Re: Add relcreated (timestamp) column to pg_class catalog to record the time an object was created

On Sat, Apr 23, 2016 at 1:03 AM, Shulgin, Oleksandr <oleksandr.shulgin@zalando.de> wrote:

>I find yor lack of proper email quoting skills disturbing..

I am sorry you are disturbed, but thank you for pointing that out. I have revised my style to make you feel more comfortable.

>Yes, but that means that the timestamps must be part of the dump file, which means in turn they can be altered before the restore or--if implemented as separate commands like ALTER TABLE ... SET TIMESTAMP--at any point in time, so there is little to zero value in having the timestamps in the first place. This was already discussed in earlier threads.

No, when restoring to a new database, then by definition, object creation time is when they are loaded into the new database, which is essentially after the new database is created.

>> I also mentioned that this is already in the catalogs of Oracle and SQL Server.
>This sort of argument doesn't help the discussion: there should be a good reason to add the feature and merely pointing out that others already doing that is not a good reason, IMO.

Actually, it does. One of the reasons PostgreSQL is growing in popularity is that many companies are looking to switch away from Oracle and SQL Server and switch to open source PostgreSQL to save money. In doing so,  consideratopn of the compatibilty and features is of prime concern.

>If rogue users can create objects in your schema you have bigger problems than tables created outside of working hours
...
>Your better bet is to keep rogue users out: using GRANTs and pg_hba.conf.

So you are saying users never be allowed to have access at late hours, or per business needs create their own sort tables?

The president of a company that works late would have a problem with that.

Also, the same applies for developers in the development database, but sometimes they forget to drop experimental tables and/or document them.

The point is to be able to review without hindering.

>If you are going to review the logs anyway, why not just set log_statements=ddl and use logs

That is already done, but a SQL query is faster than a grep, awk and sort of a large log.

>No one is going to make even a "simple change" just for you to try

I am not asking for a change just for me, I am asking to consider the benefit to the PostgreSQL community.

>the idea fails in practice due to all the other things you didn't think about.

Really, please state how this will not work. I believe I have successfully countered all negative arguments so far.

>you can make such a change for yourself and run a patched version for a while if you want--no one can stop you here.

I have already stated I cannot. I am not a C coder and attempting to ALTER the pg_class system catalog causes an error.

I am hoping the PostgreSQL developers http://www.postgresql.org/community/contributors/ will review and either say yea or provide a specific reason for rejection.