Christopher Browne cbbrowne
Mon Apr 25 15:49:20 PDT 2005
Hannu Krosing wrote:

>On P, 2005-04-24 at 04:00 +0300, Hannu Krosing wrote:
>How much change would it be to have an option to not used LISTEN/NOTIFY
>in slony ?
>
>On heavily changing databases with some long-running transactions
>pg_listener can easily grow  to tens of megabytes in size and this can
>become the main bottlenect to slony's performance, as pg_listener is a
>table with no indexes.
>
>And even if there were indexes, sending notifications is un-needed
>overhead for databases which have above certain amount of constant
>traffic.
>
>Of course this could be also done by slon's in a dynamic way, i.e when
>getting heavy stream of sl_events, just UNSUBSCRIBE and after N queries
>in a row getting nothing, do a SUBSCRIBE, similar to how gigabit
>ethernet drivers for some cards work under linux, switching to polling
>mode after a certain number of interrupts/sec is reached.
>  
>
I had a chat with Jan on this one; have some notes.

It definitely fits into post-1.1 work...


More information about the Slony1-general mailing list