Adam wrote:

>> I work at a project related to a possibility to have the index
>> statistics auto-updated and humbly I ask your opinion.
>>
>
>
> m. Th.,
>
> Just a thought, but a pie chart doesn't really tell you much. Everyone
> who doesn't understand 100% of the issues is going to click somewhere
> between indifferent and very useful.
>

That's why I put the last option: "What is this about? (or I don't know
what 'statistics' are)" - and IMHO this is the most dangerous, that now
15% from the users chosen this... here is the problem IMHO.

> My (limited) understanding is as follows:
>
> The job of the optimiser is to analyse the query and determine the
> least cost method of solving it, making use of any available index.
> The usefulness of an index or a combination of indices is determined
> by their true selectivity. The statistics tell the optimiser about the
> selectivity of the index.
>
> As new records are inserted, and old records are updated and deleted,
> the true selectivity of an index may change. Consider an indexed field
> with unique values (although not enforced by constraint). If I ran the
> query 'update table set field=1;', the true selectivity of the index
> would (after garbage collection etc) be poor, but the statistics
> provided to the optimiser would indicate it was a good one to use.
>

Yes.

> Yet there are people selecting that they are against the feature. I
> would like to know their reasons? Performance concerns? Something else?
>

You're right. But until now only 2 voted here. If there are more then is
very "interesting", because, in fact, this is a trap. Normally, anyone
who doesn't want this feature should choose " *I PREFER* my own custom
way to update...". If there are more then perhaps we'll build another
poll to see. But then we need some ready-made answers. I know already
two: 1. Performance 2. Is not worth bother - I know how to do it by
hand. (but this one I put it already inside...)

Someone knows other reasons?

hth,

m. th.

>
>