Hi, Actually, how does namenode handle when block report is sent from datanodes?
After removing snapshot the file size = 0.5*BLOCK_SIZE. but the block report says it is 0.6*BLOCK_SIZE,(if we do not shrink the block) there is a mismatch. On Tue, Oct 21, 2014 at 8:34 PM, Pushparaj Motamari <pushpara...@gmail.com> wrote: > Hi, > > Generally the user expectation is to get back to the previous state of > FileSystem, so we should see the old state of block(the block size should > shrink). I couldn't find the semantics of this kind of operations in design > document either. > > Thank You > > Pushparaj > > On Tue, Oct 21, 2014 at 8:23 PM, Andrew Wang <andrew.w...@cloudera.com> > wrote: > >> Hi Pushparaj, >> >> Based on those steps, isn't the expectation that you end up with a >> 0.6*size >> block at the end? Blocks are append only (i.e. only grow in length), so >> deleting a snapshot cannot result in a block shrinking. >> >> Best, >> Andrew >> >> On Tue, Oct 21, 2014 at 7:20 AM, Pushparaj Motamari < >> pushpara...@gmail.com> >> wrote: >> >> > Hi, >> > >> > Consider the scenario below. >> > >> > /A/file.txt in HDFS >> > >> > 1. file.txt size = 0.5*(Size of Block). >> > >> > 2. Take Snapshot on directory A >> > >> > 3. Append to the file.txt, making its size as 0.6*(Size of Block) >> > >> > 4. Remove/delete the snapshot on folder taken in step2. >> > >> > My Question >> > -------------------------------- >> > 1. How does HDFS handle when it has to decrease the size of block, as >> above >> > when we delete the snapshot the block size has to be reduces to its >> > previous size. This code is handled at datanode or Namnode? >> > >> > Thank You >> > >> > >
