Yes, depends on a E2E chain.  
Stop the agent, go to /tmp/name-of-collector and remove all directories there ( 
should be sent, sending, tmp and another one ). 
Then echo > your-file (erase them) and restart flume. You will lost the logs 
from, but I didnt know another solution yet. 
Hope it helps. 

Alex

sent via my mobile device

On Dec 1, 2011, at 8:43 PM, Mark Lewandowski <mark.e.lewandow...@gmail.com> 
wrote:

> Hmmm, the only problem is that these boxes are EC2 instances, and because 
> flume is writing such large files I'm filling 100% of my disk space before 
> retransmission would even occur.  The files are being put in the sent 
> directory, which I believe means that they've been successfully sent to the 
> collector.
> 
> I tried stopping the flume process, truncating the log file I'm writing to, 
> and wiping the sent/, writing/, logged/, sending/ directories in tmp (where 
> flume writes while doing E2E), and restarting the process that writes the 
> original log.  Even after all that, when I restart flume it starts trying to 
> send over these huge chunks of data.
> 
> Is there something else I can try blowing away that would reset the WAL?
> 
> -Mark
> 
> On Thu, Dec 1, 2011 at 11:01 AM, Eric Sammer <esam...@cloudera.com> wrote:
> Mark:
> 
> This is due to some kind of wacky state in the WAL (I'm assuming this is E2E) 
> if I remember correctly. The issue is that events keep piling up in the WAL 
> and it never gets drained properly. I don't have the details in front of me 
> but I vaguely remember this being related to downstream ACK issues (or could 
> be a remanent thereof). You could try taking a look at the logs on the errant 
> host to confirm the retransmission and lack of ACK.
> 
> Let us know.
> 
> 
> On Thu, Dec 1, 2011 at 10:53 AM, Mark Lewandowski 
> <mark.e.lewandow...@gmail.com> wrote:
> Hi all,
> 
> I've got 9 identical boxes configured in my flume cluster.  8 of them are 
> working correctly, with exactly the same config (source: tail, sink: 
> AutoE2EChain).  1 box is trying to send way too much data through my 
> collector.  I'm tailing a log file that's currently ~200k, but when I start 
> the flume agent, and I the writing and sending directories I see that flume 
> is trying to send files of ~1GB.  A new 1GB file is created in this directory 
> about every 10s.
> 
> Any ideas?
> 
> -Mark
> 
> 
> 
> -- 
> Eric Sammer
> twitter: esammer
> data: www.cloudera.com
> 

Reply via email to