UrlRewriteFilter might cause sudden CPU spikes

• Type: Bug
• Status: Closed
• Resolution: Fixed
• Affects Version/s: 6.1.30 EE GA3, 6.1.X EE, 6.2.X EE, 7.0.0 M3
• Fix Version/s: 6.1.X EE, 6.2.3 CE GA4, 6.2.X EE, 7.0.0 M3
• Component/s: Performance
• Labels:
  • ee-ts
  • liferay-fixpack-portal-34-6130
  • liferay-fixpack-portal-39-6210
  • performance
• Branch Version/s:
  6.2.x, 6.1.x
• Backported to Branch:
  Committed
• Story Points: 22
• Fix Priority: 4
• Git Pull Request: https://github.com/brianchandotcom/liferay-portal/pull/22749

Description

UrlRewriteFilter doesn't ignore ComboServlet and processes its URLs unnecessarily. When those URLs which are addressing ComboServlet are long (~1Kb), parsing them by UrlRewriteFilter becomes expensive and causes sudden CPU spikes.

Steps to reproduce

1) Set the following JVM options in setenv.sh

export HPROF_START=$(date +%Y%m%d.%H%M%S)
export JAVA_OPTS="$JAVA_OPTS -Xrunhprof:cpu=samples,doe=n,thread=n,format=a,file=${CATALINA_HOME}/logs/java.${HPROF_START}.hprof"

2) Start the portal

3) Reset CPU samples by issuing

% kill -QUIT <pid>

4) Generate load by accessing the following URL concurently

/combo/?browserId=other&minifierType=&languageId=nl_BE&b=6130&t=1396260298000&p=/html/js&m=/aui/aui-char-counter/aui-char-counter-min.js&m=/aui/aui-datatype/aui-datatype-min.js&m=/aui/datatype-date-parse/datatype-date-parse-min.js&m=/aui/datatype-date-format/lang/datatype-date-format_en-US.js&m=/aui/datatype-date-format/datatype-date-format-min.js&m=/aui/widget-locale/widget-locale-min.js&m=/aui/aui-calendar/aui-calendar-min.js&m=/aui/aui-datepicker/aui-datepicker-base-min.js&m=/aui/aui-state-interaction/aui-state-interaction-min.js&m=/aui/aui-button-item/aui-button-item-min.js&m=/aui/aui-datepicker/aui-datepicker-select-min.js&m=/aui/dd-ddm-base/dd-ddm-base-min.js&m=/aui/dd-drag/dd-drag-min.js&m=/aui/dd-ddm/dd-ddm-min.js&m=/aui/dd-ddm-drop/dd-ddm-drop-min.js&m=/aui/dd-drop/dd-drop-min.js&m=/aui/dd-drop-plugin/dd-drop-plugin-min.js&m=/aui/dd-delegate/dd-delegate-min.js&m=/aui/datasource-local/datasource-local-min.js&m=/aui/datasource-io/datasource-io-min.js&m=/aui/datasource-get/datasource-get-min.js&m=/aui/datasource-function/datasource-function-min.js&m=/aui/cache-base/cache-base-min.js&m=/aui/datasource-cache/datasource-cache-min.js&m=/aui/dataschema-base/dataschema-base-min.js&m=/aui/dataschema-json/dataschema-json-min.js&m=/aui/datasource-jsonschema/datasource-jsonschema-min.js&m=/aui/dataschema-xml/dataschema-xml-min.js&m=/aui/datasource-xmlschema/datasource-xmlschema-min.js&m=/aui/dataschema-array/dataschema-array-min.js&m=/aui/datasource-arrayschema/datasource-arrayschema-min.js&m=/aui/dataschema-text/dataschema-text-min.js&m=/aui/datasource-textschema/datasource-textschema-min.js&m=/aui/datasource-polling/datasource-polling-min.js&m=/aui/aui-form/aui-form-field-min.js&m=/aui/aui-form/aui-form-textfield-min.js

5) Gather CPU samples again by issuing

% kill -QUIT <pid>

6) Check the last CPU SAMPLES section of the HPROF file, you should see something like this

CPU SAMPLES BEGIN (total = 266475) Fri Apr  4 10:55:43 2014
 rank   self  accum   count trace method
    1 40.90% 40.90%  108975 303724 java.util.regex.Pattern$GroupTail.match
    2 30.92% 71.81%   82383 303727 java.util.regex.Pattern$Curly.match0
    3  8.13% 79.94%   21664 303741 java.util.regex.Pattern$Curly.match0
    4  3.21% 83.15%    8559 303735 java.util.regex.Pattern$Curly.match0
    5  1.84% 85.00%    4911 302874 java.util.zip.Deflater.deflateBytes
    6  1.53% 86.53%    4081 302451 java.net.PlainSocketImpl.socketAccept
    7  1.23% 87.76%    3280 304394 com.ctc.wstx.io.ReaderBootstrapper.bootstrapInput
    8  0.84% 88.59%    2230 304399 org.tuckey.web.filters.urlrewrite.utils.Log.isUsingLog4j
    9  0.74% 89.34%    1985 304393 com.liferay.portal.service.ServiceContextFactory.getInstance
   10  0.52% 89.86%    1375 304655 com.liferay.portal.sharepoint.SharepointFilter.isSharepointRequest
   ...
CPU SAMPLES END

UrlRewriteFilter gets applied to ComboServlet and causes high CPU usage, java.util.regex.Pattern$GroupTail.match is in the top of the profiler output
UrlRewriteFilter ignores ComboServlet, doesn't cause high CPU usage and java.util.regex.Pattern$GroupTail.match disappears from the top of the profiler output