[Zope] Interest in continued Oracle support in Zope

[Anthony Baxter]

On Wed, 1 Sep 2004 21:26:07 +0200, Dieter Maurer <dieter at handshake.de> wrote:
> When I had to work with Oracle 3 years ago,
> I have been *VERY* disappointed (this was Oracle 8i).
> Despite its high costs (100.000 USD), it was almost
> unfunctional:

I agree that Oracle is extremely expensive, but there's a few key advantages:

 - It's designed for large systems and has very good tools for
managing databases.
 - It's easy to find DBAs that are extremely knowledgable about Oracle. 

The second, for me, is key. Software developers should not also be
trying to manage an RDBMS.

>   *  memory corruption inside the Oracle client libraries
>      brought our Zope down within minutes of use

Haven't seen this since we stopped using Oracle on Linux, and even
then, it was far far less frequent than that - maybe once a month or
so.

>   *  the Oracle server died often, non-deterministically, after
>      some amount of work

Haven't seen that with Oracle on Solaris, ever. On Linux, we'd _very_ _very_
infrequently get strange things going on - every couple of months.

>   *  the Oracle client library caught SIGCHILD making
>      "system" and "popen" unreliable -- this could be fixed
>      with a special configuration option, deeply hidden
>      in the tons of Oracle documentation

Can't see I've ever seen that.

>   *  full-text reindexing after large imports often crashed
>      non-deterministically

Again, haven't seen that, ever.

>   *  upgrading from one Oracle version to the next Oracle
>      subversion was a nightmare

See, I've found the opposite - one of the strengths of Oracle is that
this stuff is _well_ understood and documented. We do upgrades,
sidegrades, point fixes and the like, all the time. Well, when I say
"we", I mean "our DBAs". I wouldn't let them write code, and in return
I don't expect to manage the database. They're different jobs,
requiring different skills.

>   *  some combinations of relational and full text subqueries
>      let Oracle forget about its indexes.
>      For example: while a query "Q1 and Q2" took seconds
>      (Oracle recognized that it had indexes) the query
>      "Q1 or Q2" (with the same "Q1" and "Q2") took days (!)
>      (because Oracle had forgotten about the indexes and used
>      full table scans -- which is a bad idea with hundread of
>      millions of records and hundreds of gigabyte of data).

Did you use the "explain plan" functionality? Was this with the
rule-based or cost-based optimiser?

>      The Oracle support was unable to fix this problem within
>      half a year. Oracle recommended to avoid the combination
>      of relational and full text subqueries !

Oracle's support can be annoying, yes. Fortunately, I just let the
DBAs deal with it - they know how to get the information from Oracle's
TAC.

> Now, I use Postgres -- and I am *MUCH* happier:
> I have at most 1 per cent of the problems I have had with Oracle.
> And this with no costs and no support
> (but the amount of data is smaller now and there are no full text
> queries (that were responsible for most of Oracles problems)).

I like postgres, but my main problem is finding DBAs with sufficient
PG knowledge. This, to me, is a key thing. If the business _depends_
on the database, you really really need a high clue level in the
people managing it.

Having said all that, and lest I seem like a reflexive Oracle
supporter, it often drives me screaming with frustration up a wall.
But I have the same sorts of issues when using postgres - usually it's
the query optimiser doing something stupid, and you can examine the
optimiser and figure out what's going on. Usually it's just an index
hint that's needed.

And Oracle's approach of "Oh, that _obvious_ piece of missing
functionality? You need enterprise edition for that" pisses me off, no
end. EE costs an utterly utterly indefensible amount of money.

Anthony