Solr Configuration

The Widget Framework uses Solr for a variety of look-up operations, and in a production environment it will exert a heavy load on Solr, especially if your Data Sources use many Solr-based query types such as Query by section, Query by tag and Query by search. You are therefore strongly recommended to enable Solr response caching. This will significantly reduce the load on Solr and improve its performance.

When configuring Solr you need to ensure:

• That the Solr caches are large enough

• That Solr caches search results for a suitable length of time

• That Solr adds the correct HTTP caching headers to its search query responses

Below are some suggested settings for the Solr configuration file (solrconfig.xml) that you can start with, which should give reasonable results. If you want to optimize your installation further, then you should consult the Solr documentation.

• <autoCommit>
    <maxDocs>100</maxDocs>
    <maxTime>300000</maxTime>
  </autoCommit>

  This tells Solr to cache up to 100 query results for a period of 5 minutes.

• <query>
    ...
    <queryResultCache
      class="solr.LRUCache"
      size="2048"
      initialSize="512"
      autowarmCount="0"/>
  
    <documentCache
      class="solr.LRUCache"
      size="20000"
      initialSize="1024"
      autowarmCount="0"/>
    ...
  </query>

  This sets Solr's internal caches to a suitable size for holding the required number of results.

• <requestDispatcher handleSelect="true">
    ...
    <httpCaching lastModFrom="openTime" etagSeed="Solr">
      <cacheControl>max-age=30, public</cacheControl>
    </httpCaching>
    ...
  </requestDispatcher>

  This tells the Widget Framework to cache supplied results for 30 seconds.

With these settings in place, the first time a particular query is submitted to Solr, it will perform the search, save the results in its cache and also return the results to the Widget Framework with an HTTP header containing an instruction to cache the result for 30 seconds. If the same request is repeated within 30 seconds, then the results will be returned from the Widget Framework's cache. If the same request is repeated after 30 seconds have elapsed, then the Widget Framework will again submit a request to Solr, but with an HTTP caching header asking for the response in its cache to be revalidated. If the result is still in the Solr cache, then Solr will return an HTTP 304 Not Modified response (since this is more efficient than returning the cached response again). The Widget Framework can then use the response in its own cache and revalidate it for another 30 seconds. If the Solr cache has expired in the mean time then Solr will perform the requested search again, cache the results and return them with the same caching header.

How much effect this configuration has on Solr performance will obviously depend on usage in your particular installation, and achieving optimum results may require some adjustment. For more information about Solr caching, see the Solr documentation.