drill-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (DRILL-5089) Skip initializing all enabled storage plugins for every query
Date Fri, 17 Nov 2017 21:29:01 GMT

    [ https://issues.apache.org/jira/browse/DRILL-5089?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16257601#comment-16257601

ASF GitHub Bot commented on DRILL-5089:

Github user chunhui-shi commented on a diff in the pull request:

    --- Diff: exec/java-exec/src/main/java/org/apache/drill/exec/store/dfs/WorkspaceSchemaFactory.java
    @@ -150,14 +152,30 @@ public WorkspaceSchemaFactory(
        * @return True if the user has access. False otherwise.
       public boolean accessible(final String userName) throws IOException {
    -    final FileSystem fs = ImpersonationUtil.createFileSystem(userName, fsConf);
    +    final DrillFileSystem fs = ImpersonationUtil.createFileSystem(userName, fsConf);
    +    return accessible(fs);
    +  }
    +  /**
    +   * Checks whether a FileSystem object has the permission to list/read workspace directory
    +   * @param fs a DrillFileSystem object that was created with certain user privilege
    +   * @return True if the user has access. False otherwise.
    +   * @throws IOException
    +   */
    +  public boolean accessible(DrillFileSystem fs) throws IOException {
         try {
    -      // We have to rely on the listStatus as a FileSystem can have complicated controls
such as regular unix style
    -      // permissions, Access Control Lists (ACLs) or Access Control Expressions (ACE).
Hadoop 2.7 version of FileSystem
    -      // has a limited private API (FileSystem.access) to check the permissions directly
    -      // (see https://issues.apache.org/jira/browse/HDFS-6570). Drill currently relies
on Hadoop 2.5.0 version of
    -      // FileClient. TODO: Update this when DRILL-3749 is fixed.
    -      fs.listStatus(wsPath);
    +      /**
    +       * For Windows local file system, fs.access ends up using DeprecatedRawLocalFileStatus
which has
    +       * TrustedInstaller as owner, and a member of Administrators group could not satisfy
the permission.
    +       * In this case, we will still use method listStatus.
    +       * In other cases, we use access method since it is cheaper.
    +       */
    +      if (SystemUtils.IS_OS_WINDOWS && fs.getUri().getScheme().equalsIgnoreCase(FileSystemSchemaFactory.LOCAL_FS_SCHEME))
    --- End diff --
    Yes. it was tested in windows unit tests.

> Skip initializing all enabled storage plugins for every query
> -------------------------------------------------------------
>                 Key: DRILL-5089
>                 URL: https://issues.apache.org/jira/browse/DRILL-5089
>             Project: Apache Drill
>          Issue Type: Improvement
>          Components: Query Planning & Optimization
>    Affects Versions: 1.9.0
>            Reporter: Abhishek Girish
>            Assignee: Chunhui Shi
>            Priority: Critical
> In a query's lifecycle, at attempt is made to initialize each enabled storage plugin,
while building the schema tree. This is done regardless of the actual plugins involved within
a query. 
> Sometimes, when one or more of the enabled storage plugins have issues - either due to
misconfiguration or the underlying datasource being slow or being down, the overall query
time taken increases drastically. Most likely due the attempt being made to register schemas
from a faulty plugin.
> For example, when a jdbc plugin is configured with SQL Server, and at one point the underlying
SQL Server db goes down, any Drill query starting to execute at that point and beyond begin
to slow down drastically. 
> We must skip registering unrelated schemas (& workspaces) for a query. 

This message was sent by Atlassian JIRA

View raw message