Re: [Postgres-xc-general] question about pgxc_ctl and failover

[Koichi S. <koi...@gm...>]

You can add gtm slave by "add gtm slave" command of pgxc_ctl, then you can
start it by "start gtm slave" command.   For details, please refer to
pgxc_ctl document, available at
http://postgres-xc.sourceforge.net/docs/1_1/pgxc-ctl.html

Good luck;
---
Koichi Suzuki



2013/8/20 Sean Hogan <se...@co...>

>  Sorry, meant to CC the list with my reply.
>
>
> On 13-08-20 08:52 AM, Sean Hogan wrote:
>
> On 13-08-19 10:25 PM, Koichi Suzuki wrote:
>
> I have several questions on your configuration.
>
>  1. Did you configure your XC cluster with pgxc_ctl?   If not, I'm afraid
> pgxc_ctl may not work correctly if pgxc_ctl configuration file is exactly
> matches your manual configuration.
>
>
> Yes I used pgxc_ctl.
>
>
>  2. I assume that you're configuring GTM high-available, because GTM
> slave does not help if coordinator or datanode crashes.   It helps only
> when GTM master fails.    When you find GTM master fails, you can issue
> gtm_ctl to GTM slave to promote it to the new master.  (pgxc_ctl failover
> command will take care of this process).
>
>
> Yes, using a slave GTM.  My question is the same: after I use the failover
> command to make the slave the new master, how can I "demote" the previous
> master into being the slave?  Otherwise, after the failover the GTM becomes
> a single point of failure.
>
>
>  3. You need to configure GTM proxy to reconnect to the new master, or
> you need to stop coordinator and datanode, configure their postgresql.conf
> files to target to the new master, and restart it.  I recommend to
> configure GTM proxy.   Pgxc_ctl provide reconnect command to change GTM
> proxy target to the new master.
>
>
> I thought the failover command took care of reconnecting the GTM proxies,
> but okay.
>
>
>  4. If you'd like to make coordinator/datanode high-available, you need
> their slaves too.
>
>
> Yes, sorry I wasn't clear.  Just like the diagram on page 36 of
> http://wiki.postgresql.org/images/4/44/Pgxc_HA_20121024.pdf, I have two
> servers:
>
>    - server 1: coordinator 1 (master), datanode 1 (master), coordinator 2
>    (slave) and datanode 2 (slave), and GTM proxy
>     - server 2: coordinator 2 (master), datanode 2 (master), coordinator
>    1 (slave) and datanode 1 (slave), and GTM proxy
>
> GTM and GTM slave are running elsewhere.
>
>
> So after I issue "failover coordinator cn1", server 1 still has the
> coordinator 1 data directory that could be reconfigured as a slave for the
> new master for cn1 that is now running on server 2.  My question is: is
> there a simple way to accomplish that reconfiguration?  I.e. to change the
> role of server 1 so that it is running a slave for cn1, and update server 2
> so that the new cn1 master replicates to the new slave?
>
>  Thanks again,
> Sean
>
>
>  Regards;
> ---
> Koichi Suzuki
>
>
>
> 2013/8/19 Sean Hogan <se...@co...>
>
>>  Hi,
>>
>> I am investigating Postgres-XC with the coordinators and data nodes on
>> two servers.  My goal is primarily high availability rather than raw
>> performance, so I was planning to use one coordinator and one datanode on
>> each server, with the GTM master and slave elsewhere.  Just like on page 36
>> of http://wiki.postgresql.org/images/4/44/Pgxc_HA_20121024.pdf.
>>
>> I've been trying to use pgxc_ctl (the C program, not the Bash script at
>> https://github.com/koichi-szk/PGXC-Tools), but I'm having some
>> conceptual problems with the failover process.  I understand that
>> coordinator and datanode slaves are promoted and become identical to the
>> former masters (true?), but I'm unclear what happens to the former
>> masters.  "monitor all" no longer shows them.
>>
>> I suppose my question is: what do I need to do, to make the former
>> masters into new slaves?  To me it would make sense to be able to failover
>> node1 once and then again, and be left with more or less the same
>> configuration as in the beginning.  It would be okay if there is some magic
>> command I can run to reconfigure a former master as the new slave.
>>
>> Are these reasonable expectations, or am I asking too much of the
>> software?
>>
>> I also have a potential issue with what happens when one of my servers
>> fails completely.  The master coordinators and datanodes are running with
>> synchronous_commit = on and one value in synchronous_standby_names.
>> Doesn't that mean that transactions on the good server's master coordinator
>> and master datanode will block trying to replicate to their slaves on the
>> dead server?
>>
>> Sorry if these are RTFM items, I couldn't find clear answers in any of
>> the documentation.
>>
>> Thanks,
>> Sean
>>
>>
>> ------------------------------------------------------------------------------
>> Get 100% visibility into Java/.NET code with AppDynamics Lite!
>> It's a free troubleshooting tool designed for production.
>> Get down to code-level detail for bottlenecks, with <2% overhead.
>> Download for free and get started troubleshooting in minutes.
>>
>> http://pubads.g.doubleclick.net/gampad/clk?id=48897031&iu=/4140/ostg.clktrk
>> _______________________________________________
>> Postgres-xc-general mailing list
>> Pos...@li...
>> https://lists.sourceforge.net/lists/listinfo/postgres-xc-general
>>
>>
>
>
>