HBase, mail # user - scan filtering column familly return wrong cell

scan filtering column familly return wrong cell

Damien Hardy 2012-11-09, 15:59

Hello,

I am a bit confused here...

I try to execute a M/R to import data in HBase table 'Consultation'.

Running on CDH4.1.2

map function emits context.write(ImmutableBytesWritable, KeyValue)

conf summary :
   job.setOutputFormatClass(TableOutputFormat.class);
   job.setInputFormatClass(DataDrivenDBInputFormat.class);
   job.getConfiguration().set(TableOutputFormat.OUTPUT_TABLE,
"Consultation");
   job.setOutputKeyClass(ImmutableBytesWritable.class);
   job.setOutputValueClass(KeyValue.class);
The reduce class is :

  static class ImportReducer
  extends TableReducer<ImmutableBytesWritable, KeyValue,
ImmutableBytesWritable> {
   @Override
   public void reduce(ImmutableBytesWritable row, Iterable<KeyValue> kvs,
Reducer<ImmutableBytesWritable, KeyValue, ImmutableBytesWritable,
Writable>.Context context)
   throws java.io.IOException, InterruptedException {
   Put p = new Put(row.copyBytes());
   int i = 0;
   byte[] rk = null;
   for (KeyValue kv: kvs) {
   p.add(kv);
   if ( Bytes.compareTo(CF_VISITED, 0, CF_VISITED.length,
kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength() ) == 0 ) {
   i++;
   }
   }
   p.add(CF_COUNTER,QA_COUNTER,Bytes.toBytes(i));
   context.write(new ImmutableBytesWritable(row),p);
   }
  }
hbase(main):038:0> scan 'Consultation', {COLUMNS=> *'visiting_tl'*, LIMIT
=> 10 }
ROW
COLUMN+CELL

00070db1aa26d1906a078a1e03f788cb-\x00\x13\x80\x15 column=*
visited_tl:*\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7,
timestamp=1266998781000,
value=\x00\x00\x00\x00

001316263fc8b454bbd86dff1587a347-\x00>t\x05 column=*
visited_tl:*\x7F\xFF\xFE\xD7\x0F\xB8u_\x00\x08\xE1\xA0,
timestamp=1275341540000,
value=\x00\x00\x00\x00

001497e68d7c71a3cd281860484fa6be-\x00/\x0E^ column=*
visited_tl:*\x7F\xFF\xFE\xD8\x06\x9B\xB0\xB7\x00(3S,
timestamp=1271199453000,
value=\x00\x00\x00\x00

001845aac2462a1c24b36eb90ab698cf-\x00\x04\x1E\xF5 column=*
visited_tl:*\x7F\xFF\xFE\xD6\xA8\xB9-\xEF\x002Po, timestamp=1277069546000,
value=\x00\x00\x00\x01

0019cec2c1f38c42b1c540ef7708c6a9-\x00;\xE0\x97 column=*
visited_tl:*\x7F\xFF\xFE\xD8\xF9\xC7\x0C_\x00\x02?.,
timestamp=1267119748000,
value=\x00\x00\x00\x00

001de6b92754b0ef44ee10bf2bdfe3c3-\x00%\x1AV column=*
visited_tl:*\x7F\xFF\xFE\xD6\xE4H\x99\xC7\x00\x0F\x7F9,
timestamp=1276070291000,
value=\x00\x00\x00\x01

00217f082f96eb12108c139b99a3ccb7-\x00\x02w\x08 column=*
visited_tl:*\x7F\xFF\xFE\xD8\xEB\x1B\x95\xEF\x00\x0A7\x19,
timestamp=1267365866000,
value=\x00\x00\x00\x00

0021cbfd559f56dd298e4b4fee7626a9-\x00r\xBF\xFA column=*
visited_tl:*\x7F\xFF\xFE\xD6\xA1\x0B-\x0F\x00\x03\xBC\x8B,
timestamp=1277198390000,
value=\x00\x00\x00\x02

00266c02a60f9a6efb5d24317e6032a0-\x00\x0E]+ column=*
visited_tl:*\x7F\xFF\xFE\xD6\xBC\x0D\xD1\x7F\x00/ q,
timestamp=1276745232000,
value=\x00\x00\x00\x01

0026dbbd6562da5b79f1b09e94e3b973-\x00C[\x93 column=*
visited_tl:*\x7F\xFF\xFE\xD7\xB0\xFA\xB7/\x00\x02~\x09,
timestamp=1272636066000,
value=\x00\x00\x00\x01

