[ClusterLabs] RRP is broken ... in what ways?

[Jan Pokorný]

On 10/04/19 15:31 +0200, Jan Pokorný wrote:
> On 08/04/19 19:08 +0200, Jan Friesse wrote:
>> Anyway. RRP is broken,
> 
> Since this is repeatedly rehashed on this list without summarizing
> *what* is that much broken[1] while serving (rather well, I think)
> last 8+ years, perhaps it would deserve some wider dissemination
> incl. *why* is that.
> 
> Closest thing to come to mind is this, though I don't think
> it's a well known reference:
> https://corosync.github.io/corosync/doc/presentations/2017-Kronosnet-The-new-face-of-corosync-communications.pdf#page=3
> 
> In addition, there are minor notes here and there to be found, e.g.,
> on the topic of the networks being presumably of comparable
> characteristics with RRP [2,3]:
> 
>> It should probably be documented somewhere but may not be, with
>> redundant ring, both rings must be of approximately the same
>> performance.
> 
>> The protocol is designed to limit to the speed of the slowest ring
>> - perhaps this is not working as intended.
> 
> [1] https://lists.clusterlabs.org/pipermail/users/2019-March/025491.html
> [2] https://discuss.corosync.narkive.com/jUynoGxv/totem-implementation-is-unreliable-ring-faulty-and-retransmit-list#post2
> [3] https://lists.clusterlabs.org/pipermail/pacemaker/2011-August/058536.html
> 
> Can you perhaps give a definite answer here to arising "what" and "why",
> also perhaps to provide a credible incentive to upgrade out of something
> just sketchily referred to as "broken", please?

No conclusive summary on the subject (so far taken rather for a fact
here, where I guess corosync experts are a vast minority) to afford us
the liberty of educated choice regarding the upgrade beyond allegedly
broken RRP?

-- 
Jan (Poki)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <https://lists.clusterlabs.org/pipermail/users/attachments/20190417/b75a82c4/attachment.sig>