Re: [Postgres-xc-general] General config/arch questions

Hi Lionel,

Just to complete a little bit the answers of my colleague...

On Sat, Mar 5, 2011 at 3:36 PM, Abbas Butt <abb...@te...> wrote:

> My actual setup is :
>> pgxc1 for GTM, Coordinator and datanode
>> pgxc2 for datanode only
>>
>
> You mean you will have 2 computers, one will run GTM, Coordinator and 1st
> datanode and the other the 2nd data node. If yes then this would be fine.
>
> BTW what Linux distribution will you be using?
>
> Please remember that for the first time gtm has to be started with -x
> option and it is better to start gtm before you initdb for the coordinator
> and the 2 data nodes. For successive runs you should skip -x option. As a
> general rule you should always start gtm first then the data nodes and then
> the coordinator.
>
Basically, you can initialize your data directory when you want.
The only point you have to take care is to start GTM before the other nodes
as when you start a node it tries to register on GTM.
After that it doesn't matter if you start Coordinators or Datanodes first.

For GTM startup, you have to set for the first time -x which permit to set
the first value of GXID GTM will feed to Postgres-XC nodes.
If for instance you stopped you cluster, you can restart GTM from the same
data folder as before with having to precise the first GXID value as this
value will be taken automatically from the GTM data folder if the file where
last GXID was written exists.


>
>> 1.General : Is a coordinator needed for each node, or one coordinator
>> 'to rule them all' can be setup ?
>>
> You can use one Coordinator with 100 Datanodes if you desire.
It may be better if the ratio Coordinator/Datanode is close to 1, but we
also found that if you set one Coordinator and one Datanode on the same
machine, Coordinator was using 30% of ressources and Datanode 70%.
With such numbers, a ratio of 0.5 is also possible.

>
>
>>
>> 2. Configuration :  What should be the differences between
>> postgresql.conf in /datanode and /coordinator, if there are any?
>
> Coordinator uses a connection pooler process. So it needs to know the
following parameters:
coordinator_hosts, data_nodes_hosts, coordinator_ports, datanode_ports,
pooler_port, data_node_num, coord_num, pgxc_node_id, gtm_port, gtm_host,
datanode_user, datanode_passd, coord_passwd, coord_user.
You have also pooler connection parameters to set but I forgot all the
names.
For datanode, you have to take care of GTM connection parameter and
pgxc_node_id (used to register on GTM).

In your cluster you can have a Coordinator 1 and a Datanode 1 as the
difference between node type is made when registering nodes on GTM.


>
>> Some
>> portions are they ignored for a specific function (ex : coordinator vs
>> datanode config, etc...)
>> In this can pg_hba.conf & postgresql.conf be shared for the same
>> server (maybe using symbolic links...)
>>
> No, you have to set up each postgresql.conf and pg_hba.conf for each node
separately as each Coordinator and Datanode use different data folder.

>
>
>> 3. GTM : I get on second node an "WARNING:  Do not have a GTM snapshot
>> available", can this be related to previous config files/setup  ?
>>
> This error means that you didn't set up GTM connection parameters
correctly.

>
>
>>
>> 4. Administration : are there ways of getting the online status of the
>> nodes from one node to another
>>
>
> This is currently under development. I can provide you details later.
>
Those experimental functionalities are located on a separate branch called
ha_support in the GIT repo.

We are also thinking about adding some catalog extensions to allow
coordinator to keep an eye on Datanodes as such a view process is linker to
the connection pooling process.
This is just a thought though.
Now as we are focusing on code stability for the core, this is not a high
priority.
But as we merged with PostgreSQL 9.0, it may be possible in a close future
to use XC with HOT Standby nodes. Current streaming replication is not
synchronized so its usage is now limited in current XC.

If you have any other questions, don't hesitate.
Regards,
-- 
Michael Paquier
http://michaelpq.users.sourceforge.net