hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter Wolf <opus...@gmail.com>
Subject Re: Scan.addFamiliy reduces results -- Ah Ha!
Date Thu, 15 Mar 2012 19:41:11 GMT
No, no... I found it, and it is not the filter.

Here is my Maven dependencies

<dependency>
<groupId>org.apache.hbase</groupId>
<artifactId>hbase</artifactId>
<version>0.90.4-cdh3u3</version>
</dependency>
<dependency>
<groupId>org.apache.hadoop</groupId>
<artifactId>hadoop-core</artifactId>
<version>0.20.2-cdh3u3</version>
</dependency>
<dependency>
<groupId>org.apache.zookeeper</groupId>
<artifactId>zookeeper</artifactId>
<version>3.3.4-cdh3u3</version>
</dependency>


Here is my actual code (nothing too secret)

         Scan scan = new Scan(calculateStartRowKey(nativeLanguage, 
accountID), calculateEndRowKey(nativeLanguage, accountID));
         scan
                 .addFamily(metaFamily)
                 .addFamily(request1Family)
                 .addFamily(request2Family)
                 .addFamily(serverFamily)
                 .addFamily(queueFamily)
                 .addFamily(scoreFamily)
                 .addFamily(audioFamily)
                 .addFamily(wordsFamily)
-->           //.addFamily(researchFamily) <--
         ;
         Filter filter = new SingleColumnValueFilter(request1Family, 
request1AccountID, CompareFilter.CompareOp.EQUAL, Bytes.toBytes(accountID));
         scan.setFilter(filter);
-->     scan.setCaching(10000); <--
         scanSessions(processor, scan);

         ResultScanner scanner = null;
         try {
             scanner = hTable.getScanner(scan);
             try {
                 Iterator<Result> it = scanner.iterator();
                 while (it.hasNext()) {
                     try {
                         Result result = it.next();
                         process(result);
                     } catch (Throwable e) {
                         System.out.println("WARNING: " + e);
                     }
                 }
             } finally {
                 scanner.close();
             }
         } catch (IOException e) {
             new RuntimeException(e);
         }



Notice the two lines marked by -->

If I uncomment the .addFamily(researchFamily) I get 4258 results, if I 
comment it out I get 24258.  The difference is exactly 20000.

Notice the setCaching(10000).  If I change it to 1000, I get 24258 results

There seems to be a connection between the caching value and the number 
of results returned.  Furthermore, adding more data to the results by 
adding families reduces the number of results.  Note that the data in my 
researchFamily is quite large.

Did I find a bug?

