mgr: die on bind() failure

[jcsp]

Previously, the daemon would get wedged if it
competed for the same port as another daemon
on the same host and lost.

Fixes: https://tracker.ceph.com/issues/23037
Signed-off-by: John Spray john.spray@redhat.com

[wjwithagen]

Why not do the same as with binding in MON and OSD.
Now systems that do not have systemd wll have to start wrapping mgr with something to keep autostarting it. :(

[jcsp]

@wjwithagen I thought the mon and osd were also exiting when they had a bind failure?

[wjwithagen]

@jcsp

I know there is, because FreeBSD and linux do not see eye to eye on socket reuse and the opetions used for that. So there is some code in
./src/msg/async/AsyncMessenger.cc:96
That part loops on bind() problems and tries several ports several times before giving up.

Otherwise restart MON and OSD would fail, due to socket still in use.

[jcsp]

Since ceph-mgr is using the same underlying messenger code, it would get the same ms_bind_retry_count though, so I still don't see what the difference is between what mgr is doing and what mon/osd are doing?

[wjwithagen]

@jcsp
Neither do I then. Both Simple as Async play this trick.

So this is more or less only possible when there is only one MGR port that is retried several times, after which the trying fails.
Or there are more ports and all of those were taken.
So it feels more like a configuration problem???

Let alone that this PR does end up doing what MON and OSD do in the end: abort and die.

[tchaikov]

http://pulpito.ceph.com/kchai-2018-02-27_10:33:49-rados-wip-kefu-testing-2018-02-27-1348-distro-basic-mira/