Manuel Federanko <[email protected]> writes:

> The partition that gets a vote is not randomly chosen, but depends on
> the configuration of the qdevice in corosync.conf.
>
> A partner asked about predictable behavior for vote casting by the
> QDevice. The documentation of corosync and ours didn't align.
>
> Setting tie_breaker influences how a partition is chosen for votes,
> which was tested with a 2-node test-cluster.
>
> Signed-off-by: Manuel Federanko <[email protected]>
> ---
>  pvecm.adoc | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/pvecm.adoc b/pvecm.adoc
> index 0ed1bd2..899a4de 100644
> --- a/pvecm.adoc
> +++ b/pvecm.adoc
> @@ -1266,8 +1266,10 @@ Tie Breaking
>  ^^^^^^^^^^^^
>  
>  In case of a tie, where two same-sized cluster partitions cannot see each 
> other
> -but can see the QDevice, the QDevice chooses one of those partitions randomly
> -and provides a vote to it.
> +but can see the QDevice, the QDevice chooses the partition which has
> +the lowest node id and provides a vote to it. This behavior can be tuned with
> +the configuration option `tie_breaker` (see `man corosync-qdevice` for more
> +information) and requires a restart of `corosync-qdevice.service` on all 
> nodes.
>  
>  Possible Negative Implications
>  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

I would personally only mention the man page so to avoid running out of
sync, but this looks good to me as-is.

Reviewed-by: Maximiliano Sandoval <[email protected]>

-- 
Maximiliano



Reply via email to