|
| Description: |
*Steps to reproduce:*
# Have a huge number of documents indexed (like a million documents) with a modest amount of RAM (like 6 GB) and internal Solr.
# Perform some searches with a lot of results (like {{*}}, searching for all documents)
*Expected result:*
The wiki is still working and most memory is still available to XWiki.
*Actual result:*
The wiki crashes because of an OOM error and most memory is consumed by Solr's field cache. This can be analyzed with the following Groovy snippet:
{noformat}
{{groovy wiki="false"}}
import org.apache.solr.common.params.ModifiableSolrParams;
def solr = services.component.getInstance(org.xwiki.search.solr.internal.api.SolrInstance)
ModifiableSolrParams params = new ModifiableSolrParams();
params.set("group", "core");
params.set("prefix", "CACHE.core.fieldCache");
params.set("qt", "/admin/metrics");
solr.query(params)
{{/groovy}}
{noformat}
In an instance, it reported 5.3 GiB total size. More information about the problems of this cache can be found at https://support.hcl-software.com/community?id=community_blog&sys_id=ddedbfb11be58914a2f48661cd4bcbd8
In the concrete example, the fields used the following amounts of RAM:
{noformat}
id: 5.02 GB
space_facet: 446.97 MB
space_exact: 266.94 MB
class: 95.76 MB
attauthor: 95.76 MB
mimetype: 95.70 MB
title_sort: 81.11 MB
name_exact: 44.72 MB
creator: 27.56 MB
author: 27.56 MB
locale: 15.09 MB
doclocale: 15.01 MB
type: 8.96 MB
wiki: 5.84 MB
{noformat} |
|
