Hi,
On Thu, Apr 10, 2008 at 12:05:18PM +0200, Bernd Schubert wrote:
> On Wednesday 09 April 2008 22:20:16 Lars Marowsky-Bree wrote:
> > On 2008-04-09T20:26:02, Bernd Schubert <[EMAIL PROTECTED]> wrote:
> > > I still think there is another bug in heartbeat, though. There is simply
> > > no reason for heartbeat to wait $deadtime on initial startup of the
> > > heartbeat services, when it knows all heartbeat nodes are are up.
> > > If I at least could manually force it to online the nodes, I would have
> > > no problem with an initial-deadtime == deadtime.
> >
> > That _should_ work, indeed. If both sides are up, it should proceed
> > immediately. Do you have autojoin enabled? Which version?
>
> This is 2.1.2, but after quickly grepping through the sources, I think this
> problem is also in tip. There is simply presently no way to mark a node
> online until the initial deadtime is over:
>
>
> polled_input_dispatch:
> check_for_timeouts();
>
> check_comm_isup();
>
>
>
>
>
> /* See if any nodes or links have timed out */
> static void
> check_for_timeouts(void)
> [...]
> if (heartbeat_comm_state != COMM_LINKSUP) {
> /*
> * Compute alternative dead_ticks value for very first
> * dead interval.
> *
> * We do this because for some unknown reason
> * sometimes the network is slow to start working.
> * Experience indicates that 30 seconds is generally
> * enough. It would be nice to have a better way to
> * detect that the network isn't really working, but
> * I don't know any easy way.
> * Patches are being accepted ;-)
> */
> dead_ticks
> = msto_longclock(config->initial_deadtime_ms);
>
> [...]
> mark_node_dead(hip);
>
> Then in
>
> static void
> check_comm_isup(void)
> {
> struct node_info * hip;
> int j;
> int heardfromcount = 0;
>
>
> if (heartbeat_comm_state == COMM_LINKSUP) {
> return;
> }
>
> if (config->rtjoinconfig != HB_JOIN_NONE
> && !init_deadtime_passed){
> return;
> }
>
Thanks for the analysis. I'll have to check the history of the
code.
Cheers,
Dejan
>
> Thanks,
> Bernd
>
> --
> Bernd Schubert
> Q-Leap Networks GmbH
> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
