On 12-04-24 10:40 AM, Mirko Vogt wrote:
> Hey all!
>
> I successfully setup my first slony replication - however quite a few
> conceptual questions raised on the way:
>
> I have the following setup (taken from slon_tools.conf):
>
> add_node(node     =>  1,
>           host     =>  'master.foo',
>           dbname   =>  'foo',
>           port     =>  5432,
>           user     =>  'slony',
>           password =>  'XXX');
>
> add_node(node     =>  11,
>           host     =>  'slave1.foo',
>           dbname   =>  'foo',
>           port     =>  6254,
>           user     =>  'slony',
>           password =>  'XXX');
>
> add_node(node     =>  12,
>           host     =>  'slave2.foo',
>           dbname   =>  'foo',
>           port     =>  2254,
>           user     =>  'slony',
>           password =>  'XXX');
>
> This config got deployed on every node, every pg_hba.conf-file contained
> a line which allowed all other servers to connect as user slony.
>
> Everything was working - until I tried to optimize.
>
> I thought, well, there needs to be a connection between the master and
> the slaves but no direct connection between the slaves - so I dropped
> the access lines in pg_hba.conf on the slaves for the other slave
> respectively.
>
> The setup seemed to still work, however on the slaves I noticed error
> messages like:
>    FATAL:  no pg_hba.conf entry for host "slave1.foo", user "slony",
> database "foo"
>
> Okay, fine, obviously they try to connect to each other: so I purged out
> the respective node-definitions out of the slon_tools.conf file on the
> slaves (on node 11 I deleted the definition of node 12 and vice versa).
>

It is the slon process for each of the slaves that needs to connect to 
the other slaves.  Some people run the slon process on a different 
server than the database.



> After restarting slony the slaves still tried to connect to each other.
> Where do they have the connect information from? And why are they trying
> to connect to each other at all?

Slony stores the paths that the slon uses to connect from one node to 
another in the sl_path table.  Do not edit this table directly (but you 
can view it). Use the STORE PATH command to change the connection 
information  http://www.slony.info/documentation/2.0/stmtstorepath.html

There is no slonik_store_path command in the alt perl tools (patches 
always welcome)



>
> Anyway, next thought: if one node gets hacked the attacker shouldn't be
> able to access the database on the other nodes. Idea was: The slaves do
> not need to access the master with a user who has write access to that
> database (slony). That's why I created a read-only user on the master
> (slony_ro) and tried to tell the slaves - by changing the user 'slony'
> to 'slony_ro' within the slon_tools.conf-files - to connect as 'slony_ro'.
> However also that change didn't show any effect after restarting slony.
>
> It seems to me - by initializing the cluster, creating and subscribing
> to the going-to-be-replicated sets - the information got pushed to the
> slaves from the master.
>
> That raises 2 (sub-)questions:
> a) Where is this information stored?
> b) why there is the need of a slon_tools.conf file if its data is not
> used anyway (at least on the slaves)?

Some of the data in slon_tools.conf is used in initializing the cluster 
but then changes later to the .conf file don't get reflected, the 
slon_tools.conf file is a configuration file for the TOOLS not for slony 
itself.



>
> Maybe somebody could lighten me up here? I didn't find any information
> able to clear my confusion about that yet :/
>
> Cheers, thanks a lot in advance and have a nice week!
>
>    mirko
> _______________________________________________
> Slony1-general mailing list
> Slony1-general at lists.slony.info
> http://lists.slony.info/mailman/listinfo/slony1-general