[Pgcluster-general] reading on one node creates traffic on other

[Boris Schulz]

Hi,

I am currently testing cybercluster on a two nodes plus one master setup.
There is an observation that I would be glad if someone could explain it to
me.

When I open 10 connections to one node to read from it ps shows me 10
postgres instances with connections from the replication master on the
second node as well, mainly "idle in transaction".
Why is this happening? I could just test this, but as I am asking already:
does this mean that the sum of all connections in a cluster is opened for
every node?

Now, what bothers me is this: the second node executes some selects as well.
Not all selects that I do on the first node get executed on the second node
but some.
I did an strace on the process and know that tables get read to wich I do
not write but that I read from on the first node. IO clearly indicates that
the data is read as well.

I am quite unsure what to make of this but as additional IO is a performance
issue I would like to solve this. Could I have messed up some config
parameter? Is this normal behaviour? Something else?

greets, B.S.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://pgfoundry.org/pipermail/pgcluster-general/attachments/20080109/9c9014be/attachment.html