hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Anoop Sam John (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-18160) Fix incorrect logic in FilterList.filterKeyValue
Date Thu, 22 Jun 2017 05:56:00 GMT

    [ https://issues.apache.org/jira/browse/HBASE-18160?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16058805#comment-16058805

Anoop Sam John commented on HBASE-18160:

Another side note. Again not related but we should try fix.
private Cell referenceCell = null;

   * When filtering a given Cell in {@link #filterKeyValue(Cell)},
   * this stores the transformed Cell to be returned by {@link #transformCell(Cell)}.
   * Individual filters transformation are applied only when the filter includes the Cell.
   * Transformations are composed in the order specified by {@link #filters}.
  private Cell transformedCell = null;
referenceCell is been kept till the next call to transformCell().  So there is no usage of
a prev set cell after having an RPC shipped(Reaching a batch size or caching rows count) and
so no functional problems.  Have to reset it to null after the check in transformCell() or
else this will keep the ref to a cell even after it is been shipped in RPC. So keeping ref
to the backing HFile block byte[] preventing it from GC.  A small reset one liner can avoid
such a possible issue!
Same for transformedCell 

> Fix incorrect  logic in FilterList.filterKeyValue
> -------------------------------------------------
>                 Key: HBASE-18160
>                 URL: https://issues.apache.org/jira/browse/HBASE-18160
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Zheng Hu
>            Assignee: Zheng Hu
>         Attachments: HBASE-18160.branch-1.1.v1.patch, HBASE-18160.branch-1.v1.patch,
HBASE-18160.v1.patch, HBASE-18160.v2.patch, HBASE-18160.v2.patch
> As HBASE-17678 said, there are two problems in FilterList.filterKeyValue implementation:

> 1.  FilterList did not consider INCLUDE_AND_SEEK_NEXT_ROW case( seems like INCLUDE_AND_SEEK_NEXT_ROW
is a newly added case, and the dev forgot to consider FilterList), So if a user use INCLUDE_AND_SEEK_NEXT_ROW
in his own Filter and wrapped by a FilterList,  it'll  throw  an IllegalStateException("Received
code is not valid."). 
> 2.  For FilterList with MUST_PASS_ONE,   if filter-A in filter list return  INCLUDE and
filter-B in filter list return INCLUDE_AND_NEXT_COL,   the FilterList will return  INCLUDE_AND_NEXT_COL
finally.  According to the mininal step rule , It's incorrect.  (filter list with MUST_PASS_ONE
choose the mininal step among filters in filter list. Let's call it: The Mininal Step Rule).

This message was sent by Atlassian JIRA

View raw message