Some errors migrating to solr cloud

[tedsolr <ts...@sciquest.com>]

I had a fairly simple plan for migrating my single solr instance with
multiple cores, to a solrcloud implementation where core => collection. My
testing locally (windows) worked fine, but the first linux (development)
environment I tried to migrate had some failures. This is v5.2.1.

The setup: Single linux box for two solr nodes: ports 8983 & 8984. Both are
children of the only SOLRHOME folder. Just using the embedded ZK. Single
shard collections with replication factor of 2.
SOLRHOME/server/solr (port 8983 with embedded ZK on 9983)
- solr start -c
SOLRHOME/server/solr2 (port 8984)
- solr start -c -p 8984 -s solr2 -z localhost:9983
 
The plan: Start solr in cloud mode; upload config to ZK; create collections
via the collections API; stop solr; copy the "data" folders from the old
cores into the new collections on 8983 (/solr); start solr again

The first symptom of the problem was trying to stop all nodes with "solr
stop -all". It only shut down node 8983. When I then tried "solr stop -p
8984" it had to kill it. Then I noticed the errors in the log: "Error while
trying to recover. Server refused connection at:
http://10.0.5.213:8984/solr" & "Error while trying to recover. No registered
leader was found after waiting for 4000ms".

All the indexes I moved were very small - less than 1MB. Only 2 out of 5
collections replicated when solr restarted. The only "clue" (unless its just
coincidence) is the two that worked had 8983 as their leader node. The other
3 collections had 8984 - which doesn't have a ZK. Its confusing because this
same plan worked on my local machine - even when a collection had 8984 as
the leader. Is there a flaw in my plan? Maybe I have to force the leaders to
be the same node with the ZK? Why didn't "solr stop -all" work?



--
View this message in context: http://lucene.472066.n3.nabble.com/Some-errors-migrating-to-solr-cloud-tp4243594.html
Sent from the Solr - User mailing list archive at Nabble.com.