[ClusterLabs] Is 20 seconds to complete redis switchover to be expected?

[Strahil Nikolov]

On April 1, 2020 7:56:15 AM GMT+03:00, steven prothero <steven at marimo-tech.com> wrote:
>Hello Strahil,
>
>I will increase the :
>
> redis monitoring interval:
>op monitor interval="1s"
>
>as you suggest and run the tests again
>
>I will also research about the power off... when I run it manually it
>seems very fast.  Perhaps the fence software has some settings I can
>change. I will investigate further.  I assume it is sending the OFF
>command and then waiting and then sending the STATUS command to make
>sure its off and then returns....
>
>This corosync.conf was setup from staff in the past... I wonder if the
>token & consensus should be changed and also "expected_votes: 2" since
>this is a 2 node cluster is that supposed to be 1? or maybe i can
>comment that out?
>
>I am looking at:
>https://linux.die.net/man/5/corosync.conf
>https://manpages.debian.org/testing/corosync/corosync.conf.5.en.html
>
>I will adjust the corosync.conf to match the example and run tests
>again.  Not sure what "amf" is....
>
>My corosync.conf looks like this:
>
>totem {
>        version: 2
>        token: 3000
>        token_retransmits_before_loss_const: 10
>        join: 60
>        consensus: 3600
>        vsftype: none
>        max_messages: 20
>        clear_node_high_bit: yes
>        secauth: off
>        threads: 0
>        rrp_mode: passive
>        interface {
>                ringnumber: 0
>                bindnetaddr: 10.24.x.y
>                mcastport: 5405
>        }
>        transport: udpu
>}
>nodelist {
>        node {
>                ring0_addr: 10.24.x.y
>                nodeid: 1
>                name: node1.com
>        }
>        node {
>                ring0_addr: 10.24.x.z
>                nodeid: 2
>                name: node2.com
>        }
>}
>amf {
>        mode: disabled
>}
>quorum {
>         provider: corosync_votequorum
>         expected_votes: 2
>         two_node: 1
>}
>aisexec {
>        user:   root
>        group:  root
>}
>logging {
>        fileline: off
>        to_stderr: yes
>        to_logfile: yes
>        to_syslog: no
>        syslog_facility: daemon
>        debug: off
>        logfile: /var/log/corosync/corosync.log
>        timestamp: on
>        logger_subsys {
>                subsys: AMF
>                debug: off
>                tags: enter|leave|trace1|trace2|trace3|trace4|trace6
>        }
>}
>
>Cheers
>
>Steven

The defaults for RHEL  is  a little bit lower , but it depwnds if those  are VMs .
I had issues with low token/consensus values when the VM is migrated from 1 host to another.
In your case, you are  using hardware so you can lower them to 1000/1200 but keep in mind that thia means  that any network disruption over 1s will affect your cluster.

The new versions of corosync support the 'two_node' option and you won't need  to specify the expected  votes  (as we  got only 2 nodes :)  ).

The monitor  interval could be set to 5s for example, but this  will increase the reaction time.

One silly question from my side... Can't redis be used as in a master-master setup ?

Best Regards,
Strahil Nikolov