Hi, I've playing around HDFS code, trying to do something like Avatar Nodes but using the backup node. I have modified datanodes to forward heartbeats, block reports and block received to backup node . So those are my questions:
1) Until now I did not force the BackupNode to stay on safe mode. So as the first blocks were written ( using the namenode), the backup node told the datanode to delete them. Why did it happened ? If I put it on safe mode it would not happen, right? Does putting the backup node in safe mode prevent it from sending any commands that would contradict namenode and corrupt data? 2) In the avatar node solution, the blockReceived method returns a list of blocks that the avatar datanode should retry to send to the namenode. Why is it necessary? 3) If a client connects to the backup node, can it perform any actions it would perform if it were a namenode ? Thanks for your time and regards, André Oriani
