Hi, Having started repeatedly some mini clusters on EC2 today, I ran several times into the following problem.
The install phase of Whirr takes quite some time to download all the stuff (java, zookeeper, hadoop, hbase), often going over 10 minutes in time. I noticed that after 10 minutes, if the install phase script has not finished yet, Whirr happily prints "Script run completed" and goes on with the configure phase (I noticed the same thing yesterday when adding the blocking wait-for-namenode thing before starting HBase services). The processes on the servers clearly show that the script from the install phase is still running then. The configure scripts fail quickly because some of the software is not yet available. I know there's a curl timeout of 10 minutes, but this is unrelated. Sometimes after 10 minutes it was still doing the apt-get of java. Does this ring a bell to someone? -- Bruno Dumon Outerthought http://outerthought.org/