Steve Singer ssinger at
Tue Nov 23 06:31:33 PST 2010
This idea is to try and deal with the following problem.

A user has a multi-node cluster and needs to bring one of the nodes down 
for a while (say a few day).  During this time slony collects the 
transactions to replicate in sl_log_1 and sl_log_2.

The problem is that sl_log_1 or sl_log_2 will bloat to be very big. 
Generating SYNC's will take a lot longer and reading SYNC's for other 
nodes can take a lot longer.

Slony can get into a state where it can't keep up/catch up with 
replication because the sl_log table is so large.

Does this problem bite people often enough in the real world for us to 
devote effort to fixing?

If instead of having 2 sl_log tables we could have a dynamic number of 
them.   During a log switch if no sl_log table is ready to be truncated 
we could instead create a new one.

This would complicate the  logic where we select from sl_log_1 and UNION 
with sl_log_2.

More information about the Slony1-general mailing list