You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@pig.apache.org by Jakub Glapa <ja...@gmail.com> on 2013/01/17 23:11:23 UTC

Advanced HDFS operations from Python embedded scripts

Hi,
my pig script is going to produce a set of files that will be an input for
a different process. The script would be running periodically so the number
of files would be growing.
I would like to implement an expiry mechanism were I could remove files
that are older than x or the number of files has reached some threshold.

I know a crazy way were in bash script you can call "hadoop fs -ls ...",
parse the output and then execute "rmr" on matching entries.

Is there a "human" way to do this from under python script? Pig.fs()
doesn't come in handy because it doesn't return anything to the script but
maybe I'm missing something?
How could I approach that differently other than writing a java program or
using shell? Python looks like a great idea but seems a bit limited at
least in version 0.10.1.

I appreciate any help!

--
regards,
Jakub Glapa

Re: Advanced HDFS operations from Python embedded scripts

Posted by Jakub Glapa <ja...@gmail.com>.
that looks promising, thanks Clement!



--
regards,
pozdrawiam,
Jakub Glapa


On Fri, Jan 18, 2013 at 9:12 AM, Clément MATHIEU <cl...@unportant.info>wrote:

> On 2013-01-17 23:11, Jakub Glapa wrote:
>
> Hi Jakub,
>
>
>  my pig script is going to produce a set of files that will be an input for
>> a different process. The script would be running periodically so the
>> number
>> of files would be growing.
>> I would like to implement an expiry mechanism were I could remove files
>> that are older than x or the number of files has reached some threshold.
>>
>> I know a crazy way were in bash script you can call "hadoop fs -ls ...",
>> parse the output and then execute "rmr" on matching entries.
>>
>> Is there a "human" way to do this from under python script? Pig.fs()
>>
>
> I had the same issue than you few months ago. The public Pig scripting API
> only exposes a FsShell object which is way too limited to do any real work.
> However it is possible to get access to the Hadoop FileSystem API from a
> Python script:
>
>
> def get_fs():
>     """Return a org.apache.hadoop.fs.**FileSystem instance."""
>     # Pig scripting API exports a FsShell but not a FileSystem object.
>     ctx   = ScriptPigContext.get()
>     props = ctx.getPigContext().**getProperties()
>     conf  = ConfigurationUtil.**toConfiguration(props)
>     fs    = FileSystem.get(conf)
>     return fs
>
>
> Once you have a FileSystem object you can do whatever you want using the
> standard Hadoop API.
>
>
> Hope this helps.
>
> -- Clément
>

Re: Advanced HDFS operations from Python embedded scripts

Posted by Clément MATHIEU <cl...@unportant.info>.
On 2013-01-17 23:11, Jakub Glapa wrote:

Hi Jakub,

> my pig script is going to produce a set of files that will be an 
> input for
> a different process. The script would be running periodically so the 
> number
> of files would be growing.
> I would like to implement an expiry mechanism were I could remove 
> files
> that are older than x or the number of files has reached some 
> threshold.
>
> I know a crazy way were in bash script you can call "hadoop fs -ls 
> ...",
> parse the output and then execute "rmr" on matching entries.
>
> Is there a "human" way to do this from under python script? Pig.fs()

I had the same issue than you few months ago. The public Pig scripting 
API only exposes a FsShell object which is way too limited to do any 
real work. However it is possible to get access to the Hadoop FileSystem 
API from a Python script:


def get_fs():
     """Return a org.apache.hadoop.fs.FileSystem instance."""
     # Pig scripting API exports a FsShell but not a FileSystem object.
     ctx   = ScriptPigContext.get()
     props = ctx.getPigContext().getProperties()
     conf  = ConfigurationUtil.toConfiguration(props)
     fs    = FileSystem.get(conf)
     return fs


Once you have a FileSystem object you can do whatever you want using 
the standard Hadoop API.


Hope this helps.

-- Clément