mon:Re join the Monitor alone to form a cluster when it's rank=0

[renhwztetecs]

Increase whether there is Monitor extra judgment. when there is Monitor information in the extra_probe_peers, it does not enter the win_standalone_election() process

Fixes: #14263
Signed-off-by: huanwen ren ren.huanwen@zte.com.cn

[renhwztetecs]

@jecluis
Increase whether there is Monitor extra judgment. when there is Monitor information in the extra_probe_peers, it does not enter the win_standalone_election() process

I'm not sure if this change is the most appropriate, please help to retrieve
thank you

[renhwztetecs]

I tested a variety of scenes are no problem, and can be modified bug
Http://tracker.ceph.com/issues/14263

[jecluis]

I need to take a better look at this - I'll do it later today.

[renhwztetecs]

@jdurgin
thank you for your attention

[gregsfortytwo]

I'm not sure what this is trying to do. extra_probe_peers get added to the monmap when we're setting up a new one (that is, when we haven't already been in a quorum), so monmap->size()==1 already implies there are no extra_probe_peers when starting up for the first time. If we're starting up after having already been in a quorum, the extra_probe_peers don't matter as we already have knowledge of the system's genuine state.

[renhwztetecs]

@gregsfortytwo
verification according to http://tracker.ceph.com/issues/14263 process:
At the start of the ceph240 ceph-mon found monmap->size () = = 1 when set mon initial members = "ceph240" in ceph.conf, and "monitor::handle probe probe()" function to obtain cluster has the monitor data, which can get extra_probe_peers= "ceph243" "ceph244"
So monitor::bootstrap process will enter：
if (monmap->size() == 1 && rank == 0){
win_standalone_election();
return;
}
note：ceph240 rank = 0

[renhwztetecs]

@gregsfortytwo
remove a monitor in the cluster, need to call extra_probe_peers.erase to remove the corresponding monitor, to prevent the cluster in only one node, can not enter this process and abnormal：
if (monmap->size() == 1 && rank == 0 && extra_probe_peers.empty()) {
win_standalone_election();
return;
}

[gregsfortytwo]

Yeah, in that ticket you set up mon240 with an existing monmap that only contains itself. That's expected behavior; you needed to tell it to join up with the others. See http://docs.ceph.com/docs/master/rados/operations/add-or-rm-mons/ for instructions.

[gregsfortytwo]

Wait, I misread the steps you followed.

Still, I think you've misdiagnosed the problem. The new monitor is somehow losing track of the existing monitors which should be in its map.

[renhwztetecs]

@gregsfortytwo @liewegas
I don't agree with this conclusion, according to the http://tracker.ceph.com/issues/14263 problem described in the construction, and add the following printed in the code:

Monitor::bootstrap()
{
......
dout(10) << " monmap is " << *monmap << dendl;
dout(10) << " extra probe peers " << extra_probe_peers << dendl;

// singleton monitor?
if (monmap->size() == 1 && rank == 0) {
win_standalone_election();
return;
}
......
}

[renhwztetecs]

@ gregsfortytwo
when a new Monitor to join the cluster, through handle_probe_probe () detects the existing Monitor, but if the handle_probe_probe () before testing, judgment entered win_standalone_election() process, then there will be a problem