Consistency across cluster


(Kinesh Satiya-2) #1

I have a cluster running which has 4 nodes. Each node has an EBS volume
attached to it, where the node backs up all the indexes it has.

So,
node1 => ebs 1
node2 => ebs 2
node3 => ebs 3
node4 => ebs 4

Now I need to shut down node 4. Since the load is less on this cluster, I
shut down one the node and have written a script to re spawn it when the
load increases on the cluster or more memory is required.

So, in the case when the node4 is re-spawned, it will attach its ebs 4 and
start loading all of these indexes stored in the shard.

Lets say some of the indexes which were initially with node 4 got updated
after node 4 is down ( node 1 has a replicated copy and node 1 updates
those indexes), then when node 4 is re-spawned, Will node 4 look for
updates and then update the indexes in its shard? or will it read the shard
and contain old indexes ( not updated) ?


(simonw-2) #2

On Wednesday, July 25, 2012 9:05:13 PM UTC+2, Kinesh Satiya wrote:

I have a cluster running which has 4 nodes. Each node has an EBS volume
attached to it, where the node backs up all the indexes it has.

So,
node1 => ebs 1
node2 => ebs 2
node3 => ebs 3
node4 => ebs 4

Now I need to shut down node 4. Since the load is less on this cluster, I
shut down one the node and have written a script to re spawn it when the
load increases on the cluster or more memory is required.

So, in the case when the node4 is re-spawned, it will attach its ebs 4 and
start loading all of these indexes stored in the shard.

Lets say some of the indexes which were initially with node 4 got updated
after node 4 is down ( node 1 has a replicated copy and node 1 updates
those indexes), then when node 4 is re-spawned, Will node 4 look for
updates and then update the indexes in its shard? or will it read the shard
and contain old indexes ( not updated) ?

if the node joins the cluster again ES will recover the index shards on
that machine and bring them up-to-date. You can try that out and see how
the cluster state changes. Just shut it down and index a couple of
documents. Then bring it up again and shut down the other node holding the
same shard. it should be in sync once recovery has finished. Make sure
recovery is done before you stop another node.

simon


(system) #3