This week I had a problem with the Solr index: http://lucene.472066.n3.nabble.com/corrupted-index-in-slave-td4054769.html ,
Today, this error started to occur constantly for almost every request, and I created the JIRA problem because I thought it was a https://issues.apache.org/jira/browse/SOLR-4707 error
As you can read, in the end it was due to a failure in the master-slave Solr replication, and now I don’t know if we should think about switching to SolrCloud, since the Solr-slave replication does not seem to meet our requirements:
- index size: ~ 20 million documents, ~ 9 GB
- ~ 1200 updates / min
- ~ 10000 requests / min (distributed over 2 subordinates) MoreLikeThis, RealTimeGet, TermVectorComponent, SearchHandler
I would like to thank you if anyone could help me answer these questions:
- Would it be wise to upgrade to SolrCloud? Will it affect replication performance?
- In this case, what would improve performance? save a copy of the index on each server or use shard servers?
- How many shards and replicas would you recommend for high availability?
Yours faithfully,
Victor
source
share