Re: [PATCH 06/24] perf daemon: Add config file support

[Jiri Olsa]

On Sat, Feb 06, 2021 at 05:05:04PM +0900, Namhyung Kim wrote:
> On Fri, Feb 5, 2021 at 9:56 PM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> >
> > On Fri, Feb 05, 2021 at 09:14:54PM +0900, Namhyung Kim wrote:
> > > On Sun, Jan 31, 2021 at 8:49 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > > [SNIP]
> > > > @@ -263,9 +605,16 @@ static int __cmd_start(struct daemon *daemon, struct option parent_options[],
> > > >         signal(SIGTERM, sig_handler);
> > > >
> > > >         while (!done && !err) {
> > > > -               if (fdarray__poll(&fda, -1)) {
> > > > +               err = daemon__reconfig(daemon);
> > >
> > > I think it's confusing since you put the reconfig function here.
> > > What not split normal and reconfig passes?
> >
> > hum, not sure what's confusing in here? I've been known to
> > produce confusing code, but this one seems clear to me
> 
> Maybe it's because of the name.  I thought reconfig is a
> special case when it detects some change.  But you call
> it in the loop unconditionally.
> 
> >
> > >
> > > I mean something like below
> > >
> > >  __cmd_start()
> > > {
> > >     setup_server_config();
> > >     daemon__run();
> >
> > what's daemon__run? the daemon operates in the while loop below
> 
> I thought about starting the sessions in the config.
> 
> >
> > >
> > >     while (!done && !err) {
> > >         ...
> > >         if (reconfig) {
> > >             daemon__kill();
> >
> > you don't kill daemon for each reconfig change,
> > we detect changed sessions and kill/restart only them
> 
> Yep, we can make it that way.
> 
> >
> > >             setup_server_config();
> > >             daemon__reconfig();
> > >         }
> > >     }
> >
> >
> > so basically the current workflow is:
> >
> >         setup_server_config                                     <--- reads config file, prepares session objects
> >
> >         while (!done) {
> >                 daemon__reconfig                                <--- check session objects states and run/stop them
> 
> Hmm.. then how about rename it to daemon__handle_state()
> or daemon__do_loop() or something?

well it handles reconfig, so I don't think that there's
better name than daemon__reconfig ;-)

apart from handle_server_socket, all the other functions
we call after poll can change session state, so we need
to 'reconfig' sessions each time we do a loop

jirka