Right now we only need 4 ES nodes due to the small data volume, and all 4
nodes are master & data nodes.
Q1:
I am wondering in this case, is it necessary to have dedicated master and
client node? Any benefit of having dedicated master node?
Some one said that dedicated master nodes (say, three master nodes) is
helpful to avoid the split brain issue, but even we have NO dedicated
master nodes, we can also avoid the split brain by setting the *discovery.zen.minimum_master_nodes
*to a appropriate value.
Q2:
Similarly, is dedicated client node really necessary in our 4 nodes case?
Any benefit of allocating dedicated client node?
Dedicated master nodes are super convenient if you have the it
infrastructure to host them on shared machines because they are very low
load and its useful to be able to restart the master nodes quickly. We
don't have that kind of infrastructure and our cluster is pretty large and
not having it has bitten us once or twice but its not a huge problem.
Right now we only need 4 ES nodes due to the small data volume, and all 4
nodes are master & data nodes.
Q1:
I am wondering in this case, is it necessary to have dedicated master and
client node? Any benefit of having dedicated master node?
Some one said that dedicated master nodes (say, three master nodes) is
helpful to avoid the split brain issue, but even we have NO dedicated
master nodes, we can also avoid the split brain by setting the *discovery.zen.minimum_master_nodes
*to a appropriate value.
Q2:
Similarly, is dedicated client node really necessary in our 4 nodes case?
Any benefit of allocating dedicated client node?
Thanks Nikolas,
How do you think about dedicated "client node" (the so called load balance
node)? Any benefit of dedicated client node? Seems to me, round robin to
data nodes is good enough.
On Friday, May 29, 2015 at 10:55:01 PM UTC+8, Nikolas Everett wrote:
Dedicated master nodes are super convenient if you have the it
infrastructure to host them on shared machines because they are very low
load and its useful to be able to restart the master nodes quickly. We
don't have that kind of infrastructure and our cluster is pretty large and
not having it has bitten us once or twice but its not a huge problem.
On Fri, May 29, 2015 at 10:44 AM, Xudong You <xudon...@gmail.com
<javascript:>> wrote:
Right now we only need 4 ES nodes due to the small data volume, and all 4
nodes are master & data nodes.
Q1:
I am wondering in this case, is it necessary to have dedicated master and
client node? Any benefit of having dedicated master node?
Some one said that dedicated master nodes (say, three master nodes) is
helpful to avoid the split brain issue, but even we have NO dedicated
master nodes, we can also avoid the split brain by setting the *discovery.zen.minimum_master_nodes
*to a appropriate value.
Q2:
Similarly, is dedicated client node really necessary in our 4 nodes case?
Any benefit of allocating dedicated client node?
If it is good enough for you, it is good enough for you. I will just give
you one anecdote: We implemented 3 dedicated clients on a 9 data node
cluster and got a 2x performance improvement. Moving the query
coordination, network io (has to receive data from every shard), and
combination of results (aggs and sorts) off of the nodes providing the
results is very helpful.
Thanks Nikolas,
How do you think about dedicated "client node" (the so called load balance
node)? Any benefit of dedicated client node? Seems to me, round robin to
data nodes is good enough.
On Friday, May 29, 2015 at 10:55:01 PM UTC+8, Nikolas Everett wrote:
Dedicated master nodes are super convenient if you have the it
infrastructure to host them on shared machines because they are very low
load and its useful to be able to restart the master nodes quickly. We
don't have that kind of infrastructure and our cluster is pretty large and
not having it has bitten us once or twice but its not a huge problem.
Right now we only need 4 ES nodes due to the small data volume, and all
4 nodes are master & data nodes.
Q1:
I am wondering in this case, is it necessary to have dedicated master
and client node? Any benefit of having dedicated master node?
Some one said that dedicated master nodes (say, three master nodes) is
helpful to avoid the split brain issue, but even we have NO dedicated
master nodes, we can also avoid the split brain by setting the *discovery.zen.minimum_master_nodes
*to a appropriate value.
Q2:
Similarly, is dedicated client node really necessary in our 4 nodes
case? Any benefit of allocating dedicated client node?
Thanks James's sharing.
Does your client node has same performance (CPU & memory) as data node or
lower perf?
On Sunday, May 31, 2015 at 12:25:58 AM UTC+8, James Macdonald wrote:
If it is good enough for you, it is good enough for you. I will just give
you one anecdote: We implemented 3 dedicated clients on a 9 data node
cluster and got a 2x performance improvement. Moving the query
coordination, network io (has to receive data from every shard), and
combination of results (aggs and sorts) off of the nodes providing the
results is very helpful.
James
On Sat, May 30, 2015 at 9:11 AM, Xudong You <xudon...@gmail.com
<javascript:>> wrote:
Thanks Nikolas,
How do you think about dedicated "client node" (the so called load
balance node)? Any benefit of dedicated client node? Seems to me, round
robin to data nodes is good enough.
On Friday, May 29, 2015 at 10:55:01 PM UTC+8, Nikolas Everett wrote:
Dedicated master nodes are super convenient if you have the it
infrastructure to host them on shared machines because they are very low
load and its useful to be able to restart the master nodes quickly. We
don't have that kind of infrastructure and our cluster is pretty large and
not having it has bitten us once or twice but its not a huge problem.
Right now we only need 4 ES nodes due to the small data volume, and all
4 nodes are master & data nodes.
Q1:
I am wondering in this case, is it necessary to have dedicated master
and client node? Any benefit of having dedicated master node?
Some one said that dedicated master nodes (say, three master nodes) is
helpful to avoid the split brain issue, but even we have NO dedicated
master nodes, we can also avoid the split brain by setting the *discovery.zen.minimum_master_nodes
*to a appropriate value.
Q2:
Similarly, is dedicated client node really necessary in our 4 nodes
case? Any benefit of allocating dedicated client node?
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.