Re: Should I care about this error "failed to link /usr/bin/psql [...] exists and it is not a symlink"?

Hi Tom,

El 3/12/19 a las 17:10, Tom Lane escribió:
> Ekaterina Amez <ekaterina(dot)amez(at)zunibal(dot)com> writes:
>> El 3/12/19 a las 16:14, Tom Lane escribió:
>>> I believe the issue here is that you've got Red Hat's postgresql packages
>>> installed,
>> As I've mentioned, this Postgres version comes with CentOs installation,
>> so I guess you're right.
>>> and they're not set up to participate in alternatives
>>> switching. I'd suggest removing those packages
>> By "removing those packages" you mean uninstall v9.2? This is a test
>> computer and it's not a problem, but I'm afraid this is going to happen
>> also in production server, by doing this won't be my cluster be unavailable?
> Um, wait a minute. Are you actually running a server with that 9.2
> installation? And the goal is to upgrade that to 9.6 using PGDG
> packages? If so, it's going to take some experimentation. I'm sure
> that whatever upgrade script the PGDG packages provide isn't going
> to work; you'll have to run pg_upgrade by hand. It's probably possible,
> but bleats from yum are the least of your worries.
>
> Once you do get the data converted over, probably the cleanest way to
> resolve the installation conflict is to remove both sets of packages
> and then reinstall only the PGDG ones.

I think it's better to explain my goal...

We have a Production server that runs PG9.2 over CentOs 7. The cluster
in this server contains several databases, one of them is almost 350GB
and growing. This database is being updated almost continuously, mainly
by a cron script that runs every 5 minutes and gathers data from other
sources and loads it in the db. We want to upgrade this server to 9.6
(it's the target version of other upgrades made in other servers) and,
after asking here, I've concluded that it is best to create a hot
standby replica of this cluster and, when data is synchronized and
tested that everything is fine, stop master and make a pg_upgrade with
--link option. In case upgrade goes wrong we have slave server that can
be promoted to master. One minor detail is we're running out of space in
this server.

I've upgraded a couple of servers before (from 8.4 to 9.6), and I've
done it via pg_upgrade without --link. I guess this is what you refer
when saying "pg_upgrade by hand", so there's no problem with this part.

>> So what's happening is that I have installed a RH version of Postgres
>> instead an official one? Am I right?
> That would depend on what you want to call "official" ;-). But Red Hat's
> postgres packages use a much different file layout than the PGDG packages,
> so they don't interoperate terribly well.
>
> regards, tom lane

I'm going to test this: having a machine with CentOs7 and Red Hat's
postgres v9.2 packages, move /usr/bin files listed in the previous log
to another locations, create symlinks to the new location in /usr/bin,
and finally install install PGDG 9.6 packages over this configuration.
We're not planning to run both versions at the same time.

When you say "Red Hat's postgres packages use a much different file
layout than the PGDG packages, so they don't interoperate terribly
well.", I guess you're talking about the installation, not the structure
of the database. If I'm right, there shouldn't be problems with
pg_upgrade (I'm worried about data corruption as I'm not sure how to
detect it).

Would it be possible to install another server with same CentOS but with
PGDG packages of postgres 9.2 and create the replica to this server and
replaces this one with main server when both are completely synchronized?

Kind regards,

Ekaterina

PS=> If PGDG is not the "official" what is it? :D