Don't set any keepalived master on first run #50
Conversation
During the first run, the fact gluster_vrrp_fqdns is empty, leading to all nodes to be set as keepalived master. This could cause race conditions on the second run when gluster volumes are created because every node has the VIP, and it is an invalid state anyway.
|
@jbonjean This would take a bit of convincing... When the list is empty, we're basically alone, and should be MASTER, no? I'll admit I haven't looked at this code passage recently... Let's look at this another way. Can you provide your configuration and LMK where the bug/issue is? Since this all WFM. |
|
@purpleidea During the first run, the list is empty for every node (I start puppet simultaneously on all nodes), so all become master and take the VIP. This is fixed on the second run, when the list is populated from the fact, and the first node is selected, at least it is how I understand it. I will provide my configuration as soon as I can but it's very standard, 4 nodes, replication 2, with a hardcoded hosts list (no exported resources). |
|
Does anything break or not work correctly because of this? |
During the first run, the fact gluster_vrrp_fqdns is empty, leading to
all nodes to be set as keepalived master. This could cause race
conditions on the second run when gluster volumes are created because
every node has the VIP, and it is an invalid state anyway.