Обсуждение: Understanding PG9.0 streaming replication feature

Hi,

I am having some trouble trying to figure out how to configure this particular scenario..

I have a pair of pg servers that I want to put in a Master/Standby configuration. Currently a script dumps the master db every hour, copies it to the standby, restores, and restarts the server. The aim is to replace the dumps/restores with streaming replication and ensure the standby is always up to date.

In this case writes are infrequent, but reads are constant, and I only need high availability for reads. I would ideally like both master and standby to be available simultaneously to allow load-balancing.
My confusion seems to be around the fact I don't need failover - my applications will detect a master down and immediately start using the standby, so there is no need to allow writes on the standby, they will just wait for the master to be available again - I am not sure what the minimum config needed for this scenario is..

cheers,

Ben

FYI, not looking for a detailed how to here.. I have read the manual twice and just can't figure which sections are relevant. The manual seems to be trying to cover all uses simultaneously which is always going to get confusing :) For example do I need I need WAL archiving or not?

Hi Ben,

On 2010/12/22 7:46, Ben Carbery wrote:
> FYI, not looking for a detailed how to here.. I have read the manual twice and just can't figure which sections are
relevant.The manual seems to be trying to cover all uses simultaneously which is always going to get confusing :) For
exampledo I need I need WAL archiving or not? 

My blog entry would be a good entry point for you.  :)

5 steps to implement a PostgreSQL replication system
http://pgsnaga.blogspot.com/2010/05/5-steps-to-implement-postgresql.html

It was written to be a guide for building a simple master-slave config.

Please take a look, including the comments.

Thanks,


On 2010/12/22 7:46, Ben Carbery wrote:
> FYI, not looking for a detailed how to here.. I have read the manual twice and just can't figure which sections are
relevant.The manual seems to be trying to cover all uses simultaneously which is always going to get confusing :) For
exampledo I need I need WAL archiving or not? 
>
> On Tue, Dec 21, 2010 at 2:40 PM, Ben Carbery <ben.carbery@gmail.com <mailto:ben.carbery@gmail.com>> wrote:
>
>     Hi,
>
>     I am having some trouble trying to figure out how to configure this particular scenario..
>
>     I have a pair of pg servers that I want to put in a Master/Standby configuration. Currently a script dumps the
masterdb every hour, copies it to the standby, restores, and restarts the server. The aim is to replace the
dumps/restoreswith streaming replication and ensure the standby is always up to date. 
>
>     In this case writes are infrequent, but reads are constant, and I only need high availability for reads. I would
ideallylike both master and standby to be available simultaneously to allow load-balancing. 
>     My confusion seems to be around the fact I don't need failover - my applications will detect a master down and
immediatelystart using the standby, so there is no need to allow writes on the standby, they will just wait for the
masterto be available again - I am not sure what the minimum config needed for this scenario is.. 
>
>     cheers,
>
>     Ben
>
>
>
>


--
NAGAYASU Satoshi <satoshi.nagayasu@gmail.com>

Hi Ben,

load balancing is not possible with the tools that are in the postgres
installation. There is no automatic switch-over to a slave if the master
fails. The trigger file needs to be created to make a slave to the master. This
is not done automaitcally by postgres, but should be done by a cluster
software (i.e. pacemaker).

If you can live without load balancing, read the page of Satoshi.

But the slaves can handle read requests (SELECT). Maybe this helps a little
bit.

Regards
Matthias

Am Mittwoch, 22. Dezember 2010, um 00:31:14 schrieb Satoshi Nagayasu:
> Hi Ben,
>
> On 2010/12/22 7:46, Ben Carbery wrote:
> > FYI, not looking for a detailed how to here.. I have read the manual
> > twice and just can't figure which sections are relevant. The manual
> > seems to be trying to cover all uses simultaneously which is always
> > going to get confusing :) For example do I need I need WAL archiving or
> > not?
>
> My blog entry would be a good entry point for you.  :)
>
> 5 steps to implement a PostgreSQL replication system
> http://pgsnaga.blogspot.com/2010/05/5-steps-to-implement-postgresql.html
>
> It was written to be a guide for building a simple master-slave config.
>
> Please take a look, including the comments.
>
> Thanks,
>
> On 2010/12/22 7:46, Ben Carbery wrote:
> > FYI, not looking for a detailed how to here.. I have read the manual
> > twice and just can't figure which sections are relevant. The manual
> > seems to be trying to cover all uses simultaneously which is always
> > going to get confusing :) For example do I need I need WAL archiving or
> > not?
> >
> > On Tue, Dec 21, 2010 at 2:40 PM, Ben Carbery <ben.carbery@gmail.com
<mailto:ben.carbery@gmail.com>> wrote:
> >     Hi,
> >
> >     I am having some trouble trying to figure out how to configure this
> >     particular scenario..
> >
> >     I have a pair of pg servers that I want to put in a Master/Standby
> >     configuration. Currently a script dumps the master db every hour,
> >     copies it to the standby, restores, and restarts the server. The aim
> >     is to replace the dumps/restores with streaming replication and
> >     ensure the standby is always up to date.
> >
> >     In this case writes are infrequent, but reads are constant, and I
> >     only need high availability for reads. I would ideally like both
> >     master and standby to be available simultaneously to allow
> >     load-balancing. My confusion seems to be around the fact I don't
> >     need failover - my applications will detect a master down and
> >     immediately start using the standby, so there is no need to allow
> >     writes on the standby, they will just wait for the master to be
> >     available again - I am not sure what the minimum config needed for
> >     this scenario is..
> >
> >     cheers,
> >
> >     Ben