10 row(s) in 2.1130 seconds
hbase(main):036:0> get 'Consultation',
"00070db1aa26d1906a078a1e03f788cb-\x00\x13\x80\x15"
COLUMN
CELL

*visited_tl:\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7*
timestamp=1266998781000,
value=\x00\x00\x00\x00

*visited_tl:\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7*
timestamp=1266998781000,
value=\x00\x00\x00\x00

visits_count:_counter
timestamp=1352475456545,
value=\x00\x00\x02\xA1

3 row(s) in 0.3260 seconds

hbase(main):037:0> get 'Consultation',
"00070db1aa26d1906a078a1e03f788cb-\x00\x13\x80\x15", *'visiting_tl:'*
COLUMN
CELL

*visited_tl:*\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7
timestamp=1266998781000,
value=\x00\x00\x00\x00

1 row(s) in 0.1650 seconds

So I have 3 problems :

* table is only 1 VERSION enable : who can I get the cell
visited_tl:\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7 2 time for a
single row ?
* when I explicitly query for CF 'visiting_tl:' , I get a 'visited_tl:'
cell ... WTF ?
* the Counter is (int)673 ... where are my 673 visited_tl cell ? (673 is
the good value according to my source)

Cheers,

Damien HARDY
IT Infrastructure Architect

Viadeo - 30 rue de la Victoire - 75009 Paris - France
T : +33 1 80 48 39 73 – F : +33 1 42 93 22 56

Re: scan filtering column familly return wrong cell

Damien Hardy 2012-11-09, 16:52

Ok I can reply to myself ...

you have to add a clone of the KeyValue in the Put. So
p.add(kv);
becomes
p.add(kv.clone());

If not, I suppose only the last one is added in HBase (but the result is
quite weird and should be fixed IMO)

