[RFC] Performance improvements

[Swen Schillig]

Hi Volker

Thanks for having a look.

On Wed, 2018-06-20 at 11:22 +0200, Volker Lendecke wrote:
> On Mon, Jun 18, 2018 at 10:04:03AM +0200, Swen Schillig via samba-
> technical wrote:
> > Attached is the patchset in question.
> > Even though the set consists of 10 patches I wouldn't call them all
> > micro-patches..but they're as small as possible I guess.
> 
> 2/10 is an obvious RB+ by me.
> 
> 10/10 is really, really fishy. We should never handle more than one
> event from a single tevent_loop_once. Not a formal NACK, but this
> needs thorough review by people more familiar with the core of ctdb's
> request handling.

After our conversation this morning about this, I was checking again.

Actually I'm not handling more than one event, if you look again you
will see that I'm just NOT creating an additional event by calling
tevent_schedule_immediate for the data already in the buffer 
...instead I process the data right away.

So I think this is the tiny but important difference.

In addition, within CTDB the only other place where immediate events
are scheduled are for killing the queue... therefore I don't believe
we're screwing any order here.

Bu as you said... hopefully some more familiar people can have an
additional look.

Cheers Swen