Using v7.10.2 opendistro, and rolling my own cluster. I cant do either of the following:
Ingest nodes
Can only set this up if it is a data node as well. You cant have an ingest node on its own.
Data only node:
Cant set this up, unless it is a master node as well.
Interestingly, if I set up a master only node, it will use the cluster.initial_master_nodes to find other masters.
If I set up as a data node (only) , it ognores these and tries to find itself as a master node. Same happens with an ingest node.
This looks like a ‘boolean case’ issue somewhere in the node types… if it helps I am also using discovery-ec2 to get this working. maybe I ought to go off and try th eold unicast_hosts file instead… see if its discovery-ec2 causing issues.
Would be interesting to see if anyone else can help.
On ingest only this fails: node.roles: [ingest]
But this works: node.roles: [data,ingest,master]
On data only this fails: node.roles: [data]
But this works: node.roles: [data,master]
By 'fails’I mean it refuses to find or join a current cluster.
Yep… thanks.That was what I tried first, its the very old way of dealing with node types. You cant mix node.roles with th enode.master, node.data, node.ingest method, and when you try to use the old method, it fails in a similar way.
Esseentially…
all data nodes will be masters also.
all ingest nodes have to be data nodes, and so also master nodes
you can have sytandalone master nodes.
Having only last year set up a ~30Bn doc cluster using elasticos open source version… I am now heading back to that. opendistro also doesn;t support PIT (Point in Time scrolling) yet, which is a huge negative for my use case. Its not actually fit for anything, apart from mayv=be learning, and testing on a small scale. imvho
Yep… that was the killer. You cant have an ingest only node… it has to be a data node as well… which in turn will become a master node also. Its all wrong.
I was not trying anything ‘strange’… seperate node types have worked since v0.x back in the day…the fork has changed/ignored some of teh more salient issues regarding ‘always ab=vailable’ … and on a very large cluster would be hugely expensive.
As I said… have thrown oipendistro out now… its not worth the effort… if they cant get the basics right, what will happen with the rest? NOt worth the risk imvho
hi, i am in version opensearch 1.1.0. i tried
node1:
node.master: true
node.data: true
node.ingest: false
node2:
node.master: false
node.data: false
node.ingest: true
then i got “nodes: {total: 1}” when run “GET _nodes/stats”. that means ingest node should be data node? thanks:)