Cheers,

--
Damien
2012/11/9 Damien Hardy <[EMAIL PROTECTED]>

> Hello,
>
> I am a bit confused here...
>
> I try to execute a M/R to import data in HBase table 'Consultation'.
>
> Running on CDH4.1.2
>
> map function emits context.write(ImmutableBytesWritable, KeyValue)
>
> conf summary :
> job.setOutputFormatClass(TableOutputFormat.class);
> job.setInputFormatClass(DataDrivenDBInputFormat.class);
> job.getConfiguration().set(TableOutputFormat.OUTPUT_TABLE,
> "Consultation");
> job.setOutputKeyClass(ImmutableBytesWritable.class);
> job.setOutputValueClass(KeyValue.class);
>
>
> The reduce class is :
>
> static class ImportReducer
> extends TableReducer<ImmutableBytesWritable, KeyValue,
> ImmutableBytesWritable> {
> @Override
> public void reduce(ImmutableBytesWritable row, Iterable<KeyValue> kvs,
> Reducer<ImmutableBytesWritable, KeyValue, ImmutableBytesWritable,
> Writable>.Context context)
> throws java.io.IOException, InterruptedException {
> Put p = new Put(row.copyBytes());
> int i = 0;
> byte[] rk = null;
> for (KeyValue kv: kvs) {
> p.add(kv);
> if ( Bytes.compareTo(CF_VISITED, 0, CF_VISITED.length,
> kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength() ) == 0 ) {
> i++;
> }
> }
> p.add(CF_COUNTER,QA_COUNTER,Bytes.toBytes(i));
> context.write(new ImmutableBytesWritable(row),p);
> }
> }
>
>
> hbase(main):038:0> scan 'Consultation', {COLUMNS=> *'visiting_tl'*, LIMIT
> => 10 }
> ROW
> COLUMN+CELL
>
> 00070db1aa26d1906a078a1e03f788cb-\x00\x13\x80\x15 column=*
> visited_tl:*\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7,
> timestamp=1266998781000,
> value=\x00\x00\x00\x00
>
> 001316263fc8b454bbd86dff1587a347-\x00>t\x05 column=*
> visited_tl:*\x7F\xFF\xFE\xD7\x0F\xB8u_\x00\x08\xE1\xA0,
> timestamp=1275341540000,
> value=\x00\x00\x00\x00
>
> 001497e68d7c71a3cd281860484fa6be-\x00/\x0E^ column=*
> visited_tl:*\x7F\xFF\xFE\xD8\x06\x9B\xB0\xB7\x00(3S,
> timestamp=1271199453000,
> value=\x00\x00\x00\x00
>
> 001845aac2462a1c24b36eb90ab698cf-\x00\x04\x1E\xF5 column=*
> visited_tl:*\x7F\xFF\xFE\xD6\xA8\xB9-\xEF\x002Po,
> timestamp=1277069546000,
> value=\x00\x00\x00\x01
>
> 0019cec2c1f38c42b1c540ef7708c6a9-\x00;\xE0\x97 column=*
> visited_tl:*\x7F\xFF\xFE\xD8\xF9\xC7\x0C_\x00\x02?.,
> timestamp=1267119748000,
> value=\x00\x00\x00\x00
>
> 001de6b92754b0ef44ee10bf2bdfe3c3-\x00%\x1AV column=*
> visited_tl:*\x7F\xFF\xFE\xD6\xE4H\x99\xC7\x00\x0F\x7F9,
> timestamp=1276070291000,
> value=\x00\x00\x00\x01
>
> 00217f082f96eb12108c139b99a3ccb7-\x00\x02w\x08 column=*
> visited_tl:*\x7F\xFF\xFE\xD8\xEB\x1B\x95\xEF\x00\x0A7\x19,
> timestamp=1267365866000,
> value=\x00\x00\x00\x00
>
> 0021cbfd559f56dd298e4b4fee7626a9-\x00r\xBF\xFA column=*
> visited_tl:*\x7F\xFF\xFE\xD6\xA1\x0B-\x0F\x00\x03\xBC\x8B,
> timestamp=1277198390000,
> value=\x00\x00\x00\x02
>
> 00266c02a60f9a6efb5d24317e6032a0-\x00\x0E]+ column=*
> visited_tl:*\x7F\xFF\xFE\xD6\xBC\x0D\xD1\x7F\x00/ q,
> timestamp=1276745232000,
> value=\x00\x00\x00\x01
>
> 0026dbbd6562da5b79f1b09e94e3b973-\x00C[\x93 column=*
> visited_tl:*\x7F\xFF\xFE\xD7\xB0\xFA\xB7/\x00\x02~\x09,
> timestamp=1272636066000,
> value=\x00\x00\x00\x01
>
> 10 row(s) in 2.1130 seconds
>
>
> hbase(main):036:0> get 'Consultation',
> "00070db1aa26d1906a078a1e03f788cb-\x00\x13\x80\x15"
> COLUMN
> CELL
>
> *visited_tl:\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7*
> timestamp=1266998781000,
> value=\x00\x00\x00\x00
>
> *visited_tl:\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7*
> timestamp=1266998781000,
> value=\x00\x00\x00\x00
>
> visits_count:_counter
Damien HARDY
IT Infrastructure Architect

Viadeo - 30 rue de la Victoire - 75009 Paris - France
T : +33 1 80 48 39 73 – F : +33 1 42 93 22 56

Re: scan filtering column familly return wrong cell

Varun Sharma 2012-11-11, 20:46

I have not look at this in detail but does this eventually use the
ColumnCountGetFilter - if yes, then this will actually also include upto
one older version since filters run before version tracking - see JIRA
https://issues.apache.org/jira/browse/HBASE-5257 which has a fix - Remember
that versions are always kept in memstore and only cleaned up when memstore
is flushed out as an HFile.

On Fri, Nov 9, 2012 at 8:52 AM, Damien Hardy <[EMAIL PROTECTED]> wrote:

