[
https://issues.apache.org/jira/browse/HBASE-20565?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16471467#comment-16471467
]
Zheng Hu commented on HBASE-20565:
----------------------------------
Add some log by debug.diff, and the result is shown in debug.log. I think the
bug is not from FilterList, but is from ColumnPaginationFilter#filterKeyValue.
{code}
public ReturnCode filterKeyValue(Cell v)
{
if (columnOffset != null) {
if (count >= limit) {
return ReturnCode.NEXT_ROW;
}
byte[] buffer = v.getQualifierArray();
if (buffer == null) {
return ReturnCode.SEEK_NEXT_USING_HINT;
}
int cmp = 0;
// Only compare if no KV's have been seen so far.
if (count == 0) {
cmp = Bytes.compareTo(buffer,
v.getQualifierOffset(),
v.getQualifierLength(),
this.columnOffset,
0,
this.columnOffset.length);
}
if (cmp < 0) {
return ReturnCode.SEEK_NEXT_USING_HINT;
} else {
count++;
return ReturnCode.INCLUDE_AND_NEXT_COL;
}
} else {
if (count >= offset + limit) {
return ReturnCode.NEXT_ROW;
}
ReturnCode code = count < offset ? ReturnCode.NEXT_COL :
ReturnCode.INCLUDE_AND_NEXT_COL;
count++; <------------------- the count increment even if return
NEXT_COL.
return code;
}
}
{code}
The count increment even if we return NEXT_COL, so after the filter checked
the column=Family:0, count=1, when filter is checking the column=Family:5,
the count is 5 now, so return a NEXT_ROW...
we should define our count as the number of included column in
ColumnPaginationFilter (just as the def in if (columnOffset != null) {} ),
rather than the index of column we encountered. ..
> ColumnRangeFilter combined with ColumnPaginationFilter can produce incorrect
> result since 1.4
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-20565
> URL: https://issues.apache.org/jira/browse/HBASE-20565
> Project: HBase
> Issue Type: Bug
> Components: Filters
> Affects Versions: 1.4.4
> Reporter: Jerry He
> Assignee: Zheng Hu
> Priority: Major
> Attachments: debug.diff, debug.log, test-branch-1.4.patch
>
>
> When ColumnPaginationFilter is combined with ColumnRangeFilter, we may see
> incorrect result.
> Here is a simple example.
> One row with 10 columns c0, c1, c2, .., c9. I have a ColumnRangeFilter for
> range c2 to c9. Then I have a ColumnPaginationFilter with limit 5 and offset
> 0. FileterList is FilterList(Operator.MUST_PASS_ALL, ColumnRangeFilter,
> ColumnPaginationFilter).
> We expect 5 columns being returned. But in HBase 1.4 and after, 4 columns
> are returned.
> In 1.2.x, the correct 5 columns are returned.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
