Media Wiki CirrusSearch

From NovaOrdis Knowledge Base
Revision as of 02:09, 22 November 2021 by Ovidiu (talk | contribs) (→‎Installation)
Jump to navigation Jump to search

External

Internal

Installation

Install elasticsearch. Use the latest recommended major version elasticsearch. Do not use 7.x if they recommend 6.x.

Elasticsearch Installation

Install Elastica extension:

Elastica Extension Installation

Download the correct CirrusSearch version from https://www.mediawiki.org/wiki/Special:ExtensionDistributor/CirrusSearch

Unzip it in the extensions directory:

tar -xzf CirrusSearch-REL1_35-95b958b.tar.gz -C /var/www/mediawiki-x.y.z/extensions

Add the following statement to LocalSettings.php:

wfLoadExtension( 'CirrusSearch' );

Smoke test: go to Special:Version and verify that the extension is successfully installed.

Follow the installation instructions delivered with /var/www/mediawiki-1.35.0/extensions/CirrusSearch/README.

In LocalSettings.php:

wfLoadExtension( 'Elastica' );
wfLoadExtension( 'CirrusSearch' );
$wgDisableSearchUpdate = true;


As root:

php /var/www/mediawiki-.../extensions/CirrusSearch/maintenance/UpdateSearchIndexConfig.php

Remove

$wgDisableSearchUpdate = true;

from LocalSettings.php.

php /var/www/mediawiki-.../extensions/CirrusSearch/maintenance/maintenance/ForceSearchIndex.php --skipLinks --indexOnSkip
php /var/www/mediawiki-.../extensions/CirrusSearch/maintenance/maintenance/ForceSearchIndex.php --skipParse

In LocalSettings.php:

$wgSearchType = 'CirrusSearch';

SELinux Setup

If SELinux is enabled on the host, the httpd process may not be able to invoke into the elasticsearch server, with the following symptoms:

[CirrusSearch] Search backend error during near_match search for 'test' after 5: unknown: Couldn't connect to host, Elasticsearch down?

/var/log/audit/audit.log will contain something similar to:

type=AVC msg=audit(1637546254.160:75): avc:  denied  { name_connect } for  pid=1216 comm="httpd" dest=9200 scontext=system_u:system_r:httpd_t:s0 tcontext=system_u:object_r:wap_wsp_port_t:s0 tclass=tcp_socket permissive=0

Use audit2allow to generate the SELinux policy to allow the denied operation. Follow:

Troubleshooting, Diagnosing and Fixing SELinux Problems

The summary is there:

  • Try a search operation that will fail.
  • grep httpd /var/log/audit/audit.log | audit2allow -M mysepolicy
  • semodule -i mysepolicy.pp

Troubleshooting

Enable debugging as shown in:

MediaWiki Troubleshooting
[CirrusSearchRequests] near_match search for 'test' against {index} took 5 millis. Requested via web for 5b7883dcb7cd761f62397caa67819012 by executor 2078481799
[CirrusSearch] Search backend error during near_match search for 'test' after 5: unknown: Couldn't connect to host, Elasticsearch down?