lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pranaya Behera <behera.pran...@gmail.com>
Subject Nodes goes down but never recovers.
Date Thu, 20 Apr 2017 09:43:34 GMT
Hi,
     Through SolrJ I am trying to upload configsets and create
collections in my solrcloud.

Setup:
1 Standalone zookeeper listening on 2181 port. version 3.4.10
-- bin/zkServer.sh start
3 Starting solr nodes. (All running from the same solr.home) version
6.5.0 and as well in 6.2.1
-- bin/solr -c -z localhost:2181 -p 8983
-- bin/solr -c -z localhost:2181 -p 8984
-- bin/solr -c -z localhost:2181 -p 8985

After first run of my java application to upload the config and create
the collections in solr through zookeeper is seemless and working
fine.
Here is the clusterstatus after the first run.
https://gist.github.com/shadow-fox/5874f8b5de93fff0f5bcc8886be81d4d#file-3nodes-json

Stopped one solr node via:
-- bin/solr stop -p 8985
clusterstatus changed to:
https://gist.github.com/shadow-fox/5874f8b5de93fff0f5bcc8886be81d4d#file-3nodes1down-json

Till now everything is as expected.

Here is the remaining part where it confuses me.

Bring the down node back to life. Clusterstatus changed from 2 node
down with 1 node not found to 3 node down including the new node that
just brought up.
https://gist.github.com/shadow-fox/5874f8b5de93fff0f5bcc8886be81d4d#file-3nodes3down-json
Expected result should be all the other nodes should be in active mode
and this one would be recovery mode and then it would be active mode,
as this node had data before i stopped it using the script.

Now I added one more node to the cluster via
-- bin/solr -c -z localhost:2181 -p 8986
The clusterstatus changed to:
https://gist.github.com/shadow-fox/5874f8b5de93fff0f5bcc8886be81d4d#file-4node3down-json
This one just retains the previous state and adds the node to the cluster.


When bringing up the removed node which was previously in the cluster
which was registered to the zookeeper and has data about the
collections be registered as active rather than making every other
node down ? If so what is the solution to this ?

When we add more nodes to an existing cluster, how to ensure that it
also gets the same collections/data i.e. basically synchronizes with
the other nodes which are present in the node rather than manually
create collection for that specific node ? As you can see from the
lastly added node's clusterstate it is there in the live_nodes but
never got the collections into its data dir.
Is there any other way to add a node with the existing cluster with
the cluster data ?

For the completion here is the code that is used to upload config and
create collection through CloudSolrClient in Solrj.(Not full code but
part of it where the operation is happening.)
https://gist.github.com/shadow-fox/5874f8b5de93fff0f5bcc8886be81d4d#file-code-java
Thats all there is for a collection to create: upload configsets to
zookeeper, create collection and reload collection if required.

This I have tried in my local Mac OS Sierra and also in AWS env which
same effect.



-- 
Thanks & Regards
Pranaya PR Behera

Mime
View raw message