talloc: Other minor issues/queries

[Jeremy Allison]

On Thu, Oct 22, 2020 at 11:31:33AM +0200, Ralph Boehme wrote:
> Am 10/22/20 um 10:57 AM schrieb Stefan Metzmacher via samba-technical:
> > Maybe we could abstract the low level allocation (malloc,realloc,free) in talloc and just
> > implement the other allocation schemes below the main talloc logic.
> > 
> > In order to be able to handle much higher IOPS in our SMB server, we need to find ways
> > to optimize our talloc/tevent_req handling.
> > 
> > These are some ideas, but it will be a long road to get there, so I guess we need to
> > find incremental mini steps...
> 
> stuffin all this extra stuff into talloc() is going to make to add a lot
> of extra complexity that we'll have to maintain in the future as a team
> and ideally not just you, Jeremy and Volker. Therefor I'm asking to
> please carefully consider if this is worth it.

Well my current goal is to reduce the complexity by
removing the memlimit stuff, and adding a flag to
stop a talloc_pool being able to expand out into
memory outside the pool.

Those are the two initial proposals on the table
IMHO and the complexity removed by getting rid
of memlimit more than makes up for the small extra
complexity by adding the talloc_pool flag.

The rest is "blue-sky" thinking IMHO :-).

Are you on board with the first two steps Ralph ?

The other stuff comes later (if at all :-).

Jeremy.