On Sat, 2008-04-12 at 08:55 -0400, Bill Moran wrote:
> > > I've been putting this together over the last couple weeks for
> > > the reasons listed on the HTML page:
> > >
> http://people.collaborativefusion.com/~wmoran/PostgreSQL/slony_switchover.html
> > > 
> > > I'm interested to hear how useful this is to others, and of course
> > > suggestions for improvement are welcome.
> 
> Are you running Linux by any chance?  This script was developed and
> is deployed for us on FreeBSD, and I've verified that it works as
> advertised on that platform.  I've _attempted_ to make sure that it
> will work on Linux as well, but I've not had the opportunity to test
> it.  Testing and feedback is encouraged.

I was able to do some testing on Linux this week.  I setup three nodes
running Debian 4.0 (Etch), Postgresql 8.1, and Slony 1.2.1-1.  In the
end, I didn't succeed at moving the replication origin between nodes
using slony_switchover.sh (although slonik_move_set worked).  

I fixed the first issue I encountered (bug #3 below) by adding extra
functionality to the script. However, the problem that I got stuck on
was that sl_path didn't contain connection info between my two
subscriber nodes, node 30 and node 31.  

Here's what my sl_path table looks like:

community=# select pa_server,pa_client,pa_conninfo from sl_path ;
 pa_server | pa_client |  pa_conninfo (truncated)         
-----------+-----------+---------------------------------------
        30 |        29 | host=172.16.32.130 dbname=community 
        29 |        30 | host=172.16.32.129 dbname=community 
        29 |        31 | host=172.16.32.129 dbname=community 
        31 |        29 | host=172.16.32.131 dbname=community 
(4 rows)

Notice that there are no entries for (pa_server, pa_client) = (31, 30)
or (30, 31).  Not sure if sl_path is missing entries due to a mistake I
made in configuring replication, or perhaps a bug in the version of the
slonik_* scripts packaged for Debian 4.0.  In any case, on my FreeBSD
production machines sl_path is complete, so this isn't a blocker for my
using this in production.

Here are the bugs I found.  Only the first is Linux-specific.  Hope this
is helpful feedback.

1. absolute path to psql, line 89

psql is at /usr/bin/psql in debian

  R=`/usr/local/bin/psql -t -U $PUSER -d $DB -A -c "$1" | /usr/bin/tail
-1`

2. Comparison operator '-eq' should be '=' at line 102

JOT=`which jot`
if [ "x$JOT" -eq "x" ]  <--- 102
then
  JOT=`which seq`
fi

3. slony_switchover.sh fails when node ids don't follow the sequence 1,
2, 3, .... For example, my slons have node ids 29, 30, and 31.  As a
result, the following comparison always fails at line 109:

for I in `$JOT $NUMNODES`;
do
  if [ $ME -eq $I ]   <--- 109
  then


Dane