hadoop-hdfs-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "David Parks" <davidpark...@yahoo.com>
Subject RE: Question related to Decompressor interface
Date Mon, 11 Feb 2013 03:24:28 GMT
In the EncryptedWritableWrapper idea you would create an object that takes
any Writable object as it's parameter. 


Your EncryptedWritableWrapper would naturally implement Writable.


.         When write(DataOutput out) is called on your object, create your
own DataOutputStream which reads data into a byte array that you control
(i.e. new DataOutputStream(new myByteArrayOutputStream()), keeping
references to the objects of course).

.         Now encrypt the bytes and pass them on to the DataOutput object
you received in write(DataOutput out)


To decrypt is basically the same with the readFields(DataInput in) method.

.         Read in the bytes and decrypt them (you will probably have needed
to write out the length of bytes previously so you know how much to read

.         Take the decrypted bytes and pass them to the readFields(.) method
of the Writable object you're wrapping 


The rest of Hadoop doesn't know or care if the data is encrypted, your
Writable objects are just a bunch of bytes, you're Key and Value class in
this case are now EncryptedWritableWrapper, and you'll need to know which
type of Writable to pass it in the code.


This would be good for encrypting in Hadoop. If your file comes in encrypted
then it necessarily can't be split (you should aim to limit the maximum size
of the file on the source side). In the case of an encrypted input you would
need your own record reader to decrypt it, your description of the scenario
below is correct, extending TextinputFormat would be the way to go.


If your input is just a plain text file and your goal is to store it in an
encrypted fashion then the EncryptedWritable idea works and is a more simple




From: java8964 java8964 [mailto:java8964@hotmail.com] 
Sent: Sunday, February 10, 2013 10:13 PM
To: user@hadoop.apache.org
Subject: RE: Question related to Decompressor interface


Hi, Dave:


Thanks for you reply. I am not sure how the EncryptedWritable will work, can
you share more ideas about it?


For example, if I have a text file as my source raw file. Now I need to
store it in HDFS. If I use any encryption to encrypt the whole file, then
there is no good InputFormat or RecordReader to process it, unless whole
file is decrypted first at runtime, then using TextInputFormat to process
it, right?


What you suggest is  when I encrypted the file, store it as a SequenceFile,
using anything I want as the key, then encrypt each line (Record), and
stores it as the value, put both (key, value) pair into the sequence file,
is that right? 


Then in the runtime, each value can be decrypted from the sequence file, and
ready for next step in the by the EncryptedWritable class. Is my
understanding correct?


 In this case, of course I don't need to worry about split any more, as each
record is encrypted/decrypted separately.


I think it is a valid option, but problem is that the data has to be
encrypted by this EncryptedWritable class. What I was thinking about is
allow data source to encrypt its data any way they want, as long as it is
supported by Java security package, then only provide the private key to the
runtime to decrypt it.




From: davidparks21@yahoo.com
To: user@hadoop.apache.org
Subject: RE: Question related to Decompressor interface
Date: Sun, 10 Feb 2013 09:36:40 +0700

I can't answer your question about the Decompressor interface, but I have a
query for you.


Why not just create an EncryptedWritable object? Encrypt/decrypt the bytes
on the read/write method, that should be darn near trivial. Then stick with
good 'ol SequenceFile, which, as you note, is splittable. Otherwise you'd
have to deal with making the output splittable, and given encrypted data,
the only solution that I see is basically rolling your own SequenceFile with
encrypted innards. 


Come to think of it, a simple, standardized EncryptedWritable object out of
the box with Hadoop would be great. Or perhaps better yet, an
EncryptedWritableWrapper<T extends Writable> so we can convert any existing
Writable into an encrypted form.





From: java8964 java8964 [mailto:java8964@hotmail.com] 
Sent: Sunday, February 10, 2013 3:50 AM
To: user@hadoop.apache.org
Subject: Question related to Decompressor interface




Currently I am researching about options of encrypting the data in the
MapReduce, as we plan to use the Amazon EMR or EC2 services for our data.


I am thinking that the compression codec is good place to integrate with the
encryption logic, and I found out there are some people having the same idea
as mine.


I google around and found out this code:




It doesn't seem maintained any more, but it gave me a starting point. I
download the source code, and try to do some tests with it.


It doesn't work out of box. There are some bugs I have to fix to make it
work. I believe it contains 'AES' as an example algorithm.


But right now, I faced a problem when I tried to use it in my testing
MapReduer program. Here is the stack trace I got:


2013-02-08 23:16:47,038 INFO
org.apache.hadoop.io.compress.crypto.CryptoBasicDecompressor: buf length =
512, and offset = 0, length = -132967308


    at java.nio.ByteBuffer.wrap(ByteBuffer.java:352)




    at java.io.InputStream.read(InputStream.java:82)


    at org.apache.hadoop.util.LineReader.readLine(LineReader.java:173)





    at org.apache.hadoop.mapreduce.Mapper.run(Mapper.java:139)

    at org.apache.hadoop.mapred.MapTask.runNewMapper(MapTask.java:645)

    at org.apache.hadoop.mapred.MapTask.run(MapTask.java:325)

    at org.apache.hadoop.mapred.Child$4.run(Child.java:268)

    at java.security.AccessController.doPrivileged(Native Method)

    at javax.security.auth.Subject.doAs(Subject.java:396)


    at org.apache.hadoop.mapred.Child.main(Child.java:262)


I know the error is thrown out of this custom CryptoBasicDecompressor class,
but I really have questions related to the interface it implemented:


There is limited document about this interface, for example, when and how
the method setInput() will be invoked. If I want to write my own
Decompressor, what do these methods mean in the interface?

In the above case, I enable some debug information, you can see that in this
case, the byte[] array passed to setInput method, only have 512 as the
length, but the 3rd parameter of length passed in is a negative number:
-132967308. That caused the IndexOutOfBoundsException. If I check the
GzipDecompressor class of this method in the hadoop, the code will also
throw IndexOutoutBoundsException in this case, so this is a RuntimeException
case. Why it happened in my test case?


Here is my test case:


I have a simpel log text file about 700k. I encrypted it with above code
using 'AES'. I can encrypted and decrypted to get my original content. The
file name is foo.log.crypto, this file extension is registered to invoke
this CryptoBasicDecompressor in my testing hadoop using CDH4.1.2 release
(hadoop 2.0). Everything works as I expected. The CryptoBasicDecompressor is
invoked when the input file is foo.log.crypto, as you can see in the above
stack trace. But I don't know why the 3rd parameter (length) in setInput()
is a negative number at runtime.


In additional to it, I also have further questions related to use
Compressor/Decompressor to handle the encrypting/decrypting file. Ideally, I
wonder if the encrypting/decrypting can support file splits. This maybe
depends the algorithm we are using, is that right? If so, what kind of
algorithm can do that? I am not sure if it likes the compressor cases, most
of them do not support file split. If so, it maybe not good for my


If we have a 1G file, encrypted in the Amazone S3, after it copied to the
HDFS of Amazon EMR, can each block of the date be decrypted independently by
each mapper, then passed to the underline RecorderReader to be processed
totally concurrently? Does any one do this before? If so, what encryption
algorithm does support it? Any idea?





View raw message