Re: pgsql start implies initdb = bad

[David Brownlee <abs%netbsd.org@localhost>]

2009/11/25 Jan Danielsson <jan.m.danielsson%gmail.com@localhost>:
> David Brownlee wrote:
> [---]
>> Just to clarify - the issue here is the psql data directory being on a
>> soft or interruptable nfs mount, a restart when the mount is down
>> causing initdb to run on the local filesystem, and then the mount
>> coming back and psql scribbling 'new db data' over the production
>> database?
>
> Â That's the core of the issue, yes. Information about "new" data gets
> read into memory. "Old" database appears where "new" database was when
> new mount completes. Database does confusing things to "Old" database
> because of partial "new" database information.

Thats what I thought - a particularly painful error mode that... :/

>> Running a database on an interruptable or soft mounted filesystem
>> manages to make using it on a normal nfs filesystem seem quite
>> sensible :)
>
> Â This is what happened: I noticed that template1 doesn't default to
> unicode, so I asked on a postgresql IRC channel if something has changed
> recently (since I had used utf-8 and template1 previously). After they
> asked a few questions I replied that initdb is automatically run as a
> part of "start" if no database has been initialized. Then came the,
> rather unexpected - I must admit, "ARE YOU FRIGGIN' INSANE?!", etc.
> Seeing as these guys have seen much more database experiences than have,
> and that they have more than one documented case were exactly what is
> done in default pkgsrc resulted in disaster, I thought I'd bring it up.
> I have done so. I leave it up to the people who know more about these
> things to determine if it's relevant. I'm not going to try to force the
> issue.

Its certainly an interesting case :)

The epic fail case is only going to be hit when someone has configured
their system in a hideously broken fashion anyway, but I can imagine
situations where people would prefer 'no database' to 'empty database'
when a system 'loses' its psql data dir, so certainly an option to the
rc.d file to disable the auto initdb (maybe by just setting something
in rc.conf) might be of interest.