Re: Save The Date: Cluster-Hackers meeting May 21st

On Thu, Jan 17, 2013 at 5:24 PM, Koichi Suzuki <koichi(dot)szk(at)gmail(dot)com> wrote:

> Cluster summit should focus on what PG core should do to support
> various clustering use case and external tools. I think logical
> replication as well as DDL trigger should be given longer time to
> discuss.
>

+1

I won't be able to attend, unfortunately. But I hope that another
conversation happens at PgCon:

I am very interested in defining terms and setting up repeatable testing
for replication. I've started this wiki page:

http://wiki.postgresql.org/wiki/Binary_Replication_Tools

What I've found is that we don't use consistent terminology for these
tools, and they largely don't provide complete test cases that would help
someone setting up replication, backups and restores.

I've made some very rough shell scripts to test each one, but it needs
quite a bit more work. My ultimate goal is to see these kinds of scripts
incorporated into a Jenkins-like system that can automatically test new
releases against new versions of PostgreSQL. I find it really problematic
that we don't have a consistent way of doing this. Sysadmins who are
supporting these tools are particularly disadvantaged because they are
often supporting other databases and systems that have wildly different
backup procedures.

Just the other day, a friend told me about failing over and not having
secondary slaves follow the new master in a 9.0 system. The information
about which versions of Postgres support this feature is not easy to find.
I plan on incorporating that information in this wiki page over time -- but
again, we need to standardize terminology across all of the third-party
tools (open source and otherwise) so that people using these are not caught
by surprise when they are the most vulnerable to data loss.

-selena

--
http://chesnok.com