[pve-devel] [RFC cluster 5/6] node join: use new corosync link parameters

[Thomas Lamprecht]

On 5/29/19 10:50 AM, Fabian Grünbichler wrote:
> I wonder whether we should support clusters without link0 as well? might
> happen easily:
> 
> cluster only has link0 configured
> network topology change, add new link1
> deconfigure link0
> 
> cluster joining no longer possible without specifying a bogus link0
> address and manually cleaning up corosync.conf afterwards?

man corosync says:

> linknumber
>  This  specifies  the link number for the interface.  When using the knet
>  protocol,  each  interface  should  specify  separate  link  numbers  to
>  uniquely  identify to the membership protocol which interface to use for
>  which link.  The linknumber must start at 0. For UDP the only  supported
>  linknumber is 0.

While a live addition seems not to problematic, a full stop start of
corosync shows:
> Jun 11 14:10:44 dev6 corosync[564072]:   [TOTEM ] knet_link_set_ping_timers for nodeid 1, link 0 failed: Invalid argument (22)
> Jun 11 14:10:44 dev6 corosync[564072]:   [TOTEM ] knet_link_set_pong_count for nodeid 1, link 0 failed: Invalid argument (22)

So, as the man page specifies it and corosync behaviour is a bit questionable,
and at least not error free, I'm not really sure the "no special treatment for
link 0" holds, could get worse in the future - as they specifically ..

I do not get above errors when having a link 0 configured..

> 
> in general, we probably want to remove any special handling of link0/1,
> and just set fallback defaults on link0 if no other link is specified..
>