Hi,
We were getting pretty low ingest throughput when using replicas. We traced it
to the following logic in indexing code:
- Read SolrInputDocument from stream
- Index a document
- Add to ConcurrentUpdateSolrServer instance for sending to replica
So, we
On 10/17/2014 8:50 AM, Cp Mishra wrote:
So, we changed the logic to:
-Read SolrInputDocument objects from stream in batches of 500.
-Add documents to ConcurrentUpdateSolrServer instance
-Index documents in a loop
This has improved indexing speed significantly.
What are the caveats to this
Pretty sure that's never been made configurable.
I've seen anecdotal evidence of a 30-40% slowdown when adding the
first replica, from there the penalty is much less.
Cp Mishra:
Any time you change code you're absolutely invited to open a JIRA and
attach the code for people to look at. Please