On Wed, Dec 22, 2010 at 8:31 AM, Satoshi Nagayasu
<satoshi.nagayasu@gmail.com> wrote:
> My blog entry would be a good entry point for you.  :)
>
> 5 steps to implement a PostgreSQL replication system
> http://pgsnaga.blogspot.com/2010/05/5-steps-to-implement-postgresql.html

Or
http://wiki.postgresql.org/wiki/Binary_Replication_Tutorial
http://wiki.postgresql.org/wiki/Streaming_Replication#How_to_Use

Regards,

--
Fujii Masao
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center

Thanks for the responses all, I have this working now. I had to create a base backup before copying to the standby for replication to start, but the main sticking point was actually understanding the terms and concepts involved..

I think the Binary Replication Tutorial page on the wiki basically explains everything. Unfortunately the actual pg manual is still about as clear as mud even though I now have a vague idea of how this all works. I think this is worth mentioning given the majority of the pg manual is actually of an unusually high standard - probably among the best technical manuals I have read in terms of being both comprehensive and concise, so it's a shame that this section doesn't meet that standard (IMO). Hopefully this will get a rewrite at some point!

cheers,

Ben

Ben Carbery wrote:
> Thanks for the responses all, I have this working now. I had to create a
> base backup before copying to the standby for replication to start, but the
> main sticking point was actually understanding the terms and concepts
> involved..
>
> I think the Binary Replication Tutorial page on the wiki basically explains
> everything. Unfortunately the actual pg manual is still about as clear as
> mud even though I now have a vague idea of how this all works. I think this
> is worth mentioning given the majority of the pg manual is actually of an
> unusually high standard - probably among the best technical manuals I have
> read in terms of being both comprehensive and concise, so it's a shame that
> this section doesn't meet that standard (IMO). Hopefully this will get a
> rewrite at some point!

Can you give some concrete suggestions on what needs to be added?  The
current documentation is here:

    http://developer.postgresql.org/pgdocs/postgres/index.html

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +

Dan Birken wrote:
> (I am not the OP, but recently went through the same thing so I'll chime in)
>
> Reading through the documentation now (albeit with a now pretty good
> understanding of how everything works), I think the main confusing thing is
> how different bits which apply to file-base log shipping, streaming
> replication and both of them are thrown together on this
> page<http://developer.postgresql.org/pgdocs/postgres/warm-standby.html>,
> making it difficult to figure out what you need to know if you are just
> looking to implement streaming replication.

Nice idea to use color to highlight stuff.  You are right that the
streaming docs were added later, and seem out of place.

I have applied the attached patch so the text is more fluid.  Let me
know if you have additional suggestions.

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +
diff --git a/doc/src/sgml/high-availability.sgml b/doc/src/sgml/high-availability.sgml
index 717347c..ce3f264 100644
*** a/doc/src/sgml/high-availability.sgml
--- b/doc/src/sgml/high-availability.sgml
*************** protocol to make nodes agree on a serial
*** 486,503 ****
    <para>
     Directly moving WAL records from one database server to another
     is typically described as log shipping. <productname>PostgreSQL</>
!    implements file-based log shipping, which means that WAL records are
!    transferred one file (WAL segment) at a time. WAL files (16MB) can be
     shipped easily and cheaply over any distance, whether it be to an
     adjacent system, another system at the same site, or another system on
     the far side of the globe. The bandwidth required for this technique
     varies according to the transaction rate of the primary server.
!    Record-based log shipping is also possible with streaming replication
!    (see <xref linkend="streaming-replication">).
    </para>

    <para>
!    It should be noted that the log shipping is asynchronous, i.e., the WAL
     records are shipped after transaction commit. As a result, there is a
     window for data loss should the primary server suffer a catastrophic
     failure; transactions not yet shipped will be lost.  The size of the
--- 486,504 ----
    <para>
     Directly moving WAL records from one database server to another
     is typically described as log shipping. <productname>PostgreSQL</>
!    implements file-based log shipping by transfering WAL records
!    one file (WAL segment) at a time. WAL files (16MB) can be
     shipped easily and cheaply over any distance, whether it be to an
     adjacent system, another system at the same site, or another system on
     the far side of the globe. The bandwidth required for this technique
     varies according to the transaction rate of the primary server.
!    Record-based log shipping is more granular and streams WAL changes
!    incrementally over a network connection (see <xref
!    linkend="streaming-replication">).
    </para>

