maria-developers team mailing list archive

[Sergey Vojtovich <svoj@xxxxxxxxxxx>]

Hi Sergei,

On Thu, Nov 26, 2015 at 02:55:49PM +0100, Sergei Golubchik wrote:
> Hi, Sergey!
> 
> On Nov 26, Sergey Vojtovich wrote:
> > > > +
> > > > +  if ((user= my_strdup(user_host_buff, MYF(0))) &&
> > > > +      !my_atomic_casptr((void **) &shutdown_user,
> > > > +                        (void **) &expected_shutdown_user, user))
> > > > +    my_free(user);
> > > > +}
> > > 
> > > Interesting. Why is that?
> > It supposed to be safe concurrent shutdown.
> 
> Okay, but please add a comment about it.
Ok.

> 
> > > > @@ -1888,7 +1912,13 @@ static void __cdecl kill_server(int sig_ptr)
> > > >    if (sig != 0) // 0 is not a valid signal number
> > > >      my_sigset(sig, SIG_IGN);                    /* purify inspected */
> > > >    if (sig == MYSQL_KILL_SIGNAL || sig == 0)
> > > > -    sql_print_information(ER_DEFAULT(ER_NORMAL_SHUTDOWN),my_progname);
> > > > +  {
> > > > +    char *user= (char *) my_atomic_loadptr((void**) &shutdown_user);
> > > > +    sql_print_information(ER_DEFAULT(ER_NORMAL_SHUTDOWN), my_progname,
> > > > +                          user ? user : "unknown");
> > > 
> > > can it, really, be "unknown" here? when?
> > Yes, in a few cases. E.g. killed by signal. 
> 
> Oh, right, sorry. I thought that if() filters that out (because signals
> have a special message ER_DEFAULT(ER_GOT_SIGNAL) below).
> 
> Any other cases? Otherwise I'd suggest
> 
> -               user ? user : "unknown");
> +               user ? user : STRINGIFY_ARG(MYSQL_KILL_SIGNAL))
No user info in following cases:
- if listening socket is closed and SIGNALS_DONT_BREAK_READ is defined (which
  is never seem to be defined), kill_server is called with MYSQL_KILL_SIGNAL
- if signal SIG{TERM|QUIT|KILL} received and USE_ONE_SIGNAL_HAND defined (is it
  guaranteed to be defined properly, kill_server is called with 0
- special handling of SIGINT
- wsrep applier thread may call kill_server() without thd, but that's probably
  fixable

Regards,
Sergey