On Tue, Jun 15, 2010 at 9:53 AM, Dhruba Borthakur <[email protected]> wrote:
> You can start downloading and using the code form the hadoop-0.20-append. > > Though it's still missing some patches, so I'd recommend SVN checkouts and watching the branch :) -Todd > thanks, > dhruba > > On Tue, Jun 15, 2010 at 8:24 AM, Todd Lipcon <[email protected]> wrote: > > > On Tue, Jun 15, 2010 at 5:31 AM, Jan Stöcker <[email protected]> > > wrote: > > > > > Hello, > > > > > > > > > > > > in the hdfs-default.xml, I found the warning that the property > > > > > > "dfs.support.append" is false by default, because of the "append > > > > > > code" containing bugs. > > > > > > For our test case (which is not very complicated), I tried it out > > > > > > nevertheless (in a C program using libhdfs), and I did not > > > > > > encounter any problems yet. The only thing I noticed is that the > > > > > > modification date of the file to which I append is not changed. > > > > > > > > > > > > Could someone indicate which other bugs may occur? And in > > > > > > which cases? Because the append could be quite useful for me, > > > > > > and I would like to know the risk. > > > > > > > Mostly in failure handling cases, but not entirely. The result is usually > > truncated files, sometimes truncated as if your append didn't happen, > > sometimes the entire last block going missing. Look for the 0.20-append > fix > > version on JIRA for a more significantly list. > > > > I'd recommend waiting a couple weeks and using the hadoop-0.20-append > > branch > > which will contain fixes for many of these bugs. > > > > -Todd > > > > > > -- > > Todd Lipcon > > Software Engineer, Cloudera > > > > > > -- > Connect to me at http://www.facebook.com/dhruba > -- Todd Lipcon Software Engineer, Cloudera
