All shards are placed to the first node

[avlukanin]

I used the default settings, so 3 nodes are created. After this command:

runner.createIndex("test_index", Settings.builder().build());

I see that the data/node_1/elasticsearch-cluster-runner/nodes/0/indices/test_index folder has all the shards folders, but data/node_2/elasticsearch-cluster-runner/nodes/0/ and data/node_2/elasticsearch-cluster-runner/nodes/0/ folders has no indices subfolder at all.

If I add these settings:

cluster.routing.allocation.total_shards_per_node: 1
index.number_of_shards: 3
index.number_of_replicas: 0

I get this exception:

org.codelibs.elasticsearch.runner.ClusterRunnerException: ensureYellow timed out, cluster state:

version: 5
state uuid: 0tU2XEO6SeagUQEIui6JRg
from_diff: false
meta data version: 3
nodes: 
   {Node 1}{HLJ9sxVTR2OlxA321H1n2Q}{127.0.0.1}{127.0.0.1:9301}{master=true}, local, master
routing_table (version 3):
-- index [test_index]
----shard_id [test_index][0]
--------[test_index][0], node[HLJ9sxVTR2OlxA321H1n2Q], [P], v[2], s[STARTED], a[id=ye9MxH2LRNeCif-c_G8-CQ]
----shard_id [test_index][1]
--------[test_index][1], node[null], [P], v[0], s[UNASSIGNED], unassigned_info[[reason=INDEX_CREATED], at[2016-06-17T09:28:15.461Z]]
----shard_id [test_index][2]
--------[test_index][2], node[null], [P], v[0], s[UNASSIGNED], unassigned_info[[reason=INDEX_CREATED], at[2016-06-17T09:28:15.461Z]]

routing_nodes:
-----node_id[HLJ9sxVTR2OlxA321H1n2Q][V]
--------[test_index][0], node[HLJ9sxVTR2OlxA321H1n2Q], [P], v[2], s[STARTED], a[id=ye9MxH2LRNeCif-c_G8-CQ]
---- unassigned
--------[test_index][1], node[null], [P], v[0], s[UNASSIGNED], unassigned_info[[reason=INDEX_CREATED], at[2016-06-17T09:28:15.461Z]]
--------[test_index][2], node[null], [P], v[0], s[UNASSIGNED], unassigned_info[[reason=INDEX_CREATED], at[2016-06-17T09:28:15.461Z]]

If I download Elasticsearch 2.3.3, run 3 nodes and add an index, all 3 shards are allocated evenly at 3 nodes.

I also tried to download elasticsearch-cluster-runner, update dependency to 2.3.3 and run the test with the updated version. The same bug.

[avlukanin]

I've copied these settings from your test code and now the index is splitted evenly:

                settingsBuilder.put("http.cors.enabled", true);
                settingsBuilder.put("http.cors.allow-origin", "*");
                settingsBuilder.putArray("discovery.zen.ping.unicast.hosts", "localhost:9301-9305");

[avlukanin]

So, it is better to add these settings automatically if more than one shard is added to the cluster. I wonder why standalone Elasticsearch does not require these settings? Because it assigns ports automatically?

[marevol]

Our plugin sets transport.tcp.port to specific port(9301-9303) for 3 nodes,
but standalone elasticsearch uses 9300-9400(range).
I think it's difficult to change the default value of our plugin
because the plugin manages an available port for transport.