cluster and replication doubt

[PaulVM]

I can't understand why replication stops if I disconnect the cable in the "local" network while it continue if I disconnect the "internet" network.
I supposed it uses the "local" link for replication and the alternate link if the "local" fail.

I have this test cluster (3 nodes, 2 nic):
corosync.conf: logging { debug: off to_syslog: yes } nodelist { node { name: pve08 nodeid: 1 quorum_votes: 1 ring0_addr: 172.16.0.8 ring1_addr: 5.5.7.8 } node { name: pve09 nodeid: 2 quorum_votes: 1 ring0_addr: 172.16.0.9 ring1_addr: 5.5.7.9 } node { name: pve10 nodeid: 3 quorum_votes: 1 ring0_addr: 172.16.0.10 ring1_addr: 5.5.7.10 } } quorum { provider: corosync_votequorum } totem { cluster_name: pve23 config_version: 3 interface { linknumber: 0 } interface { linknumber: 1 } ip_version: ipv4-6 link_mode: passive secauth: on version: 2 }

What am I missing?
Thanks, P.

 

[LnxBil]

cluster communication and replication are not connected. AFAIK, replication uses the hostname of your other node to connect and has most probably only one IP accociated with it in your DNS or /etc/hosts file, so it'll run over ONE network not both. If the non-cluster network fails, it'll also fail the replication.

 

[UdoB]

What am I missing?

Look into "Datacenter --> Options --> Migrations Settings" to configure a specific network to use.

Best regards

 

[PaulVM]

So, migration and replication only is done using 1 network by default (the default WAN network)?
And I can only optionalluy chose which network use?
I supposed that confugiring multiple networks in the cluster configuration automatically it uses them as needed.

Thanks, P.

 

[LnxBil]

Normally you put network high availability one layer down, so that you only see only one network on top and the underlying network configuration will take care of everything else. This can be done e.g. via bonding. Most setups I encounter nowadays are MLAG-based LACP trunks with cluster and/or storage networking as VLAN on top of that.