> Ok I can reply to myself ...
>
> you have to add a clone of the KeyValue in the Put. So
> p.add(kv);
> becomes
> p.add(kv.clone());
>
> If not, I suppose only the last one is added in HBase (but the result is
> quite weird and should be fixed IMO)
>
> Cheers,
>
> --
> Damien
>
>
> 2012/11/9 Damien Hardy <[EMAIL PROTECTED]>
>
> > Hello,
> >
> > I am a bit confused here...
> >
> > I try to execute a M/R to import data in HBase table 'Consultation'.
> >
> > Running on CDH4.1.2
> >
> > map function emits context.write(ImmutableBytesWritable, KeyValue)
> >
> > conf summary :
> > job.setOutputFormatClass(TableOutputFormat.class);
> > job.setInputFormatClass(DataDrivenDBInputFormat.class);
> > job.getConfiguration().set(TableOutputFormat.OUTPUT_TABLE,
> > "Consultation");
> > job.setOutputKeyClass(ImmutableBytesWritable.class);
> > job.setOutputValueClass(KeyValue.class);
> >
> >
> > The reduce class is :
> >
> > static class ImportReducer
> > extends TableReducer<ImmutableBytesWritable, KeyValue,
> > ImmutableBytesWritable> {
> > @Override
> > public void reduce(ImmutableBytesWritable row, Iterable<KeyValue>
> kvs,
> > Reducer<ImmutableBytesWritable, KeyValue, ImmutableBytesWritable,
> > Writable>.Context context)
> > throws java.io.IOException, InterruptedException {
> > Put p = new Put(row.copyBytes());
> > int i = 0;
> > byte[] rk = null;
> > for (KeyValue kv: kvs) {
> > p.add(kv);
> > if ( Bytes.compareTo(CF_VISITED, 0, CF_VISITED.length,
> > kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength() ) == 0 ) {
> > i++;
> > }
> > }
> > p.add(CF_COUNTER,QA_COUNTER,Bytes.toBytes(i));
> > context.write(new ImmutableBytesWritable(row),p);
> > }
> > }
> >
> >
> > hbase(main):038:0> scan 'Consultation', {COLUMNS=> *'visiting_tl'*, LIMIT
> > => 10 }
> > ROW
> > COLUMN+CELL
> >
> > 00070db1aa26d1906a078a1e03f788cb-\x00\x13\x80\x15 column=*
> > visited_tl:*\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7,
> > timestamp=1266998781000,
> > value=\x00\x00\x00\x00
> >
> > 001316263fc8b454bbd86dff1587a347-\x00>t\x05 column=*
> > visited_tl:*\x7F\xFF\xFE\xD7\x0F\xB8u_\x00\x08\xE1\xA0,
> > timestamp=1275341540000,
> > value=\x00\x00\x00\x00
> >
> > 001497e68d7c71a3cd281860484fa6be-\x00/\x0E^ column=*
> > visited_tl:*\x7F\xFF\xFE\xD8\x06\x9B\xB0\xB7\x00(3S,
> > timestamp=1271199453000,
> > value=\x00\x00\x00\x00
> >
> > 001845aac2462a1c24b36eb90ab698cf-\x00\x04\x1E\xF5 column=*
> > visited_tl:*\x7F\xFF\xFE\xD6\xA8\xB9-\xEF\x002Po,
> > timestamp=1277069546000,
> > value=\x00\x00\x00\x01
> >
> > 0019cec2c1f38c42b1c540ef7708c6a9-\x00;\xE0\x97 column=*
> > visited_tl:*\x7F\xFF\xFE\xD8\xF9\xC7\x0C_\x00\x02?.,
> > timestamp=1267119748000,
> > value=\x00\x00\x00\x00
> >
> > 001de6b92754b0ef44ee10bf2bdfe3c3-\x00%\x1AV column=*
> > visited_tl:*\x7F\xFF\xFE\xD6\xE4H\x99\xC7\x00\x0F\x7F9,
> > timestamp=1276070291000,
> > value=\x00\x00\x00\x01
> >
> > 00217f082f96eb12108c139b99a3ccb7-\x00\x02w\x08 column=*
> > visited_tl:*\x7F\xFF\xFE\xD8\xEB\x1B\x95\xEF\x00\x0A7\x19,
> > timestamp=1267365866000,
> > value=\x00\x00\x00\x00
> >
> > 0021cbfd559f56dd298e4b4fee7626a9-\x00r\xBF\xFA column=*
> > visited_tl:*\x7F\xFF\xFE\xD6\xA1\x0B-\x0F\x00\x03\xBC\x8B,
> > timestamp=1277198390000,
> > value=\x00\x00\x00\x02
> >
> > 00266c02a60f9a6efb5d24317e6032a0-\x00\x0E]+ column=*

Re: scan filtering column familly return wrong cell

Damien Hardy 2012-11-12, 09:38

I don't know if HBase shell scan command use ColumnCountGetFilter.
The absence of compaction could explain the 2 same cell displayed.
But when I filter on one colfam, I get only 1 cell ... from the wrong
colfam (like if the cell is stored in the wrong HFile) ...

When I add clone of my KeyValues in my Put in reduce the data is well
writen (I get my 2 colfam filled).

It sound strange that client mapReduce can set such a mess in the storage...

