I have 20 large machines with 256gb of memory on each and I configured the
nodes in the following way:
Total Nodes: 60 (3 per physical machine)
Memory for each node: 30g
Discovery: Unicast
discovery.zen.ping.unicast.hosts: ["host1[9300-9400], host2[9300-9400],
....................host20[9300-9400]"]
discovery.zen.minimum_master_nodes: 31 (TO PREVENT SPLIT BRAIN)
Total Master+Data Nodes:40, DataOnly Nodes:20
Q1: Is this the right way to setup the Unicast? Or should I just specify
one or two hosts in the "unicast.hosts" list and Start those before
starting the rest of the nodes?
When I configured Unicast hosts with all the available hosts, My cluster
is taking very long time to come up and also some of the shards are getting
stuck at "INITIALIZING" state. Also, some of the nodes don't even come up.
I see some warnings in the logs saying cluster initialization not set after
30seconds.
Q2: In large cluster like this, Is there a benefit in making some of the
nodes just Master with no data? If so, how much memory should I give to
Master only nodes?
Are you just running multiple instances of the elasticsearch binary on each
node?
If yes, what is your network interface on each of the physicals? Chances
are you're saturating the interfaces when bringing all the nodes up at once.
I have 20 large machines with 256gb of memory on each and I configured the
nodes in the following way:
Total Nodes: 60 (3 per physical machine)
Memory for each node: 30g
Discovery: Unicast
discovery.zen.ping.unicast.hosts: ["host1[9300-9400], host2[9300-9400],
....................host20[9300-9400]"]
discovery.zen.minimum_master_nodes: 31 (TO PREVENT SPLIT BRAIN)
Total Master+Data Nodes:40, DataOnly Nodes:20
Q1: Is this the right way to setup the Unicast? Or should I just specify
one or two hosts in the "unicast.hosts" list and Start those before
starting the rest of the nodes?
When I configured Unicast hosts with all the available hosts, My cluster
is taking very long time to come up and also some of the shards are getting
stuck at "INITIALIZING" state. Also, some of the nodes don't even come up.
I see some warnings in the logs saying cluster initialization not set after
30seconds.
Q2: In large cluster like this, Is there a benefit in making some of the
nodes just Master with no data? If so, how much memory should I give to
Master only nodes?
I have Gig eth on all of the nodes and the only software running on each
node is 3 instances of ES from the same binary.
On Tuesday, December 17, 2013 3:32:19 AM UTC-5, Mark Walkom wrote:
Are you just running multiple instances of the elasticsearch binary on
each node?
If yes, what is your network interface on each of the physicals? Chances
are you're saturating the interfaces when bringing all the nodes up at once.
On 17 December 2013 11:44, Search User <feed...@gmail.com <javascript:>>wrote:
Hello,
I have 20 large machines with 256gb of memory on each and I configured
the nodes in the following way:
Total Nodes: 60 (3 per physical machine)
Memory for each node: 30g
Discovery: Unicast
discovery.zen.ping.unicast.hosts: ["host1[9300-9400], host2[9300-9400],
....................host20[9300-9400]"]
discovery.zen.minimum_master_nodes: 31 (TO PREVENT SPLIT BRAIN)
Total Master+Data Nodes:40, DataOnly Nodes:20
Q1: Is this the right way to setup the Unicast? Or should I just specify
one or two hosts in the "unicast.hosts" list and Start those before
starting the rest of the nodes?
When I configured Unicast hosts with all the available hosts, My cluster
is taking very long time to come up and also some of the shards are getting
stuck at "INITIALIZING" state. Also, some of the nodes don't even come up.
I see some warnings in the logs saying cluster initialization not set after
30seconds.
Q2: In large cluster like this, Is there a benefit in making some of the
nodes just Master with no data? If so, how much memory should I give to
Master only nodes?
I have Gig eth on all of the nodes and the only software running on each
node is 3 instances of ES from the same binary.
On Tuesday, December 17, 2013 3:32:19 AM UTC-5, Mark Walkom wrote:
Are you just running multiple instances of the elasticsearch binary on
each node?
If yes, what is your network interface on each of the physicals? Chances
are you're saturating the interfaces when bringing all the nodes up at once.
I have 20 large machines with 256gb of memory on each and I configured
the nodes in the following way:
Total Nodes: 60 (3 per physical machine)
Memory for each node: 30g
Discovery: Unicast
discovery.zen.ping.unicast.hosts: ["host1[9300-9400], host2[9300-9400],
....................host20[9300-9400]"]
discovery.zen.minimum_master_nodes: 31 (TO PREVENT SPLIT BRAIN)
Total Master+Data Nodes:40, DataOnly Nodes:20
Q1: Is this the right way to setup the Unicast? Or should I just specify
one or two hosts in the "unicast.hosts" list and Start those before
starting the rest of the nodes?
When I configured Unicast hosts with all the available hosts, My
cluster is taking very long time to come up and also some of the shards are
getting stuck at "INITIALIZING" state. Also, some of the nodes don't even
come up. I see some warnings in the logs saying cluster initialization not
set after 30seconds.
Q2: In large cluster like this, Is there a benefit in making some of the
nodes just Master with no data? If so, how much memory should I give to
Master only nodes?
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.