flume-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Brock Noland <br...@cloudera.com>
Subject Re: HDFS Sink log rotation on the basis of time of writing
Date Mon, 05 Nov 2012 15:57:00 GMT

Yes you are correct, I suggest running a MR job once an hour to merge
those 60 files into one file.


On Mon, Nov 5, 2012 at 9:49 AM, Pankaj Gupta <pankaj@brightroll.com> wrote:
> Hi Brock,
> But then if I rotate frequently e.g. every minute, the total number of files in a single
folder of HDFS will go into thousands very quickly. I am not sure how/if that will affect
HDFS namenode performance and I worry that it may suffer. I don't have a lot of experience
with HDFS, do you happen to know if having thousands of files in a single directory in HDFS
is common?
> Thanks,
> Pankaj
> On Nov 5, 2012, at 7:30 AM, Brock Noland <brock@cloudera.com> wrote:
>> Hi,
>> If you just did not bucket the data at all, it would be organized by
>> the time they arrived at the sink.
>> Brock
>> On Fri, Nov 2, 2012 at 6:08 PM, Pankaj Gupta <pankaj@brightroll.com> wrote:
>>> Hi,
>>> Is it possible to organize files written to HDFS into buckets based on the
>>> time of writing rather than the timestamp in the header? Alternatively, is
>>> it possible to insert the timestamp injector just before the HDFS Sink?
>>> My use case is  to organize files such that they are organized
>>> chronologically as well as alphabetically by name and that there is only one
>>> file being written to at a time. This will make it easier to look for newly
>>> available data so that MapReduce jobs can process them.
>>> Thanks in Advance,
>>> Pankaj
>> --
>> Apache MRUnit - Unit testing MapReduce - http://incubator.apache.org/mrunit/

Apache MRUnit - Unit testing MapReduce - http://incubator.apache.org/mrunit/

View raw message