Regards,

--
Damien

2012/11/11 Varun Sharma <[EMAIL PROTECTED]>

> I have not look at this in detail but does this eventually use the
> ColumnCountGetFilter - if yes, then this will actually also include upto
> one older version since filters run before version tracking - see JIRA
> https://issues.apache.org/jira/browse/HBASE-5257 which has a fix -
> Remember
> that versions are always kept in memstore and only cleaned up when memstore
> is flushed out as an HFile.
>
> On Fri, Nov 9, 2012 at 8:52 AM, Damien Hardy <[EMAIL PROTECTED]>
> wrote:
>
> > Ok I can reply to myself ...
> >
> > you have to add a clone of the KeyValue in the Put. So
> > p.add(kv);
> > becomes
> > p.add(kv.clone());
> >
> > If not, I suppose only the last one is added in HBase (but the result is
> > quite weird and should be fixed IMO)
> >
> > Cheers,
> >
> > --
> > Damien
> >
> >
> > 2012/11/9 Damien Hardy <[EMAIL PROTECTED]>
> >
> > > Hello,
> > >
> > > I am a bit confused here...
> > >
> > > I try to execute a M/R to import data in HBase table 'Consultation'.
> > >
> > > Running on CDH4.1.2
> > >
> > > map function emits context.write(ImmutableBytesWritable, KeyValue)
> > >
> > > conf summary :
> > > job.setOutputFormatClass(TableOutputFormat.class);
> > > job.setInputFormatClass(DataDrivenDBInputFormat.class);
> > > job.getConfiguration().set(TableOutputFormat.OUTPUT_TABLE,
> > > "Consultation");
> > > job.setOutputKeyClass(ImmutableBytesWritable.class);
> > > job.setOutputValueClass(KeyValue.class);
> > >
> > >
> > > The reduce class is :
> > >
> > > static class ImportReducer
> > > extends TableReducer<ImmutableBytesWritable, KeyValue,
> > > ImmutableBytesWritable> {
> > > @Override
> > > public void reduce(ImmutableBytesWritable row, Iterable<KeyValue>
> > kvs,
> > > Reducer<ImmutableBytesWritable, KeyValue, ImmutableBytesWritable,
> > > Writable>.Context context)
> > > throws java.io.IOException, InterruptedException {
> > > Put p = new Put(row.copyBytes());
> > > int i = 0;
> > > byte[] rk = null;
> > > for (KeyValue kv: kvs) {
> > > p.add(kv);
> > > if ( Bytes.compareTo(CF_VISITED, 0, CF_VISITED.length,
> > > kv.getBuffer(), kv.getFamilyOffset(), kv.getFamilyLength() ) == 0 ) {
> > > i++;
> > > }
> > > }
> > > p.add(CF_COUNTER,QA_COUNTER,Bytes.toBytes(i));
> > > context.write(new ImmutableBytesWritable(row),p);
> > > }
> > > }
> > >
> > >
> > > hbase(main):038:0> scan 'Consultation', {COLUMNS=> *'visiting_tl'*,
> LIMIT
> > > => 10 }
> > > ROW
> > > COLUMN+CELL
> > >
> > > 00070db1aa26d1906a078a1e03f788cb-\x00\x13\x80\x15 column=*
> > > visited_tl:*\x7F\xFF\xFE\xD9\x00\xFC\xDB\xB7\x001\xC5\xA7,
> > > timestamp=1266998781000,
> > > value=\x00\x00\x00\x00
> > >
> > > 001316263fc8b454bbd86dff1587a347-\x00>t\x05 column=*
> > > visited_tl:*\x7F\xFF\xFE\xD7\x0F\xB8u_\x00\x08\xE1\xA0,
> > > timestamp=1275341540000,
> > > value=\x00\x00\x00\x00
> > >
> > > 001497e68d7c71a3cd281860484fa6be-\x00/\x0E^ column=*
> > > visited_tl:*\x7F\xFF\xFE\xD8\x06\x9B\xB0\xB7\x00(3S,
> > > timestamp=1271199453000,
> > > value=\x00\x00\x00\x00
> > >
> > > 001845aac2462a1c24b36eb90ab698cf-\x00\x04\x1E\xF5 column=*
> > > visited_tl:*\x7F\xFF\xFE\xD6\xA8\xB9-\xEF\x002Po,
> > > timestamp=1277069546000,
> > > value=\x00\x00\x00\x01
> > >
> > > 0019cec2c1f38c42b1c540ef7708c6a9-\x00;\xE0\x97 column=*
> > > visited_tl:*\x7F\xFF\xFE\xD8\xF9\xC7\x0C_\x00\x02?.,