    <para>
!    It should be noted that log shipping is asynchronous, i.e., the WAL
     records are shipped after transaction commit. As a result, there is a
     window for data loss should the primary server suffer a catastrophic
     failure; transactions not yet shipped will be lost.  The size of the
*************** protocol to make nodes agree on a serial
*** 505,512 ****
     <varname>archive_timeout</varname> parameter, which can be set as low
     as a few seconds.  However such a low setting will
     substantially increase the bandwidth required for file shipping.
!    If you need a window of less than a minute or so, consider using
!    streaming replication (see <xref linkend="streaming-replication">).
    </para>

    <para>
--- 506,513 ----
     <varname>archive_timeout</varname> parameter, which can be set as low
     as a few seconds.  However such a low setting will
     substantially increase the bandwidth required for file shipping.
!    Streaming replication (see <xref linkend="streaming-replication">)
!    allows a much smaller window of data loss.
    </para>

    <para>

On Wed, Jan 26, 2011 at 02:22:41PM -0800, Dan Birken wrote:
> > Can you give some concrete suggestions on what needs to be added?  The
> > current documentation is here:
> >
> >        http://developer.postgresql.org/pgdocs/postgres/index.html


I had some difficulty getting the keepalives syntax in recovery.conf.

 http://developer.postgresql.org/pgdocs/postgres/warm-standby.html#STREAMING-REPLICATION

might stand an example.  When you click on the link there it takes you here:

 http://developer.postgresql.org/pgdocs/postgres/runtime-config-connection.html#GUC-TCP-KEEPALIVES-IDLE

which doesn't help to much.  Somewhere I found a comment that did it this way:

 primary_conninfo = 'host=test user=fred keepalives_idle= 60 '

The example I found for this in "PostgreSQL 9 Admin Cookbook" didn't
fly: primary_conninfo = '....options="-c tcp_keepalives_idle= 60" ...'
I didn't check the erratta.  Just sayin....

Ray Stell wrote:
> On Wed, Jan 26, 2011 at 02:22:41PM -0800, Dan Birken wrote:
> > > Can you give some concrete suggestions on what needs to be added?  The
> > > current documentation is here:
> > >
> > >        http://developer.postgresql.org/pgdocs/postgres/index.html
>
>
> I had some difficulty getting the keepalives syntax in recovery.conf.
>
>  http://developer.postgresql.org/pgdocs/postgres/warm-standby.html#STREAMING-REPLICATION
>
> might stand an example.  When you click on the link there it takes you here:
>
>  http://developer.postgresql.org/pgdocs/postgres/runtime-config-connection.html#GUC-TCP-KEEPALIVES-IDLE
>
> which doesn't help to much.  Somewhere I found a comment that did it this way:
>
>  primary_conninfo = 'host=test user=fred keepalives_idle= 60 '
>
> The example I found for this in "PostgreSQL 9 Admin Cookbook" didn't
> fly: primary_conninfo = '....options="-c tcp_keepalives_idle= 60" ...'
> I didn't check the erratta.  Just sayin....

Uh, the sentence is:

    On systems that support the keepalive socket option, setting
    tcp_keepalives_idle, tcp_keepalives_interval and tcp_keepalives_count
    helps the primary promptly notice a broken connection.

I believe there they are talking about setting those values in the
master/primary so you see the slave has disconnected --- this is not
related to setting in the connection string, I think.

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +

On Wed, Jan 26, 2011 at 09:18:01PM -0500, Bruce Momjian wrote:
>
> Uh, the sentence is:
>
>     On systems that support the keepalive socket option, setting
>     tcp_keepalives_idle, tcp_keepalives_interval and tcp_keepalives_count
>     helps the primary promptly notice a broken connection.
>
> I believe there they are talking about setting those values in the
> master/primary so you see the slave has disconnected --- this is not
> related to setting in the connection string, I think.


Yep.  Simon shows it the other way in his book:

 "If you want a Standby to notice that the connection to the Master has
 dropped, you need to set the keepalives in the primary_conninfo in the
 recovery.conf"

That works, also.

On Wed, Jan 26, 2011 at 09:02:24PM -0500, Ray Stell wrote:
> On Wed, Jan 26, 2011 at 02:22:41PM -0800, Dan Birken wrote:
> > > Can you give some concrete suggestions on what needs to be added?  The
> > > current documentation is here:


It seems like there is a departure in postgresql/pg_hba.conf with 9.x
at standby activation.  I'm running 8.x pitr standbys with the same conf
files on both systems.  This is not possible in 9.x, right?

After the standby is activated, don't you have to restart with the a
different config file in order to get a new standby into recovery?

I suggest a statement about actions needed after a standby is activated.
This could be added in the 25.2.x somewhere.

wal_level and max_wal_senders are not dynamic and are not set as needed on
the new primary.  Could this should be hacked into the activation code?