Обсуждение: Will switchover still need a checkpoint in 9.1 SR Hot Standby

In 9.0 (as in earlier versions) a former standby host has to do a full
checkpoint before becoming available as an independent database instance
in either switchover or failover scenarios. 

For most combinations of of bigger than minimal shared buffers and
non-memory-speed disks this can take from several seconds to tens of
minutes on busy systems.

Is the pre-activation checkpoint still required in 9.1 ?

The reason I ask is that I'm looking into making planned switchovers and
adding partitions more instantaneous. 

Streaming replication provides replica lag of only millisecond vs. high
tens or even hundreds of milliseconds achievable on trigger-based
replication systems like Londiste. 

Where Londiste currently wins, is the ability to become available
instantaneously without any checkpoint or other lengthy maintenance
operations.

-- 
-------
Hannu Krosing
PostgreSQL Infinite Scalability and Performance Consultant
PG Admin Book: http://www.2ndQuadrant.com/books/

On Sun, Aug 7, 2011 at 8:57 AM, Hannu Krosing <hannu@2ndquadrant.com> wrote:
> In 9.0 (as in earlier versions) a former standby host has to do a full
> checkpoint before becoming available as an independent database instance
> in either switchover or failover scenarios.
>
> For most combinations of of bigger than minimal shared buffers and
> non-memory-speed disks this can take from several seconds to tens of
> minutes on busy systems.

For switchover, you issue a checkpoint first, to reduce this time as
much as possible.

> Is the pre-activation checkpoint still required in 9.1 ?

Yes, but I've found a way to remove them in 9.2 and will be patching that soon.

--
 Simon Riggs                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services

On Aug 7, 2011, at 11:01 AM, Simon Riggs wrote:

> On Sun, Aug 7, 2011 at 8:57 AM, Hannu Krosing <hannu@2ndquadrant.com> wrote:
>> In 9.0 (as in earlier versions) a former standby host has to do a full
>> checkpoint before becoming available as an independent database instance
>> in either switchover or failover scenarios.
>>
>> For most combinations of of bigger than minimal shared buffers and
>> non-memory-speed disks this can take from several seconds to tens of
>> minutes on busy systems.
>
> For switchover, you issue a checkpoint first, to reduce this time as
> much as possible.
>
>> Is the pre-activation checkpoint still required in 9.1 ?
>
> Yes, but I've found a way to remove them in 9.2 and will be patching that soon.



hi simon,

this is highly interesting. this is am important issue for big iron.
can you share the idea you have in mind?
many thanks,
    hans


--
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt, Austria
Web: http://www.postgresql-support.de