[PATCH]: Inconsistent recmaster during election.

[Martin Schwenke]

On Tue, 12 Jan 2016 17:02:25 +1100, Martin Schwenke <martin at meltin.net>
wrote:

> OK.  Still thinking about answers to this...  :-)

I'm still trying to find ways around this.

I thought about modifying 49.winbind so that it no longer does
"smbcontrol winbindd ip-dropped ..." when CTDB is shutting down.
However, the problem is not that CTDB on the current node is shutting
down, it is that a node (any node!) (shutting down) has triggered a
recovery so smbcontrol is unable to connect to CTDB as a client.

We need to figure out if some changes to winbind mean that this
particular smbcontrol is no longer required.

Alternatively, we need to understand how important it is.  We could
just run it under the timeout(1) command:

  timeout 5 smbcontrol winbindd ip-dropped $ip >/dev/null 2>/dev/null

The redirections there and lack of anything looking at the exit code
suggests that we don't care a lot whether it works or not.  So, while
not a well thought out solution, the above seems to be a reasonable
workaround.

Do you want to try the above and see if your problems magically go
away?  :-)

peace & happiness,
martin