Enabling evs.version=1 problem for WAN optimization

[Donovan Sydow]

Has anyone had any luck using the evs version 1 settings? I have tried turning this on a single node through the my.cnf and it is unable to join an existing cluster.  

Trying to follow this documentation here:

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

I have tried:

wsrep_provider_options = "evs.version=1;"

and

wsrep_provider_options = "evs.version=1"

MySQL will start just fine but the node upgraded to evs.version=1 through either entry is unable to join the cluster with log entries like this:

2015-02-09 13:20:31 123233 [Note] WSREP: EVS version downgrade 1 -> 0

2015-02-09 13:20:31 123233 [Note] WSREP: declaring 11c451ca at tcp://10.0.1.2:4567 stable

2015-02-09 13:20:31 123233 [Note] WSREP: declaring 6bede4f3 at tcp://10.0.1.12:4567 stable

2015-02-09 13:20:31 123233 [Note] WSREP: declaring 7bd85abe at tcp://10.2.1.23:4567 stable

2015-02-09 13:20:31 123233 [Note] WSREP: PC protocol downgrade 1 -> 0

2015-02-09 13:20:32 123233 [Note] WSREP: Node 11c451ca state prim

2015-02-09 13:20:37 123233 [Note] WSREP: EVS version upgrade 0 -> 1

2015-02-09 13:20:37 123233 [Note] WSREP: PC protocol upgrade 0 -> 1

2015-02-09 13:20:37 123233 [Note] WSREP: view(view_id(NON_PRIM,547a3bdb,4151) memb {

547a3bdb,0

} joined {

} left {

} partitioned {

11c451ca,0

60f09b2c,0

6bede4f3,0

7bd85abe,0

})

2015-02-09 13:20:37 123233 [Note] WSREP: New COMPONENT: primary = no, bootstrap = no, my_idx = 0, memb_num = 1

2015-02-09 13:20:37 123233 [Note] WSREP: Flow-control interval: [16, 16]

2015-02-09 13:20:37 123233 [Note] WSREP: Received NON-PRIMARY.

2015-02-09 13:20:37 123233 [Note] WSREP: New cluster view: global state: 7f5b0b91-169f-11e4-9050-6234cfd53603:2695, view# -1: non-Primary, number of nodes: 1, my index: 0, protocol version -1

I even went so far as to enable this via the mysql console, which put the entire cluster in a non-primary state.

All other nodes are on the same mysql version, as well as the garbd arbiter. Anyone know if I am jumping the gun by turning this on in the Percona release, and/or know if I should just wait for the new 3.9 release to be integrated. Other suggestions on enabling the new evs version would be appreciated. 

Ultimately my goal was to use evs version 1 to use the WAN replication tips here:

http://galeracluster.com/documentation-webpages/configurationtips.html#wan-replication

It isn't clear if I really need version 1, or if I can use these with version 0 as well.

Cheers,

Donovan

[alexey.y...@galeracluster.com]

Hi Donovan,

At the moment all nodes in the cluster must speak the same protocol.

Regards,
Alex

[Donovan Sydow]

Alexey,

Thanks for the response!

Is there a way to turn it on one at a time like it is documented? Or is it currently a downtime implementation, where I'll need to shutdown the entire cluster, and bootstrap it from a single node with evs version 1 enabled?

Any idea if the WAN optimizations should work without evs version 1?

Cheers,

Donovan

[Teemu Ollakka]

Donovan,

On Monday, February 9, 2015 at 11:59:10 PM UTC+2, Donovan Sydow wrote:

Is there a way to turn it on one at a time like it is documented? Or is it currently a downtime implementation, where I'll need to shutdown the entire cluster, and bootstrap it from a single node with evs version 1 enabled?

Rolling upgrade for evs.version = 1 was possible at some point assuming that all of the nodes in the cluster had been upgraded. What you are observing is likely a bug.

 

Any idea if the WAN optimizations should work without evs version 1?

Most of the WAN optimizations should work with evs.version = 0. Version bump to 1 is necessary only for autoeviction to work.

- Teemu