is the same as the network case: rate limit the
task using too much resource, if the user requests that.
I can't see the objection to replacing something inadvertently removed
in 9.0, especially since it is a 1 line patch and is accompanied by
copious technical evidence. Sure, we can do an even better job in a
later release.
--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services