> Mark Woodward wrote:
>> I know this is a kind of stupid question, but postgresql does not
>> behave well when the system changes in a major way without at least
>> an analyze. There must be something that can be done to protect the
>> casual user (or busy sometimes absent minded developer) from these
>> dangerous edge conditions?
>
> autovacuum
That's a good simple answer, sure, but it is no different than "run
analyze," they are obvious when you know the problems, but not so when you
don't are focusing on something else.
I didn't see the "problem" because I didn't suspect HassHagg would behave
so badly, who would?
One of my biggest problems with Oracle is that there are so many ways that
it can fail. One can argue that the DBA should "know what they are doing,"
and it is a good argument, but there is a diffeence between knowing the
findimentals of server design, query design, parallel processing, I/O
bandwidth, etc. and knowing the esoterica of a particular platform. One
tends to acquire the essoterica as needed.
What I discovered with PostgreSQL was a failure. It had run away memory
cconsuption, this is bad behavior. On Linux it was killed, while I don't
want to have that discussion, it is a real world fact that saying "turn
OOM off," is not acceptable.
If PostgreSQL exhibits bad behavior, it is PostgreSQL's problem.
My question was based on an observation that ANALYZE and VACUUM are
nessisary, both for different reasons. The system or tools must be able to
detect substantial changes in the database and at least run analyze if
failing to do so would cause PostgreSQL to fail badly.