fix(learner nodes): Reconnect to learner nodes after restart #7554
Conversation
When an alpha would restart, it wasn't reconnecting to the learner nodes. This PR fixes that.
jarifibrahim
requested review from
manishrjain and
vvbalaji-dgraph
as code owners
| @@ -1760,6 +1760,12 @@ func (n *node) InitAndStartNode() { | |||
| n.Connect(id, m.Addr) | |||
| } | |||
| } | |||
| for _, id := range sp.Metadata.ConfState.Learners { | |||
| m, ok := members[id] | |||
There was a problem hiding this comment.
I am not sure why this check is necessary. The loop above had the check and so I've added it.
There was a problem hiding this comment.
I think this might be needed because of the following reason. Not sure though.
Say an alpha node has a stale snapshot and the cluster has progressed (some nodes added/deleted). When we restart that node, it first gets the snapshot from the WAL and verifies that it actually exists in latest membership state. If yes, then it connects to it.
This change will also mean that learner nodes will connect among themselves. Is that expected behaviour?
There was a problem hiding this comment.
This change will also mean that learner nodes will connect among themselves. Is that expected behaviour?
That's expected. All alphas connect to all other alphas.
When an alpha would restart, it wasn't reconnecting to the learner nodes. This PR fixes that. Fixes DGRAPH-3125 (cherry picked from commit 586181f)
When an alpha would restart, it wasn't reconnecting to the learner
nodes. This PR fixes that.
Fixes DGRAPH-3125
This change is