Peter





On 3/15/12 3:25 PM, Daniel Gómez Ferro wrote:
> As I told you in the other message, if you don't addColumn() the column you are filtering
on, by default it will return any row that doesn't contain the said column: http://hbase.apache.org/apidocs/org/apache/hadoop/hbase/filter/SingleColumnValueFilter.html#setFilterIfMissing(boolean)
>
> So when you uncomment the addColumn(), the filter kicks in and actually filters values.
When the addColumn() is commented, all rows are returned.
>
> On Mar 15, 2012, at 20:05 , Peter Wolf wrote:
>
>> Huh!  That's what I was afraid you'd say.  I'm still confused :-(
>>
>> If "it will give all rows that contain _any_ of these families", then
>> why does adding a family give me *less* rows?
>>
>> Leaving my row start/stop and filtering code constant, and just
>> un-commenting an addFamily() dramatically reduces the number of results
>> returned from a scan.
>>
>> P
>>
>>
>>
>> On 3/15/12 2:42 PM, Himanshu Vashishtha wrote:
>>> " Let's also say there are 1000 rows with A,B,C and 500 rows with only B and
C.
>>>
>>> If I add families A, B and C and scan with no filter will I get 1500,
>>> 1000 or 500 results?"
>>>
>>> In this case, you will get 1000 rows. In case you add only B, you will
>>> get 500 rows.
>>>
>>> It's not like if you add families A, B and C, it will give you _only_
>>> those rows that have _all_ three families; rather it will give all
>>> rows that contain _any_ of these families.
>>>
>>> Hope this helps.
>>>
>>> Experts are welcome to chime in if I am missing something :)
>>>
>>> Thanks,
>>> Himanshu
>>>
>>>
>>> On Thu, Mar 15, 2012 at 11:48 AM, Peter Wolf<opus111@gmail.com>   wrote:
>>>> Hi Lars, still confused...
>>>>
>>>> My table *should* have values for families A, B and C.  Let's say I have
a
>>>> bug, and some rows only have values for B and C.  Let's also say there are
>>>> 1000 rows with A,B,C and 500 rows with only B and C.
>>>>
>>>> If I add families A, B and C and scan with no filter will I get 1500, 1000
>>>> or 500 results?
>>>>
>>>> Many thanks
>>>> P
>>>>
>>>>
>>>>
>>>>
>>>> On 3/15/12 1:17 PM, lars hofhansl wrote:
>>>>> Hi haijia,
>>>>>
>>>>> In that case HBase will still return the data for columns in family B
and
>>>>> C.But if you only added family A then HBase would only return "rows"
for
>>>>> which family A has any columns.
>>>>>
>>>>> -- Lars
>>>>> ________________________________
>>>>>
>>>>> From: Haijia Zhou<leonster@gmail.com>
>>>>> To: user@hbase.apache.org; lars hofhansl<lhofhansl@yahoo.com>
>>>>> Sent: Thursday, March 15, 2012 10:12 AM
>>>>> Subject: Re: Scan.addFamiliy reduces results
>>>>>
>>>>>
>>>>> I have the same confusion. Say if I added three column families A, B
anc C
>>>>> to the scan, now if a row has data for column family B and C but no data
for
>>>>> A, then it won't be returned  in the next() method?
>>>>> What if the requirement is to get row data regardless of whether there's
>>>>> data for a specific column family or not?
>>>>>
>>>>>
>>>>> On Thu, Mar 15, 2012 at 1:04 PM, lars hofhansl<lhofhansl@yahoo.com>
>>>>>   wrote:
>>>>>
>>>>> Hi Peter,
>>>>>> for HBase you have keep in mind that it is a sparse columnar (or
>>>>>> KeyValue) store: (rowkey, columnfamily, column, TS) ->     value
>>>>>>
>>>>>> A scan only returns those KeyValues that match the scan. So when
you set
>>>>>> families on your scan you'll only get those rows for which the scan
found
>>>>>> any columns.
>>>>>>
>>>>>> Makes sense?
>>>>>>
>>>>>> -- Lars
>>>>>>
>>>>>>
>>>>>>
>>>>>> ________________________________
>>>>>>   From: Peter Wolf<opus111@gmail.com>
>>>>>> To: user@hbase.apache.org
>>>>>> Sent: Thursday, March 15, 2012 9:52 AM
>>>>>> Subject: Re: Scan.addFamiliy reduces results
>>>>>>
>>>>>>
>>>>>> Thanks Doug,
>>>>>>
>>>>>> I had read that, and I just read it again.  But I am missing something...
>>>>>>
>>>>>> Why does adding a family reduce the number of results?  Is there
an
>>>>>> implied filter of some form?  Does addFamily add some constraint
on
>>>>>> which rows are returned?
>>>>>>
>>>>>> Note that all my rows *ought* to have values in all the families.
>>>>>>
>>>>>> Thanks
>>>>>> Peter
>>>>>>
>>>>>> On 3/15/12 12:39 PM, Doug Meil wrote:
>>>>>>> re:  "However, I am getting different number of results, depending
on
>>>>>>> which families are added"
>>>>>>>
>>>>>>> Yes.
>>>>>>>
>>>>>>> I'd suggest you read this in the RefGuide.
>>>>>>>
>>>>>>> http://hbase.apache.org/book.html#datamodel
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On 3/15/12 12:08 PM, "Peter Wolf"<opus111@gmail.com>  
   wrote:
>>>>>>>
>>>>>>>> Hi all,
>>>>>>>>
>>>>>>>> I am doing a scan on a table with multiple families.  My
code looks
>>>>>>>> like
>>>>>>>> this...
>>>>>>>>
>>>>>>>>            Scan scan = new Scan(calculateStartRowKey(a),
>>>>>>>> calculateEndRowKey(b));
>>>>>>>>
>>>>>>>>            scan.setCaching(10000);
>>>>>>>>            Filter filter = new SingleColumnValueFilter(xFamily,
xColumn,
>>>>>>>> CompareFilter.CompareOp.EQUAL, Bytes.toBytes(x));
>>>>>>>>            scan.setFilter(filter);
>>>>>>>>            scan
>>>>>>>>                    .addFamily(xFamily)
>>>>>>>>                    .addFamily(yFamily)
>>>>>>>>                    .addFamily(zFamily);
>>>>>>>>
>>>>>>>>            ResultScanner scanner = hTable.getScanner(scan);
>>>>>>>>
>>>>>>>>            Iterator<Result>      it = scanner.iterator();
>>>>>>>>            int resultCount = 0;
>>>>>>>>            while (it.hasNext()) {
>>>>>>>>                  Result result = it.next();
>>>>>>>>
>>>>>>>>                  resultCount++;
>>>>>>>>            }
>>>>>>>>
>>>>>>>> However, I am getting different number of results, depending
on which
>>>>>>>> families are added.  For example these give different result
counts
>>>>>>>>
>>>>>>>>            scan
>>>>>>>>                    //.addFamily(xFamily)
>>>>>>>>                    .addFamily(yFamily)
>>>>>>>>                    .addFamily(zFamily);
>>>>>>>> and
>>>>>>>>            scan
>>>>>>>>                    .addFamily(xFamily)
>>>>>>>>                    .addFamily(yFamily)
>>>>>>>>                    .addFamily(zFamily);
>>>>>>>>
>>>>>>>>
>>>>>>>> There is no error message, and I don't see anything in the
Scan
>>>>>>>> documentation.  Does anyone know what is going on?
>>>>>>>>
>>>>>>>> Thanks
>>>>>>>> Peter
>>>>>>>>
>>>>>>>>
>>>>>>>>


Mime
View raw message