Re: postgres 9 bind address for replication

From: Magnus Hagander <magnus(at)hagander(dot)net>
To: Andrew Dunstan <andrew(at)dunslane(dot)net>
Cc: Adam Crews <adam(dot)crews(at)gmail(dot)com>, pgsql-hackers(at)postgresql(dot)org
Subject: Re: postgres 9 bind address for replication
Date: 2012-08-06 09:53:19
Message-ID: CABUevEyucj3t5QgoCatL-=ppO2Xi85r_OUefzQu04=HOyt2w6Q@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-cluster-hackers pgsql-general pgsql-hackers

On Mon, Jul 23, 2012 at 8:45 PM, Andrew Dunstan <andrew(at)dunslane(dot)net> wrote:
>
> On 07/23/2012 02:23 PM, Adam Crews wrote:
>>
>> On Mon, Jul 23, 2012 at 8:34 AM, Robert Haas <robertmhaas(at)gmail(dot)com>
>> wrote:
>>>
>>> On Sat, Jul 21, 2012 at 1:24 AM, Adam Crews <adam(dot)crews(at)gmail(dot)com> wrote:
>>>>
>>>> Hello,
>>>>
>>>> I'm sorry for cross-posting, however I originally posted this to
>>>> pgsql-general list, but didnt get any replies.
>>
>> Then I posted to pgsql-cluster-hackers..., and now here.
>>
>>>>
>>>> I’m using pg 9.1.3 on CentOS 5 and have a few slave databases setup
>>>> using the built in streaming replication.
>>>>
>>>> On the slaves I set the “listen_addresses” config option to an ip
>>>> address for a virtual alias on my network interfaces. The host has an
>>>> address of 10.1.1.10, and there is a virtual alias of 10.1.1.40 that
>>>> the slave postmaster binds to.
>>>>
>>>> When the slave makes it connection to the master to start replication
>>>> the source address for the connection is the host address, not the
>>>> virtual alias address. Connections appear to come from 10.1.1.10,
>>>> instead of the slave postmaster address of 10.1.1.40.
>>>>
>>>> This seems like a bug to me. I could understand that if the
>>>> postmaster is listening on all interfaces, then it should use whatever
>>>> the IP is for the for the host, but in an instance where the
>>>> postmaster has been configured to listen to a specific address it
>>>> seems like the call to start the replication should be passed that
>>>> address so connections come from the slave postmaster’s IP, instead of
>>>> the host.
>>>>
>>>> Is there a config option that can be used to adjust this? I've looked
>>>> in the docs, but haven't found one yet.
>>>>
>>>> Is this perhaps a bug, or lack of feature?
>>>
>>> I don't think it's a bug, because the behavior you're hoping for might
>>> not be what everyone would want in a similar situation. It might
>>> qualify as an unimplemented feature.
>>>
>>> This mailing list isn't heavily used and this seems a bit off-topic
>>> for it anyway; you might want to try a different one for further
>>> discussion of this issue.
>>>
>> So, I think this, as Robert states, an unimplemented feature.
>>
>> For my situation it would be very useful to have an option to be able
>> to specify the source address for replication.
>>
>> I discovered this because I bind the listen address for postgres to a
>> single address even though the host system may have multiple
>> addresses. I then use that single address in iptables rules on other
>> systems. Since I expect the slave to be at a .40 address, but the
>> replication comes from the primary address of the interface (in this
>> case .10), my iptables rules were missing the access for the slave to
>> connect to the master.
>>
>> This site http://linux-ip.net/html/routing-saddr-selection.html
>> describes the behavior I'm seeing.
>>
>> How do I go about requesting a config option that would allow me to
>> specify the source address for the replication connections?
>>
>
> You just have :-)
>
> You could just add an iptables rule redirecting .10 packets on port 5432 (or
> whatever you're using) appropriately.
>
> We don't have any provision for binding the local end of any connection
> AFAIK. So the first question is "Do we want to?" and the second is "If yes,
> when and how?" I don't see that replication should be a special case - if
> this is worth providing for it should be applicable to all clients, ISTM.

I have an ugly patch lying around that implemented this as a libpq
connection option. It was just a quick hack to work around a situation
just like this (though not for replication), and ISTM that's the
proper place to put it.

I'll stick it on my TODO to try to clean that one up and submit for 9.3..

--
Magnus Hagander
Me: http://www.hagander.net/
Work: http://www.redpill-linpro.com/

In response to

Browse pgsql-cluster-hackers by date

  From Date Subject
Next Message Wang, Chaoyong 2012-08-21 10:40:41 How to copy Datum
Previous Message Adam Crews 2012-07-24 02:02:45 Re: postgres 9 bind address for replication

Browse pgsql-general by date

  From Date Subject
Next Message Marek Kielar 2012-08-06 12:08:43 Re: Clogging problem (was: Schema-only dump dumps no constraints, no triggers)
Previous Message Magnus Hagander 2012-08-06 09:31:39 Re: Hi,Frirends, are there any commands in pgsql/bin/ corresponding "create tablespace"?

Browse pgsql-hackers by date

  From Date Subject
Next Message Amit Kapila 2012-08-06 11:30:32 Re: [WIP] Performance Improvement by reducing WAL for Update Operation
Previous Message Magnus Hagander 2012-08-06 09:26:46 Re: Where is diskchecker.pl ?