On Thu, Jun 12, 2008 at 10:40 PM, Chris St Denis <chris@smartt.com> wrote:
> I upgraded slony on the servers from 1.2.13 to 1.2.14 and downloaded the
> 1.2.14 creation scripts to my client.
>
> I created the new empty cluster and did a "upgrade node" on the old cluster,
> selecting the new one from the box.
>
> I get "Column not found in pgSet: indname"
>
> After dismissing that error dialog I get a second one.
>
> 'Error: relation "sl_log_1_idx1" already exists.'
Apologies for the late reply. I cannot reproduce the error you quote
here, but I have found another nasty issue which I'd like to bounce
off the other hackers...
It seems that current versions of the slony scripts embed the cluster
schema name in some of the functions. This means that if we create a
dummy cluster called 'upgrade' on a node, and then upgrade a cluster
called 'pgbench', the pgbench cluster references functions directly in
the _upgrade schema. This works fine (unless something gets changed in
the _upgrade schema of course), but barfs horribly when upgrading a
cluster on a different node which doesn't have the _upgrade schema.
The obvious first-thought solution is to search 'n' replace the schema
name when we generate the upgrade script. It doesn't take more than a
minute or two to think of ways that this may easily break however. The
second option is to only allow an upgrade from a cluster with the same
name, but that makes it kinda awkward to do the upgrade.
Any other ideas? (Magnus, Guillaume, others?)
--
Dave Page
EnterpriseDB UK: http://www.enterprisedb.com