[
https://issues.apache.org/jira/browse/HADOOP-1531?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12508609
]
Jim Kellerman commented on HADOOP-1531:
---------------------------------------
In general, I have no objections to this change.
However, I do have a couple of comments:
- in HClient, the constructor for ClientScanner that does not take a filter is
no longer needed since the constructor is only called from obtainScanner and
obtainScanner(columns,startRow) just calls
obtainScanner(columns,startRow,filter) specifying null for the filter.
- in HClient.ClientScanner, shouldn't the call to server.openScanner be
conditionalized so that it either calls the HRegionServerInterface.openScanner
which takes a scanner or the one which does not? (since you can't pass a null
over an rpc) For example:
try {
if(this.filter == null) {
this.scannerId = this.server.openScanner(info.regionInfo.regionName,
this.columns, currentRegion == 0 ? this.startRow : EMPTY_START_ROW);
} else {
this.scannerId = this.server.openScanner(info.regionInfo.regionName,
this.columns, currentRegion == 0 ? this.startRow : EMPTY_START_ROW,
filter);
}
- finally, I would like to see a test case that uses a filter The existing
tests will ensure that there are no regressions.
> Add RowFilter to HRegion.HScanner
> ---------------------------------
>
> Key: HADOOP-1531
> URL: https://issues.apache.org/jira/browse/HADOOP-1531
> Project: Hadoop
> Issue Type: Improvement
> Components: contrib/hbase
> Affects Versions: 0.14.0
> Reporter: James Kennedy
> Assignee: James Kennedy
> Attachments: RowFilter.patch
>
>
> I've implemented a RowFilterInterface and a RowFilter implementation. This
> is passed to the HRegion.HScanner via HClient.openScanner() though it is an
> entirely optional parameter.
> HScanner applies the filter in the next() call by iterating until it
> encounters a row that is not filtered by the RowFilter. The filter applies
> criteria based on row keys and/or column data values.
> Null values are little tricky since the resultSet in that loop may represent
> nulls as absent columns or as DELETED_BYTES. Nevertheless null cases are
> taken care of by the filter and you can for example retrieve all rows where
> column X = null.
> The initial RowFilter implementation is limited in several ways:
> * Equality test only with literal values. No !=, <, >, etc. No col1 == col2.
> This is a straight-up byte[] comparison.
> * Multiple column criteria are treated as an implicit conjunction, no
> disjunction possible.
> * row key criteria is a regular expression only
> * row key criteria is independent of column criteria. No "if
> rowkey.matches(A) and col1==B" although the interface is created to allow
> for that.
> But it should be easy to write an improved RowFilterInterface implementation
> to take care of most of the above without having to change code elsewhere.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
