One of the common problems with Java based applications is out of memory. There have been a few posts on the topic and want to reiterate some of the symptoms and solutions.
Large number of rows
Apache Solr is a scoring engine underneath the hood. This means that it is designed to return the best ranked documents for the end user. One of the common use cases is ecommerce where the top 5-10 results need to match what the user is looking for or they will go elsewhere. Solr can also be used in the analytics space where large result sets need to be returned if matching a set of search criteria. Returning a large number of rows is possible with Solr but this cannot be done with
rows parameter for Solr can be used to return more than the default of 10 rows. I have seen users successfully set the
rows parameter to 100-200 and not see any issues. However, setting the
rows parameter higher has a big memory consequence and should be avoided at all costs. The Solr wiki has details about the memory consumption when requesting a lot of rows here. Furthermore, a high
rows parameter means that in each shard in a Solr Cloud setup will need to return that many rows to the leader for final sorting and processing. This can significantly slow down the query even if not running into any memory problems.
An out of memory (OOM) error typically occurs after a query comes in with a large
rows parameter. Solr will typically work just fine up until that query comes in. Sometimes this can be hard to track down but looking at the Solr logs will help here. It may be tempting to increase heap to resolve this error but be aware that that will not solve the issue. Some ideas for solutions to this problem will be shown below.
Large start for paging results
As with large number of rows above, Solr can be used to return a lot of rows. One of the patterns that I’ve seen is not using a large
rows parameter but instead using a large
start parameter like
start=1000000. By using the
start parameters, many users think they will be able to get millions of results back from Solr. This will work in a test environment but for actual millions of rows this pattern will fail. This will most likely not cause an OOM memory problem, but results will not be returned quickly. The same solutions can help in this situation as well. For more details see below solutions.
Apache Solr has a fantastic feature called faceting that allows for counts of terms in the index. This can be used in a variety of different ways. Sorting is also core to Solr in that we want to be able to return relevant results based on a user’s sort criteria. The problem with faceting and sorting is that they are done on the uninverted values stored in the index. What that means to users who are not familiar with Lucene, is that the original tokens need to be available and not just the location in the document.
When it comes to memory usage, if a field in Solr needs to be faceted or sorted and the uninverted representation is not available it will be built on heap. This result is then cached in the FieldCache. Every time the index is changed the field needs to be uninverted again and stored on heap. This can take up a lot of heap and caused out of memory (OOM). This may not happen immediately but over time the more uninverted fields are put on the FieldCache.
This symptom comes up often during analytics when someone wants to explore an index with a tool like Banana. Banana is built on faceting and sorted so it requires that most fields in the index be uninverted. This can cause a previously working Solr setup to crash due to OOM errors.
There is good news since starting from Apache Lucene/Solr 4.0, there is a new feature called DocValues. There is more information in the solutions section below.
solr, oom, faceting, sorting, grouping oom, large number of rows, paging, large number of results, debugging solr, rows parameter, start parameter, docValues, field cache, streaming expressions, parallel sql
About this pageThis is a preview of a SAP Knowledge Base Article. Click more to access the full version on SAP ONE Support launchpad (Login required).
Search for additional results
Visit SAP Support Portal's SAP Notes and KBA Search.