How to monitor arbiter status within galera cluster

[Steffen Boehme]

Hi folks,

I wonder how it would be possible to just do a simple monitoring on the status of a galera cluster arbiter.

We run Percona XtraDB Cluster 5.7 and have some setups with 1 arbiter and 2 data nodes.

I wonder that in the data nodes, where I'm able to fetch status and variables, is no hint about the arbiter and also not the availability of this arbiter node.

I only found, that the wsrep_cluster_size is counted to 3 where the wsrep_incoming_addresses status shows only 2 addresses and one additional comma, like this:

wsrep_incoming_addresses        db97.prv.fra.sprd.net:5719,db98.prv.fra.sprd.net:5719,

sometimes also like this

wsrep_incoming_addresses        db97.prv.fra.sprd.net:5719,,db98.prv.fra.sprd.net:5719

or like this

wsrep_incoming_addresses        ,db97.prv.fra.sprd.net:5719,db98.prv.fra.sprd.net:5719

Well, I could use this ... but this looks not really reliable for me and I wonder, if there is a nice way to monitor arbiter status from within the data nodes.

Of course I could (and I will) monitor the arbiter process itself in another way, but always to try to use the look "from inside the cluster" to get a state about the stability and redundancy and safety against split brains.

So just to check, if the are 3+ nodes and a odd number is also possible ... but still wondering, why I cannot extract some info about a connected arbiter from the running nodes ...

Any ideas how to deal with this?

Thanks in advance

Steffen

[Steven Hancz]

Hi Steffen 

Did you even find an answer to your question? 

I too want to find out the arbiter hostname and port from the cluster. I can't find any variable or parameter that can show me that information.

Steven

[Steffen Boehme]

Hey Steven,

sorry for the late answer ... did not realize your question ...

No, I did not find.

Right now I'm playing around with using the garbd as hot backup method like explained on the galera website but even this works not for me, since the cluster I connect to gets no information about the arbiter hostname or any other information ... it a bit similar ... i found no possibility to provide this information to the cluster node using the arbiter config and parameters and so, the node does not know the hostname and does not know where to stream the data.

Its obscur ... I cant understand, why this information is not send to the cluster (nodes) ...

But it looks like the cluster just don't gets this information ...

I wonder how the arbiter mechanism works then ... can only estimate that its like a push mechanism ... the arbiter send keep alives or something like this.

I will stay on this, since I'm interested in and I also would like to have a working description how to use hot backups over the garbd binary from galera (percona?) ...

You know this page:

http://galeracluster.com/documentation-webpages/backingupthecluster.html

?

Greets

Steffen