[ https://issues.apache.org/jira/browse/DRILL-3742?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14955890#comment-14955890 ]
ASF GitHub Bot commented on DRILL-3742: --------------------------------------- Github user julienledem commented on a diff in the pull request: https://github.com/apache/drill/pull/148#discussion_r41938462 --- Diff: contrib/storage-hive/core/src/main/java/org/apache/drill/exec/expr/fn/HiveFunctionRegistry.java --- @@ -53,12 +54,14 @@ * @param config */ public HiveFunctionRegistry(DrillConfig config) { - Set<Class<? extends GenericUDF>> genericUDFClasses = PathScanner.scanForImplementations(GenericUDF.class, null); + // TODO: see if we can avoid this. We can't change the constructor right now. + ScanResult classpathScan = ClassPathScanner.fromPrescan(config); --- End diff -- no because fromPrescan has a cache (it keeps track of the last scan and reuses it if we ask for the same thing), but that's far from ideal. > Improve classpath scanning to reduce the time it takes > ------------------------------------------------------ > > Key: DRILL-3742 > URL: https://issues.apache.org/jira/browse/DRILL-3742 > Project: Apache Drill > Issue Type: Improvement > Reporter: Julien Le Dem > Fix For: Future > > > classpath scanning and function registry take a long time (seconds every > time). > We'd want to avoid loading the classes (use bytecode inspection instead) and > have a build time cache to avoid doing the scanning at startup. -- This message was sent by Atlassian JIRA (v6.3.4#6332)