Is it possible that master node is up but ES cluster is down?
How do you check the state of cluster?
If some nodes of the cluster can still serve request, I would say that it is partially up. What i don't want is the full cluster going down.
The state of cluster you can check in this way:
curl -XGET nodeX:9200
curl -XGET nodeX:9200/_cluster/health?pretty
if status is green the cluster seems OK.
In this way you can check the state of each node of cluster:
curl -XGET nodeX:9200/_cat/nodes
where: m – master node, d – data node, i – ingest node, * - current master node
Thanks for the response. The question is hypothetical. I am trying to understand the value of a "dedicated master node".
What is your exact definition of a cluster being down?
Hmm, interesting topic.
In our case we have a cluster with 3 master nodes. If current master mode will down them other will take over his role.
I would consider the cluster is down when none of the nodes are responding and serving any requests. Because of some past mistakes in contracts, we have allowed some bad queries to pass to the cluster. In recent past, we saw multiple ES failures where all nodes went down/became unresponsive. We have 3 dedicated master node now[Previously all nodes were data + master + coordinating]. I want to understand even if some queries are bad, can that bring the whole cluster down in the presence of a dedicated master node?
In order to serve queries at least some data nodes need to be available. It is potentially possible that all data nodes could be overwhelmed and crash while the dedicated master nodes remain up and available. In such scenario n requests could be served.
What is your definition of up or down?
You can have a master node up but no data nodes, in which case the cluster status is red and you can't do anything other than get the status, etc. So in that case the cluster is probably 'up', but useless.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.