Re: VACUUM FULL name is very confusing to some people (or to most non expert people)

2018-02-26 9:56 GMT+01:00 Lætitia Avrot <laetitia(dot)avrot(at)gmail(dot)com>:

> Although VACUUM and VACUUM FULL is different, then result is same (depends
>> on detail level) - the data files are optimized for other processing. You
>> should to see a VACUUM like family of commands that does some data files
>> optimizations. VACUUM, VACUUM FULL, VACUUM FREEZE, VACUUM ANALYZE, ...
>> Personally I don't think so we need to implement new synonym command for
>> this case.
>>
>
> Here's how I understand what you wrote : "Each and every vacuum operations
> are different flavours for files optimization so it's legitimate to use
> similar names". I agree for VACUUM ANALYZE and VACUUM FREEZE that can be
> seen as options to do more things than a simple VACUUM.
>
> But I disagree for VACUUM FULL that isn't an option to do one more thing
> than VACUUM does. VACUUM FULL is a complete different process.
>
> Let's take an example:
> In a production server with average production load, if you're already
> running a VACUUM, you can change it to a VACUUM ANALYZE without many risks.
> But I wouldn't dare try a VACUUM FULL without pg_repack.
>

VACUUM and VACUUM ANALYZE does same VACUUM work.

The VACUUM FREEZE is different too.

>
>
>> Why you you cannot to say your students - "VACUUM FULL is like SHRINK in
>> SQL Server"?
>>
>
> I do explain that to my students but I'm not sure they memorize it,
> because they do have a lot to memorize in a training session.
>

Maybe the core of this issue is fact so VACUUM FULL, VACUUM FREEZE is
statements based on two keywords commands. Lazy VACUUM uses just keyword.
From this perspective the lazy VACUUM "should be" renamed, because it is
inconsistent - and some databases uses some names like OPTIMIZE table (and
it sound much more optimistic :)). I teach PostgreSQL more than ten years -
and I have not the problem with this topic - the hard part is explain of
VACCUM FREEZE - but VACUUM and VACUUM FULL can be explained simply (with
some detail reduction). VACUUM recuces MVCC costs, VACUUM FULL reduces
bloating. ANALYZE is orthogonal - calculate column statistic.

>
> I keep meeting customers to who I have to explain that a simple VACUUM
> doesn't rebuild indexes. Am I the only one facing that problem ?
>

simple VACUUM (lazy VACUUM does a tasks that don't needs aggressive locks)
- rebuild indexes needs strong locks.

I agree, it is not pretty clean, because VACUUM FULL share some work with
REINDEX, but introduction new command change nothing.

>
>
>> Regards
>>
>> Pavel
>>
>
> Regards
>
> Lætitia
>